# howtobedigital — the whole site, one file > Free, plain-English lessons on the digital and AI skills working adults in the UK were never taught. One lesson, one outcome, one measured time, and one real task on your own machine at the end. No account, no paywall. **Generated** 8 August 2026 · **Canonical** https://howtobedigital.com/ · **Index for agents** https://howtobedigital.com/llms.txt · **Licence** CC BY-NC-SA 4.0 Written by `build.py` from the same content that produces the HTML, never by hand. If this file and a page disagree, the page is right and the build is broken. Prefer the per-page markdown twin at `/index.md` for a targeted read. --- # howtobedigital > Free, plain-English lessons on the digital and AI skills working adults in the UK were never taught. One lesson, one outcome, one measured time, and one real task on your own machine at the end. No account, no paywall. ## You are not behind. Nobody taught you. Short lessons on the digital and AI bits of work. Every one states what you will be able to do, tells you who should skip it, and ends with one real task on your own machine. ## Start here - [Is it safe to put this in?](/learn/ai/is-it-safe-to-put-this-in/) — You want to paste something in and you are not sure you should. 14 minutes. - [Tell a real message from a fake one](/learn/digital-skills/tell-a-real-message-from-a-fake-one/) — A message came in asking you to do something, urgently. 15 minutes. - [Ask it something daft and see what happens](/learn/play/ask-it-something-daft/) — You have never used one of these and would rather not start at work. 8 minutes. ## Where are you now - **Left out** — Somebody at work assumed you already knew how to do this, and you did not want to say. In about a month, the five things you currently ask somebody else to do, you will do yourself. - **Getting by** — You manage. You are slow, you lose things, and you are never quite sure you did it right. You will stop losing work, and stop guessing whether a message is genuine. - **Getting going** — You have used one of these tools. You want to be better at it, and safer. You will decide what is safe to hand over, and check what comes back. - **Leading** — People are asking what the organisation should do, and there is no answer yet. You will be able to answer the question your board is going to ask. ## The numbers - **8m** UK adults cannot do all eight of the tasks that make up the basic level. — Ipsos for Lloyds Banking Group, *Essential Digital Skills 2025*, Fieldwork 5 March – 2 April 2025 — https://www.ipsos.com/en-uk/lloyds-essential-digital-skills-2025 - **35%** of UK businesses with ten or more staff use at least one AI technology. — Office for National Statistics, *Artificial intelligence in UK businesses: 2023 to 2026*, 20 July 2026 — https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026 - **11%** say more than half their workforce has had any training on it. Same release, same base. — Office for National Statistics, *Artificial intelligence in UK businesses: 2023 to 2026*, 20 July 2026 — https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026 --- # About howtobedigital teaches working adults in the UK the digital and AI skills nobody taught them. Small on purpose. Free for individuals, paid for organisations. ## What this promises - Free means free. No account, no email wall, no paywall. - One lesson, one outcome, stated in the first line. - Every lesson says who should skip it. - Times measured with a stopwatch, not estimated. - Every lesson ends with one real task on your own machine. - Every statistic carries a named source. - Nothing is sold on a lesson page. - Works without JavaScript, on an old phone, on a slow connection. - Nothing measures you. ## What this does not do Not in Welsh. No text-to-speech yet. It is a website, and for adults over 65 who would rather ask a person it is not the whole solution. Everything free here is asynchronous. ## Contact hello@howtobedigital.com. Read by a person. Corrections are published. --- # The method Every lesson is built the same way and has to pass the same twelve checks. Published under CC BY-NC-SA 4.0. Take it and use it. ## Seven steps 1. **Start from a moment, not a topic** Not "email management". Somebody sent a thread with forty replies and asked what we decided. Topics are how institutions organise content. Moments are how people search. Test: could you overhear this in a break room? 2. **Name the outcome, then name the exit** One observable thing the reader can do afterwards, written as a verb they perform rather than something they understand. Then, immediately, who should skip it. If half the audience should leave, say so. 3. **Cut to the smallest true thing** If it does not change what the reader does in the next twelve minutes, it is not in this lesson. Then do the whole thing with a stopwatch, at the pace of somebody meeting it for the first time, and publish the number that comes out. 4. **Show, then let them try with the answer visible** Every step says what the reader should see afterwards. That is what stops somebody quietly concluding they have broken something. Then one worked variant with the answer shown, before anything unaided. That guided middle is where most free training has a hole. 5. **Write down what went wrong** Two to four real failures, hit while doing the lesson, and what to do about each. Required, not optional. It is the difference between teaching and documenting. 6. **One real task, today** On the reader's own machine or in their own life, under ten minutes, producing something that did not exist before. Four tests: is it real, is it today, is it one thing, does it make something. 7. **Check it, then leave one way forward** Twelve checks. A published date and a last-checked date. Then exactly one link to the next lesson, not a grid of related content. ## The lesson spine 1. Breadcrumb — track, then path 2. Mono label — lesson number, track, time, last checked 3. Title — the moment 4. Outcome — "After this you will be able to X" 5. Skip this if — honest entry criteria 6. What you need — device, account, licence, cost 7. Retrieval opener — one question about the previous lesson 8. Steps — numbered, each saying what you should see 9. Try this one first — a worked variant with the answer visible 10. If it goes wrong — two to four real failures 11. Do today — one task. The only green block on the page 12. Where else this works — two dissimilar situations 13. Practice card, check block, and exactly one next lesson ## What was deliberately left out - **Learning styles** No evidence behind them. There will never be a "visual learner" route here. - **A quiz at the end** A quiz about the lesson measures whether you read the lesson. Around twenty free UK providers end this way. It is the ending this site differentiates against. - **Completion certificates** They measure attendance. This audience has been measured enough. - **Video by default** It cannot be skimmed, it needs headphones somebody may not have, and it is wrong the moment a menu moves. Where video helps it sits alongside the text, never instead of it, always with a transcript. - **Streaks, badges, progress bars, scores** Anything that makes a learner feel measured works against the one rule this site has. - **Accounts and logins** Not now, not later. Retrieval and spacing are done with prompts instead, which needs no account. --- # Twelve checks Six run in four seconds. Six take an afternoon and need a person. | Check | Fails when | Who | | --- | --- | --- | | Outcome is observable | It says "understand" or "be aware of" | Person | | Title is a moment | You could not overhear it in a break room | Person | | Entry criteria are honest | Written so nobody would ever skip | Person | | Time was measured | No stopwatch was involved | Person | | Every step says what you should see | Any step ends without it | Person | | Failures are real | Imagined rather than hit while doing it | Person | | Required blocks present | Outcome, skip-if, failures, do-today or check missing | Build | | Reading grade | Above 13.0. Warns above 11.0 | Build | | Banned words and voice tics | Any match without an explicit allow | Build | | Every claim sourced | A statistic with no source, or one that will not resolve | Build | | Links alive, content fresh | Dead link. Warns past the stale window | Build | | Brand rules | Two primary buttons, a shadow, a rounded corner, a video without a transcript | Build | ## Today's build ``` $ python check.py BUILD 8 lessons · 3 paths · 6 tracks · 10 sources · 3 posts · 50 pages · 50 markdown twins ok required blocks present on every lesson ok every step carries its confirming observation ok reading grade max 6.8 on is-it-safe-to-put-this-in (ceiling 13.0, target 11.0) ok banned words and voice tics ok every sourceId resolves, every source has a url and a date ok every figure sits near its named source ok internal links all resolve ok brand rules: no shadow, no radius, one primary action, no amber CTA in a lesson, no icon fonts ok no personal attribution anywhere in the output ok one h1, skip link, lang, title and description on every page ok markdown twin, alternate link, llms.txt, site.md, rss, sitemap, headers ok no video embed ships without a transcript link ok fonts self-hosted, 52.8 KB / 60 KB across 6 subset faces ok third-party requests: Google Analytics only, cookieless — the privacy page says exactly that ok javascript 7.5 KB / 8 KB · css 16.1 KB / 18 KB · largest page 14.5 KB / 30 KB warn /accessibility/index.html: 3 em-dashes in 346 words (budget 2) warn /method/index.html: 14 em-dashes in 924 words (budget 4) warn /method/standards/index.html: 11 em-dashes in 411 words (budget 3) warn /for-teachers/index.html: 5 em-dashes in 413 words (budget 3) warn /learn/ai/asking-for-something-and-getting-it-first-time/index.html: 6 em-dashes in 902 words (budget 4) warn /learn/ai/checking-the-answer-before-you-send-it/index.html: 6 em-dashes in 818 words (budget 4) warn /learn/ai/is-it-safe-to-put-this-in/index.html: 7 em-dashes in 1091 words (budget 4) warn /learn/ai/what-ai-at-work-actually-means/index.html: 10 em-dashes in 904 words (budget 4) warn /learn/ai/try-a-chatbot-for-the-first-time/index.html: 9 em-dashes in 923 words (budget 4) warn /learn/play/ask-it-something-daft/index.html: 7 em-dashes in 795 words (budget 3) warn /learn/digital-skills/tell-a-real-message-from-a-fake-one/index.html: 8 em-dashes in 934 words (budget 4) warn /learn/digital-skills/find-a-file-you-have-lost/index.html: 10 em-dashes in 902 words (budget 4) warn /evidence/index.html: 6 em-dashes in 1167 words (budget 4) warn /practice/what-ai-at-work-actually-means/index.html: 4 em-dashes in 293 words (budget 2) warn 8 lessons carry an empty video slot — no transcript needed until one is attached PERSON CHECKS OUTSTANDING (all 8 lessons) [ ] steps performed end to end on a real machine [ ] time measured with a stopwatch [ ] failures recorded from failures actually hit [ ] independent read-through for reading level -> every lesson carries a draft block until these are ticked 15 passed · 15 warnings · 0 failures ``` --- # How this is made ## The AI declaration **AI may draft.** Outlines, first passes, alternative phrasings, summarising a source, checking readability. **AI may not decide.** What is taught, what is cut, what the time estimate is, whether a claim is true, or whether a step actually works. **A person does the doing.** Every step in every published lesson is performed on a real machine by a person who watches what happens. **A person checks and publishes.** Nothing is auto-published. Six of the twelve checks require a person. Four are currently unticked on all eight lessons and every lesson says so at the top. ## Why there is no name on this The claim is not that a named person wrote this. The claim is that a person did this, against a process you can read and a build you can check. The method is published, the build log is published including failures, and the address is answered. --- # Standards Two national frameworks, two different licences. ## AI foundation skills for work Skills England · 28 January 2026 · https://www.gov.uk/government/publications/ai-foundation-skills-for-work - Three domains: technical, non-technical, and responsible and ethical AI foundation skills, with six competencies across them Open Government Licence v3.0 — reusable and adaptable, including commercially, with attribution. The exact competency-to-domain mapping is pending a check against the published benchmark. ## Essential digital skills standards: 2026 Department for Education · 15 July 2026 · https://www.gov.uk/government/publications/essential-digital-skills-standards-2026 - Five levels, Entry 1 to Level 2, updated and expanded in response to technological change, including artificial intelligence Crown copyright. The qualifications behind it have not been rewritten and no implementation date has been set. Read the licence statement before adapting rather than citing. ## What is tagged so far - What AI at work actually means — Skills England: Technical AI foundation skills - Try one for the first time without giving anything away — Skills England: Technical AI foundation skills - Is it safe to put this in? — Skills England: Responsible and ethical AI foundation skills; Department for Education: Being safe and legal online, Level 2 - Asking for something and getting it first time — Skills England: Technical AI foundation skills - Checking the answer before you send it — Skills England: Responsible and ethical AI foundation skills - Find a file you saved and cannot find again — Department for Education: Handling information and content, Entry Level 3 - Tell a real message from a fake one — Department for Education: Being safe and legal online, Level 1 **The gap, stated.** A statement-by-statement mapping that anybody could audit does not exist yet. No page here claims alignment until it does. --- # Lessons 8 lessons published. All free, no account, all currently drafts. ## Digital Skills Level 0 and 1. Finding a file, telling a real message from a fake one, filling in a form that might time out. This is the most crowded ground on the site and it is here because you cannot start the AI track without it, and because a drop-in needs sheets it can print. Levels 0–1 · 2 published · re-checked every 120 days - [Find a file you saved and cannot find again](/learn/digital-skills/find-a-file-you-have-lost/) — 9 min, Level 0. After this you will be able to find any file on your own device, three ways. - [Tell a real message from a fake one](/learn/digital-skills/tell-a-real-message-from-a-fake-one/) — 15 min, Level 0. After this you will be able to check whether a message asking you to do something is genuine, and report it if it is not. ## AI Fundamentals Level 1 and 2. What these tools are, what is safe to put in them, how to ask for something and get it first time, and how to check what comes back. Aligned to the Skills England AI foundation skills for work benchmark. Levels 1–2 · 5 published · re-checked every 120 days - [What AI at work actually means](/learn/ai/what-ai-at-work-actually-means/) — 8 min, Level 1. After this you will be able to tell the difference between a chatbot, an assistant built into software you already use, and automation somebody set up. - [Try one for the first time without giving anything away](/learn/ai/try-a-chatbot-for-the-first-time/) — 10 min, Level 1. After this you will be able to have a first go on your own device, with nothing of yours in it. - [Is it safe to put this in?](/learn/ai/is-it-safe-to-put-this-in/) — 14 min, Level 2. After this you will be able to decide whether a piece of information is safe to give an AI tool, under UK data protection rules. - [Asking for something and getting it first time](/learn/ai/asking-for-something-and-getting-it-first-time/) — 12 min, Level 1. After this you will be able to write a request that includes the task, who it is for, how long it should be, and an example. - [Checking the answer before you send it](/learn/ai/checking-the-answer-before-you-send-it/) — 10 min, Level 2. After this you will be able to run three checks on anything an AI tool writes for you. ## Copilot and Agents Level 1 to 3. The most fragile track on the site: interfaces change and licensing changes. Every lesson here is re-checked every 60 days rather than every 120. Levels 1–3 · 0 published · re-checked every 60 days - Nothing published yet. ## Emerging Technologies Level 1 and 2. A technology enters this track when an ordinary person can actually touch it, not when it is announced. If you cannot write a do-today task for it, it is not ready. Levels 1–2 · 0 published · re-checked every 120 days - Nothing published yet. ## Horizon Scanning Level 2 and 3. The leaders' track. Every claim about the future is labelled as a claim and attributed to whoever made it. No lesson here predicts anything on its own authority. Levels 2–3 · 0 published · re-checked every 120 days - Nothing published yet. ## Digital Play Level 0 to 2. You cannot learn a thing you are frightened of getting wrong. Nothing in Play touches work data or anything you could get in trouble for. Free tools only. Every play lesson still teaches something nameable. Levels 0–2 · 1 published · re-checked every 120 days - [Ask it something daft and see what happens](/learn/play/ask-it-something-daft/) — 8 min, Level 0. After this you will be able to use one of these tools once, with nothing of yours in it, so the first time is not at work. --- # Digital Skills Track · Levels 0–1 · re-checked every 120 days Level 0 and 1. Finding a file, telling a real message from a fake one, filling in a form that might time out. This is the most crowded ground on the site and it is here because you cannot start the AI track without it, and because a drop-in needs sheets it can print. **Not in this track.** Anything needing a work account. Anything about one employer's software. ## Published - [Find a file you saved and cannot find again](/learn/digital-skills/find-a-file-you-have-lost/) — 9 min, Level 0 - [Tell a real message from a fake one](/learn/digital-skills/tell-a-real-message-from-a-fake-one/) — 15 min, Level 0 --- # AI Fundamentals Track · Levels 1–2 · re-checked every 120 days Level 1 and 2. What these tools are, what is safe to put in them, how to ask for something and get it first time, and how to check what comes back. Aligned to the Skills England AI foundation skills for work benchmark. **Not in this track.** Anything vendor-specific. That is Copilot and Agents. Anything about the future. That is Horizon Scanning. ## Published - [What AI at work actually means](/learn/ai/what-ai-at-work-actually-means/) — 8 min, Level 1 - [Try one for the first time without giving anything away](/learn/ai/try-a-chatbot-for-the-first-time/) — 10 min, Level 1 - [Is it safe to put this in?](/learn/ai/is-it-safe-to-put-this-in/) — 14 min, Level 2 - [Asking for something and getting it first time](/learn/ai/asking-for-something-and-getting-it-first-time/) — 12 min, Level 1 - [Checking the answer before you send it](/learn/ai/checking-the-answer-before-you-send-it/) — 10 min, Level 2 --- # Copilot and Agents Track · Levels 1–3 · re-checked every 60 days Level 1 to 3. The most fragile track on the site: interfaces change and licensing changes. Every lesson here is re-checked every 60 days rather than every 120. **Not in this track.** Building or fine-tuning models. The teaching here is for general users and implementers, not specialists. ## Published - Nothing yet. --- # Emerging Technologies Track · Levels 1–2 · re-checked every 120 days Level 1 and 2. A technology enters this track when an ordinary person can actually touch it, not when it is announced. If you cannot write a do-today task for it, it is not ready. **Not in this track.** Anything you cannot try this week. ## Published - Nothing yet. --- # Horizon Scanning Track · Levels 2–3 · re-checked every 120 days Level 2 and 3. The leaders' track. Every claim about the future is labelled as a claim and attributed to whoever made it. No lesson here predicts anything on its own authority. **Not in this track.** Opinion pieces about the future of work. ## Published - Nothing yet. --- # Digital Play Track · Levels 0–2 · re-checked every 120 days Level 0 to 2. You cannot learn a thing you are frightened of getting wrong. Nothing in Play touches work data or anything you could get in trouble for. Free tools only. Every play lesson still teaches something nameable. **Not in this track.** Anything with a right answer, a work login, or a subscription. ## Published - [Ask it something daft and see what happens](/learn/play/ask-it-something-daft/) — 8 min, Level 0 --- # What AI at work actually means `/learn/ai/what-ai-at-work-actually-means/` · AI Fundamentals · Level 1 · **8 minutes** (planned, not yet measured) · Last checked 14 July 2026 > **Draft.** The steps have not yet been performed end to end on a real machine and the time is planned rather than measured. Four of twelve checks outstanding. **After this you will be able to** tell the difference between a chatbot, an assistant built into software you already use, and automation somebody set up. **Skip this if** Skip this if you can already say which of those three a tool at your work is, without looking it up. People at work say "AI" about three completely different things. Sorting them out takes about eight minutes and it makes every conversation afterwards easier, because most of the confusion in a meeting comes from two people using one word for two things. *In a housing office last year, four people in one room were talking about "the AI" and meant four different products.* ## What you need - Nothing. You can read this on a phone on a bus. - No account, no licence, no cost. ## The steps 1. Think of a tool where you type a question into a box and get writing back. That is a chatbot. It knows nothing about your organisation unless you tell it. *You should see:* You should be able to name at least one. If you cannot, that is fine — lesson 2 is having a go with nothing at stake. 2. Now think of software you already use — email, a spreadsheet, a chat app — with a new button in it that offers to summarise or draft something. That is an assistant. It can usually see what you can see. *You should see:* You should be looking at a button or a panel inside something you already had, rather than a separate website. 3. Now think of something that happens without anybody pressing anything: a form that files itself, an alert that arrives on a schedule. That is automation. Somebody set it up once and it runs. *You should see:* You should notice that nobody is present when it happens. That is the whole difference. 4. Write the three words down, chatbot, assistant, automation, and put one tool from your own work under each. *You should see:* You should have a short list with at least one gap in it. The gap is the useful finding, not a failure. ## Try this one first Somebody says: "we've turned the AI on in Teams." That is an assistant. It sits inside software you already had, and it can generally reach what you can reach. The question that follows is not "what can it do" — it is "what can it see", and that is lesson 3. ## If it goes wrong - **The tool seems to be two of them at once.** Some are. A chatbot with access to your files is a chatbot doing an assistant's job. Name the two halves separately and the confusion goes. - **You cannot find out which one your organisation bought.** That is common and it is not your job to know. Ask whoever sent the announcement email. If nobody sent one, that is worth noticing too. - **Somebody tells you the distinction does not matter.** It matters for one reason: the three have different answers to "where does what I type end up". That is the only question this site treats as urgent. ## Do today Name the three tools at your work that fall into each category. Write the list on paper. If you cannot fill one of the three, write that down as well. **Where else this works.** The same three categories work on anything sold to you at home — a smart speaker, a banking app that flags spending, a photo app that tidies albums. **Covers:** Skills England — Technical AI foundation skills ## Sources - Ofcom, *Adults' Media Use and Attitudes 2026*, 2 April 2026 — https://www.ofcom.org.uk/media-use-and-attitudes/media-habits-adults/adults-media-use-and-attitudes/ --- Checked by a person · not auto-published · howtobedigital.com/learn/ai/what-ai-at-work-actually-means/ · CC BY-NC-SA 4.0 --- # Try one for the first time without giving anything away `/learn/ai/try-a-chatbot-for-the-first-time/` · AI Fundamentals · Level 1 · **10 minutes** (planned, not yet measured) · Last checked 14 July 2026 > **Draft.** The steps have not yet been performed end to end on a real machine and the time is planned rather than measured. Four of twelve checks outstanding. **After this you will be able to** have a first go on your own device, with nothing of yours in it. **Skip this if** Skip this if you have already typed something into a chatbot and read what came back. Any attempt counts, including one you thought went badly. **Before you start.** Without looking back: what are the three kinds of thing lesson 1 said "AI at work" usually means? (A chatbot, an assistant inside software you already use, and automation somebody set up once.) The first time you use one of these should not be at work, on something that matters, with somebody waiting. This lesson is the first time, done somewhere it cannot cost you anything. *Almost everybody's first attempt is a one-word question, and almost everybody is surprised by how much comes back.* ## What you need - A phone, tablet or computer with a browser. - No account for most free tools. If one asks you to sign up, close it and try another — you are not short of options. - No cost. ## The steps 1. Open a browser and go to any free chatbot. Do not create an account. Do not install anything. *You should see:* You should be looking at a box to type in, usually at the bottom of the screen, with some suggestions above it. 2. Type a question about something with no consequence. "Explain why bread rises" works. "What should I do about my mortgage" does not, and neither does anything with a real name in it. *You should see:* You should see your words appear, then a reply write itself out line by line rather than arriving all at once. 3. Read the answer, then type: "say that again in three sentences." *You should see:* You should get a shorter version. Notice it does not ask you to repeat the original question — it remembers, within this one conversation. 4. Type something it cannot possibly know: "what did I have for tea last night?" *You should see:* You should get an admission that it does not know, or a guess. Either is worth seeing once, deliberately, before it happens to you by accident at work. ## Try this one first "Explain how a fridge works, to somebody who left school at sixteen and is not stupid." You should get something plain and reasonably accurate. That instruction — naming who it is for — is the single most useful thing you can add to any request, and lesson 4 is about nothing else. ## If it goes wrong - **It asks you to make an account before you can type anything.** Close it and try a different one. Several are usable without signing up. You do not have to hand over an email address to have a first go. - **The answer is confidently wrong.** Good. That is the most useful thing that can happen to you today. These tools produce wrong answers in exactly the same tone as right ones, and knowing that in your body rather than in theory is what this lesson is for. - **It will not answer, or says it cannot help.** Try asking in a different way. If it still refuses, ask it why. The reason is usually a rule somebody wrote, and it will tell you. ## Do today Ask it one thing you are genuinely curious about and that involves nobody but you. Read the whole answer. Then close the tab. **Where else this works.** The habit of trying something once, somewhere it cannot cost you anything, works for every new tool you will ever be handed. **Covers:** Skills England — Technical AI foundation skills ## Sources - Ofcom, *Adults' Media Use and Attitudes 2026*, 2 April 2026 — https://www.ofcom.org.uk/media-use-and-attitudes/media-habits-adults/adults-media-use-and-attitudes/ --- Checked by a person · not auto-published · howtobedigital.com/learn/ai/try-a-chatbot-for-the-first-time/ · CC BY-NC-SA 4.0 --- # Is it safe to put this in? `/learn/ai/is-it-safe-to-put-this-in/` · AI Fundamentals · Level 2 · **14 minutes** (planned, not yet measured) · Last checked 14 July 2026 > **Draft.** The steps have not yet been performed end to end on a real machine and the time is planned rather than measured. Four of twelve checks outstanding. **After this you will be able to** decide whether a piece of information is safe to give an AI tool, under UK data protection rules. **Skip this if** Skip this if you already look up what your organisation allows before you paste anything in. If you are not sure your organisation has any rules at all, stay — that is lesson 12, and it is the more common answer. **Before you start.** Without looking back: what are the three kinds of thing lesson 1 said an AI tool actually is? (A chatbot, an assistant inside software you already use, and automation somebody set up.) This is the question people are most afraid to ask out loud, usually because they think everybody else already knows the answer. Nobody does. There is a three-step check and it takes about a minute once you have done it twice. *The most common way this goes wrong is not a data breach. It is somebody pasting a complaint letter in to get help replying, and not thinking of the name at the top as information.* ## What you need - A device you can type on, or a piece of paper. - No account, no licence, nothing to install. - Nothing real goes into anything during this lesson. ## The steps 1. Open a blank note on your own device, or take a sheet of paper, and write down the thing you were about to paste in. *You should see:* You should be looking at your own words, on your own machine, with nothing sent anywhere yet. 2. Go through it and underline anything that names a person, or could be traced back to one. *You should see:* You should have underlined names. You should also have underlined job titles, addresses, reference numbers, dates of birth, and anything about somebody's health, money or household. 3. Ask the question that decides it: if this appeared in an email to somebody outside your organisation tomorrow, would that be a problem? *You should see:* You should get a clear yes or no. If what you get is "it depends", treat it as a yes. 4. If the answer was yes, rewrite it with the specifics taken out, and check that the rewritten version still asks for what you actually need. *You should see:* You should have two versions in front of you: the original, which stays on your machine, and a general one that could go in. ## Try this one first "Can you rewrite this: Mrs Patterson at 14 Elm Road has complained twice about the boiler and we still haven't been out." Underline Mrs Patterson and 14 Elm Road. Answer to the question: yes, that would be a problem. So it does not go in as written. What does go in: "rewrite this — a tenant has complained twice about a boiler and we have not attended." The rewrite you get back is just as useful, because the tool was never using her name to do the writing. ## If it goes wrong - **You cannot tell whether a job title counts.** If only one person holds that title, it counts. "The night duty manager at the Blyth site" identifies somebody as surely as their name does. - **Your organisation has a tool it says is safe.** It may well be, and the assurance may be sound. That is a different question from whether the information is yours to share in the first place. Ask who signed it off and what it covers. - **You take the names out and the answer comes back useless.** That usually means the task genuinely needed the specifics, which means it is not a task to hand over. Do that one yourself. This is a good outcome, not a failed lesson. - **Somebody tells you it is fine because the tool does not train on your data.** That is one risk of several, and it is the one vendors answer because it is the one they can. It says nothing about who inside your organisation can see the conversation, or how long it is kept. ## Do today Find one real thing you were going to paste into an AI tool this week. Run the three steps on it. Either rewrite it so it passes, or decide not to use the tool for that one. **Where else this works.** The same three steps work before you put something into a translation site, before you forward a document to a supplier, and before you put a photo of paperwork in a group chat. **Covers:** Skills England — Responsible and ethical AI foundation skills; Department for Education — Being safe and legal online, Level 2 ## Sources - Office for National Statistics, *Artificial intelligence in UK businesses: 2023 to 2026*, 20 July 2026 — https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026 --- Checked by a person · not auto-published · howtobedigital.com/learn/ai/is-it-safe-to-put-this-in/ · CC BY-NC-SA 4.0 --- # Asking for something and getting it first time `/learn/ai/asking-for-something-and-getting-it-first-time/` · AI Fundamentals · Level 1 · **12 minutes** (planned, not yet measured) · Last checked 14 July 2026 > **Draft.** The steps have not yet been performed end to end on a real machine and the time is planned rather than measured. Four of twelve checks outstanding. **After this you will be able to** write a request that includes the task, who it is for, how long it should be, and an example. **Skip this if** Skip this if your requests already name the reader and the length. If you are getting answers that are technically right and completely unusable, this is the lesson. **Before you start.** Without looking back: what is the one question from lesson 3 that decides whether something is safe to put in? (If this appeared in an email to somebody outside your organisation tomorrow, would that be a problem?) Most disappointing answers are disappointing because the request left out something the person asking already knew and never said. There are four things worth saying every time, and adding them takes about fifteen seconds. *The single biggest jump in quality comes from naming the reader, and almost nobody does it unprompted.* ## What you need - Any device with a browser. - No account needed. - No cost. ## The steps 1. Write down what you actually want, as a verb. "Draft", "summarise", "list", "compare". Not "help me with". *You should see:* You should have a sentence starting with an instruction rather than a pleasantry. 2. Add who it is for, in your own words. "For a tenant who has complained twice already." "For my manager, who has two minutes." *You should see:* You should notice this is the part you would have said out loud to a colleague and never thought to type. 3. Add how long it should be. Words, sentences, bullet points, one paragraph — any unit works as long as it is a number. *You should see:* You should have a number in your request. "Short" is not a number and it will not do what you want. 4. Add an example of what good looks like, if you have one. Paste in a previous letter you were happy with, with the names taken out, per lesson 3. *You should see:* You should see the answer come back in a noticeably closer register than before. This is the step with the biggest effect and the one most often skipped. ## Try this one first Weak: "write something about the boiler policy." Better: "Draft a letter for a tenant who has already complained twice and is angry. Four short paragraphs. Plain English, no jargon, no apology that sounds like a template. Here is a letter I was happy with last year: …" Four additions, about fifteen seconds of typing: the verb, the reader, the length, the example. The second request will get something you can edit rather than something you have to rewrite. ## If it goes wrong - **The answer is still generic.** You almost certainly left out the reader. Add one sentence describing the person who will read it, including how they are feeling, and ask again. - **It is too long no matter what you say.** Give it a number and a unit: "four sentences", "120 words". If it overruns, reply with "half that" rather than starting again. - **It writes in a voice that is not yours.** That is what the example is for. One piece of your own past writing does more than any amount of describing the tone you want. ## Do today Take a request you made this week that came back disappointing. Add the four things. Send it again and compare the two answers side by side. **Where else this works.** The same four things work when you brief a colleague, a contractor or a designer. Most of what makes a request work has nothing to do with the tool. **Covers:** Skills England — Technical AI foundation skills --- Checked by a person · not auto-published · howtobedigital.com/learn/ai/asking-for-something-and-getting-it-first-time/ · CC BY-NC-SA 4.0 --- # Checking the answer before you send it `/learn/ai/checking-the-answer-before-you-send-it/` · AI Fundamentals · Level 2 · **10 minutes** (planned, not yet measured) · Last checked 14 July 2026 > **Draft.** The steps have not yet been performed end to end on a real machine and the time is planned rather than measured. Four of twelve checks outstanding. **After this you will be able to** run three checks on anything an AI tool writes for you. **Skip this if** Skip this if you already check every name, number and date against a source before anything goes out. **Before you start.** Without looking back: what are the four things lesson 4 said to put in every request? (The task as a verb, who it is for, how long it should be, and an example of what good looks like.) These tools produce wrong answers in exactly the same confident tone as right ones. There is no tell in the writing. So the check has to be a habit rather than a feeling, and three passes cover almost everything. *The invented ones are usually the most plausible-looking sentence on the page.* ## What you need - Something one of these tools wrote for you. If you have nothing, use the answer from lesson 4. - No account needed. ## The steps 1. First pass: names, numbers, dates. Go through and mark every one. Check each against something that is not the tool. *You should see:* You should find that most are right and it takes under two minutes. The point is the one that is not. 2. Second pass: read it as the person receiving it. Would they feel talked down to, or fobbed off? *You should see:* You should catch at least one phrase that is technically fine and lands badly. That is nearly always the apology sentence. 3. Third pass: the ownership question. Would you be content to have written this yourself, and to be asked about any line in it next week? *You should see:* You should get a straight yes or no. A no means edit it, not send it with a caveat. ## Try this one first "Under the Housing Act 2004, landlords must respond to repair requests within 14 days." Mark the Act, the year and the number. Check all three against the legislation, not against the tool. Confident, specific and legal-sounding is the exact shape an invented sentence takes. ## If it goes wrong - **You have nothing to check the facts against.** Then that sentence does not go out. Cut it, or find the source first. "I could not verify it" is a reason to remove a claim, not a reason to hedge it. - **It cites a source that sounds real.** Look it up. Invented citations are common and they are convincing because they are built out of the shapes of real ones. - **Checking takes longer than writing it yourself would have.** Then for that task it is not worth using. That is a real and reasonable finding, and lesson 16 is a list of the situations where it applies. ## Do today Take the last thing one of these tools wrote for you. Run the three passes. Count what you find. **Where else this works.** The three passes work on anything somebody else drafted for you — a colleague, an agency, a template you inherited. **Covers:** Skills England — Responsible and ethical AI foundation skills ## Sources - King's College London Policy Institute and King's Institute for AI, *Public attitudes to AI*, Published 19 May 2026 — https://www.kcl.ac.uk/news/one-in-five-britons-think-ai-will-create-civil-unrest-study-finds --- Checked by a person · not auto-published · howtobedigital.com/learn/ai/checking-the-answer-before-you-send-it/ · CC BY-NC-SA 4.0 --- # Find a file you saved and cannot find again `/learn/digital-skills/find-a-file-you-have-lost/` · Digital Skills · Level 0 · **9 minutes** (planned, not yet measured) · Last checked 14 July 2026 > **Draft.** The steps have not yet been performed end to end on a real machine and the time is planned rather than measured. Four of twelve checks outstanding. **After this you will be able to** find any file on your own device, three ways. **Skip this if** Skip this if you already use the search box on your computer or phone to find files by name. Files are not lost. They are somewhere with a name you have half-forgotten, and there are three ways to get to them. None of them involves scrolling through folders hoping. *"I've lost it" almost always turns out to mean "it saved somewhere I did not choose", and that is the machine's fault rather than yours.* ## What you need - The computer, phone or tablet where the file is. - Nothing to install. No account. ## The steps 1. By name. Find the search box — on Windows it is next to the Start button, on a Mac it is the magnifier at the top right, on a phone it is the swipe down from the middle of the home screen. Type any part of the name. *You should see:* You should get results as you type. Any word from the middle of the name works — you do not have to start at the beginning. 2. By when. Open your files and look for a way to sort by date. On a computer it is usually a column heading you can click. Sort so the newest is at the top. *You should see:* You should see the thing you were working on yesterday near the top. This is the one that works when you cannot remember the name at all. 3. By where it went on its own. Look in Downloads, and in Documents. Anything you opened from an email or a website is very likely in Downloads. *You should see:* You should find a folder with more in it than you expected. That is normal and it is where most "lost" files are. 4. When you find it, move it somewhere you chose, and give it a name you would search for. *You should see:* You should be able to find it again from the search box in one go. ## Try this one first You saved something called "Scan_20260714_0001.pdf" and cannot find it. Search for 2026. Any part of the name works. If nothing comes up, sort Downloads by date and look at 14 July. Then rename it to something you would actually type, like "boiler letter July". ## If it goes wrong - **Search finds nothing at all.** Try a shorter piece of the name, or one word from inside the document. If it still finds nothing the file may be on a different device — check the phone if you were searching the computer. - **It finds hundreds of things.** Add a second word, or use the date sort instead. Two pieces of information almost always narrows it to one screen. - **You find it but cannot open it.** That means you do not have the program it needs. Note the three letters after the dot in the name and ask for those. That is the useful thing to say to whoever helps you. ## Do today Find one file you thought you had lost. Rename it to something you would search for, and move it somewhere you chose. **Where else this works.** The three ways — by name, by date, by where things land on their own — work in your email, in a shared drive at work, and in your photos. **Covers:** Department for Education — Handling information and content, Entry Level 3 ## Sources - Ipsos for Lloyds Banking Group, *Essential Digital Skills 2025*, Fieldwork 5 March – 2 April 2025 — https://www.ipsos.com/en-uk/lloyds-essential-digital-skills-2025 --- Checked by a person · not auto-published · howtobedigital.com/learn/digital-skills/find-a-file-you-have-lost/ · CC BY-NC-SA 4.0 --- # Tell a real message from a fake one `/learn/digital-skills/tell-a-real-message-from-a-fake-one/` · Digital Skills · Level 0 · **15 minutes** (planned, not yet measured) · Last checked 14 July 2026 > **Draft.** The steps have not yet been performed end to end on a real machine and the time is planned rather than measured. Four of twelve checks outstanding. **After this you will be able to** check whether a message asking you to do something is genuine, and report it if it is not. **Skip this if** Skip this if you already check the sender's address and go to the organisation's own website rather than following the link in the message. The fakes are good now. Spelling mistakes stopped being the giveaway a while ago. What has not changed is the shape: a fake message wants you to do something quickly, using the route it has provided. *The ones that catch people are not the outlandish ones. They are a parcel delivery text on a day you are expecting a parcel.* ## What you need - The message in front of you, unopened links and all. - A way to look up the organisation separately: a browser, or the number on the back of your card. ## The steps 1. Before anything else: notice whether it is asking you to hurry. Suspended accounts, missed deliveries, fines, unpaid fees. Urgency is the mechanism, not a coincidence. *You should see:* You should be able to point at the sentence that is trying to make you move fast. Almost every fake has one. 2. Look at the sender's full address, not the display name. On a phone, tap the name to expand it. *You should see:* You should see the real address. Look at the part just before the first single slash. That is the actual organisation. Everything before it can say anything. 3. Do not tap the link. Instead, go to the organisation the way you normally would — type the address yourself, use the app, or ring the number on the back of your card. *You should see:* You should be able to see whether the thing the message claims is real. If there is a genuine problem, it will be there when you log in normally. 4. If it is fake: forward emails to report@phishing.gov.uk, forward texts to 7726, then delete it. *You should see:* You should get an automatic acknowledgement from the email address. 7726 is free on every UK network. ## Try this one first A text: "Royal Mail: your parcel is held pending a £2.99 fee. Pay here to arrange redelivery: rm-redelivery-uk.com/pay" Urgency, a small believable amount, and a link. The address is not royalmail.com. The real organisation name sits just before the first single slash, and here it is rm-redelivery-uk.com. Forward to 7726, delete. If you genuinely are expecting a parcel, check in the app you already have. ## If it goes wrong - **It is from a name you recognise.** Display names are typed by whoever sent the message. The address underneath is the part that is hard to fake. Always expand it. - **You already clicked the link.** Clicking alone is usually survivable. If you entered a password, change it now on the real site and anywhere else you used the same one. If you entered card details, ring your bank on the number on the card. - **It looks completely genuine and you cannot tell.** Then use the rule that does not need you to tell: never act through the route the message provided. Go the way you normally go. That works even when the fake is perfect. ## Do today Take the last message that asked you to do something urgently. Run the four steps. Report it if it fails, and tell one other person what you found. **Where else this works.** The rule — never act through the route the message provided — works for phone calls, for somebody at the door, and for a message from a friend's account asking for money. **Covers:** Department for Education — Being safe and legal online, Level 1 ## Sources - Ipsos for Lloyds Banking Group, *Essential Digital Skills 2025*, Fieldwork 5 March – 2 April 2025 — https://www.ipsos.com/en-uk/lloyds-essential-digital-skills-2025 - Ofcom, *Adults' Media Use and Attitudes 2026*, 2 April 2026 — https://www.ofcom.org.uk/media-use-and-attitudes/media-habits-adults/adults-media-use-and-attitudes/ --- Checked by a person · not auto-published · howtobedigital.com/learn/digital-skills/tell-a-real-message-from-a-fake-one/ · CC BY-NC-SA 4.0 --- # Ask it something daft and see what happens `/learn/play/ask-it-something-daft/` · Digital Play · Level 0 · **8 minutes** (planned, not yet measured) · Last checked 14 July 2026 > **Draft.** The steps have not yet been performed end to end on a real machine and the time is planned rather than measured. Four of twelve checks outstanding. **After this you will be able to** use one of these tools once, with nothing of yours in it, so the first time is not at work. **Skip this if** Skip this if you have already used one for anything at all. This one is only for the very first time. There is no right answer in this lesson and nothing to get wrong. That is deliberate. You cannot learn something you are frightened of failing at, so the first go happens here, where failure is not available. *People laugh during this one, which is not something that happens much in digital skills training.* ## What you need - Any device with a browser. - No account. No cost. Nothing installed. - Nothing of yours goes into it. That is the whole point. ## The steps 1. Open any free chatbot in a browser. Do not sign up. *You should see:* You should see a box to type in, usually at the bottom of the screen, and some suggested questions above it. 2. Ask it to explain something you already understand perfectly, in the style of somebody who does not. "Explain a kettle as if you are a Victorian ghost." *You should see:* You should get something silly and quite well written. Notice how confident it sounds about a thing that is complete nonsense. 3. Ask it to do the same thing again, worse. "Now do it badly." *You should see:* It will do that too. It will do anything you ask, which is both the useful part and the problem. 4. Ask it a question about itself: "what are you bad at?" *You should see:* You should get a list. Some of it will be accurate and some will be the answer it has been trained to give. Both are worth seeing. ## Try this one first "Write a short poem about a stapler that has given up." You will get a poem. It might even be quite good. The thing to notice is that it did not hesitate, ask what you meant, or say the request was odd — and that same lack of hesitation is what it brings to a question it should have refused. ## If it goes wrong - **You feel silly doing this.** Everybody does for about two minutes. Then most people ask it four more things. - **It refuses your daft request.** Ask it why, and then ask it something adjacent. Finding the edges is a genuine part of the lesson, not a detour. - **You want to ask it something real.** Not in this lesson. Go and do lesson 3 in AI Fundamentals first, the one about what is safe to put in — and then ask whatever you like. ## Do today Ask it one thing with no right answer and no consequence. Read the whole answer. Notice how sure of itself it sounds. **Where else this works.** Trying something once, somewhere nothing is at stake, is how to meet every tool you get handed from now on. --- Checked by a person · not auto-published · howtobedigital.com/learn/play/ask-it-something-daft/ · CC BY-NC-SA 4.0 --- # Paths An ordered run through some lessons. Not a course, nothing to enrol on. - [Is it safe?](/paths/is-it-safe/) — 3 lessons, about 37 minutes. You use a computer at work, somebody has switched an AI tool on, and nobody has told you what you are allowed to put in it. - [AI at work](/paths/ai-at-work/) — 4 lessons, about 44 minutes. You have used one of these tools once or twice and you want to be better at it, and safer. - [Have a go](/paths/have-a-go/) — 3 lessons, about 26 minutes. You have not used one of these and you would rather the first time was not at work with somebody watching. --- # Is it safe? You use a computer at work, somebody has switched an AI tool on, and nobody has told you what you are allowed to put in it. This is not for developers and it is not about building anything. 3 lessons, about 37 minutes. ## The lessons, in order 1. [What AI at work actually means](/learn/ai/what-ai-at-work-actually-means/) — 8 min 2. [Is it safe to put this in?](/learn/ai/is-it-safe-to-put-this-in/) — 14 min 3. [Tell a real message from a fake one](/learn/digital-skills/tell-a-real-message-from-a-fake-one/) — 15 min ## When you finish - Tell a chatbot, an assistant and automation apart - Decide whether a piece of information is safe to give an AI tool - Check whether a message asking you to do something is genuine --- # AI at work You have used one of these tools once or twice and you want to be better at it, and safer. Not for anybody who has never opened one. Do 'Have a go' first. 4 lessons, about 44 minutes. ## The lessons, in order 1. [What AI at work actually means](/learn/ai/what-ai-at-work-actually-means/) — 8 min 2. [Is it safe to put this in?](/learn/ai/is-it-safe-to-put-this-in/) — 14 min 3. [Asking for something and getting it first time](/learn/ai/asking-for-something-and-getting-it-first-time/) — 12 min 4. [Checking the answer before you send it](/learn/ai/checking-the-answer-before-you-send-it/) — 10 min ## When you finish - Decide what is safe to hand over - Write a request that gets something useful back first time - Run three checks on anything it writes for you --- # Have a go You have not used one of these and you would rather the first time was not at work with somebody watching. Not for anybody who already uses one weekly. 3 lessons, about 26 minutes. ## The lessons, in order 1. [Ask it something daft and see what happens](/learn/play/ask-it-something-daft/) — 8 min 2. [Try one for the first time without giving anything away](/learn/ai/try-a-chatbot-for-the-first-time/) — 10 min 3. [What AI at work actually means](/learn/ai/what-ai-at-work-actually-means/) — 8 min ## When you finish - Use one of these tools without anything of yours in it - Say what these tools are actually like, from having poked one - Tell a chatbot, an assistant and automation apart --- # What changed What actually changed in the tools working adults use, and whether it matters. Primary sources only. Fortnightly. - [week of 3 August 2026](/changed/2026-08-03/) — The government published a new national standard for essential digital skills in July, and almost nothing is taught against it yet. --- # What changed — week of 3 August 2026 4 minutes **If you read nothing else.** The government published a new national standard for essential digital skills in July, and almost nothing is taught against it yet. **What changed.** The Department for Education published Essential digital skills standards: 2026 on 15 July, updated and expanded to include artificial intelligence. **Does it matter to you?** Only if you buy or run training **What to do.** Nothing today. If you commission training, start asking suppliers which standard they teach against, and expect vague answers for the next year. *Source:* Department for Education, *Essential digital skills standards: 2026*, 15 July 2026 — https://www.gov.uk/government/publications/essential-digital-skills-standards-2026 **What changed.** The ONS reported that 35% of UK businesses with ten or more employees now use at least one AI technology, while 11% say more than half their workforce has had any AI training. **Does it matter to you?** Yes **What to do.** If your workplace switched something on without training anybody, you are in the gap between those two numbers, and you are not unusual. *Source:* Office for National Statistics, *Artificial intelligence in UK businesses: 2023 to 2026*, 20 July 2026 — https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026 **What changed.** The Social Tenant Access to Information Requirements commence for private registered providers on 1 October 2026, with individual request rights from 1 April 2027. **Does it matter to you?** Only if you rent from a housing association, or work for one **What to do.** If you are a tenant, nothing yet — the right arrives in October. If you work for a landlord, the question worth asking now is how a tenant who is not online will use it. *Source:* Ministry of Housing, Communities and Local Government, *Social Tenant Access to Information Requirements: policy statement*, Checked 8 August 2026 — https://www.gov.uk/government/consultations/social-tenant-access-to-information-requirements-consultation/outcome/social-tenant-access-to-information-requirements-policy-statement ## Not worth your time this week - Another model scored higher on another benchmark. It does not change anything you do on Monday. - A large amount of coverage of a product that is not available in the UK. --- # The numbers Every figure used on this site, with its source and date — including the ones that were checked and thrown out. ## The numbers used here ### Artificial intelligence in UK businesses: 2023 to 2026 Office for National Statistics · 20 July 2026 · https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026 - 35% of UK businesses with 10+ employees use at least one AI technology (June 2026) - 28% of businesses with 0–9 employees; 49% of businesses with 250+ - 11% report more than half their workforce has received AI training *Official statistics. Business Insights and Conditions Survey. Always state the employee-size base in the same sentence as the figure.* ### Essential Digital Skills 2025 Ipsos for Lloyds Banking Group · Fieldwork 5 March – 2 April 2025 · https://www.ipsos.com/en-uk/lloyds-essential-digital-skills-2025 - 8.0 million UK adults (15%) cannot do all eight Foundation Level tasks - 4.0 million (8%) lack Essential Digital Skills for Life - 7.1 million labour-force adults (17%) lack them for Work - 53% Foundation Level attainment among adults with no formal qualifications, against 93% of graduates *n=4,175. The 53%/93% split is drawn from the full report rather than the published summary page and is pending a page-number check.* ### The Economic Impact of Closing the Work Essential Digital Skills Gap Cebr for FutureDotNow, supported by DSIT · March 2025 · https://futuredotnow.uk/about-us/the-essential-digital-skills-gap/ - 52% of the UK labour force cannot complete all 20 Work Essential Digital Skills tasks - Closing the gap estimated at £23.1 billion a year in GVA *Having 'Essential Digital Skills for Work' requires one task in each of five areas. That is why 83% 'have' them while 52% cannot do all twenty. Both are true.* ### Adults' Media Use and Attitudes 2026 Ofcom · 2 April 2026 · https://www.ofcom.org.uk/media-use-and-attitudes/media-habits-adults/adults-media-use-and-attitudes/ - 54% of UK adults use AI tools, rising to 79% of 16–24s - 20% of online UK adults depend exclusively on a smartphone - 89% feel confident online overall, but only 72% are confident judging accuracy *The PDF returns 403 to automated link checks. Verified by hand.* ### Public attitudes to AI King's College London Policy Institute and King's Institute for AI · Published 19 May 2026 · https://www.kcl.ac.uk/news/one-in-five-britons-think-ai-will-create-civil-unrest-study-finds - 69% of workers are worried about AI's economic impact - 41% of the public say they are afraid of AI *Fieldwork by Opinium, 16–29 April 2026, n=4,508.* ### Essential digital skills standards: 2026 Department for Education · 15 July 2026 · https://www.gov.uk/government/publications/essential-digital-skills-standards-2026 - Five levels, Entry 1 to Level 2, updated and expanded in response to technological change, including artificial intelligence *Crown copyright. The qualifications behind it have not been rewritten and no implementation date has been set. Read the licence statement before adapting rather than citing.* ### AI foundation skills for work Skills England · 28 January 2026 · https://www.gov.uk/government/publications/ai-foundation-skills-for-work - Three domains: technical, non-technical, and responsible and ethical AI foundation skills, with six competencies across them *Open Government Licence v3.0 — reusable and adaptable, including commercially, with attribution. The exact competency-to-domain mapping is pending a check against the published benchmark.* ### Social Tenant Access to Information Requirements: policy statement Ministry of Housing, Communities and Local Government · Checked 8 August 2026 · https://www.gov.uk/government/consultations/social-tenant-access-to-information-requirements-consultation/outcome/social-tenant-access-to-information-requirements-policy-statement - Publication scheme requirement from 1 October 2026 - Individual information request rights from 1 April 2027 - Applies to private registered providers of social housing *Tenant is defined broadly, including licensees and shared owners holding under 100% equity.* ### Employer Skills Survey 2024 Department for Education · Published 24 July 2025 · https://explore-education-statistics.service.gov.uk/find-statistics/employer-skills-survey - UK employer training investment fell to £53.0bn in 2024, down 10.2% in real terms from £59.0bn in 2022 - Per-employee spend fell from £1,960 to £1,700 *22,712 employers. Publishes no headline digital or AI skills gap figure.* ### AI Skills Boost expansion Department for Science, Innovation and Technology · 28 January 2026 · https://www.gov.uk/government/news/free-ai-training-for-all-as-government-and-industry-programme-expands-to-provide-10-million-workers-with-key-ai-skills-by-2030 - Target of 10 million workers upskilled in AI by 2030 - Including at least 2 million SME employees *The '21% of workers feel confident using AI at work' figure associated with this release could not be located in a published data table and is not used here.* ## Numbers used elsewhere that are not used here **Only 21% of UK workers feel confident using AI at work.** Appears in the DSIT AI Skills Boost release of 28 January 2026. It could not be located in a published data table, and the fieldwork behind the research programme completed in March 2025 — sixteen months before it was quoted. The ONS training figure is six days old and traceable, so that is used instead. **46% of small businesses lack the skills or knowledge to use AI effectively.** Same release, attributed to the FSB in secondary coverage. The underlying publication could not be found. **60% of employers cite AI skill gaps.** This figure does not exist in the survey it is attributed to. The actual figures are 25% and 27%, on a base of 373 employers who use or plan to use AI, not all employers. The likely origin is a separate line about six in ten employers not using AI and not planning to, which is a non-adoption figure rather than a skills figure. **3.8 million lack essential digital skills for life; 7.3 million for work.** Still widely repeated in 2026 government and sector documents. These come from the 2024 Consumer Digital Index and are superseded by the 2025 figures of 4.0m and 7.1m. **Any ONS figure for internet non-use.** The ONS Internet users series was discontinued; a notice dated 3 May 2023 directs users to Ofcom. The last data is from 2020. **Any North East-specific digital skills statistic.** The most recent regional work relies on the discontinued ONS series and cannot be refreshed. There is no current North East measure of digital skills. National figures are used here and labelled national. ## What is not known - There is no official UK measure of individual adults' AI skills against a defined framework. The ONS measures business adoption. Nobody measures people. The first real measurement arrives with the DSIT survey in 2027. - There is no current regional breakdown of digital skills below nation level. - The Employer Skills Survey 2024 — 22,712 employers, the most authoritative UK employer survey — publishes no headline digital or AI skills gap figure. - There is almost no independent evidence that digital skills training works at scale. The widely-quoted £9.48-return-per-£1 figure is one organisation's evaluation of its own programmes. --- # Sources Generated at build time. Every source ID on the site resolves here. - Office for National Statistics, *Artificial intelligence in UK businesses: 2023 to 2026*, 20 July 2026 — https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026 - Ipsos for Lloyds Banking Group, *Essential Digital Skills 2025*, Fieldwork 5 March – 2 April 2025 — https://www.ipsos.com/en-uk/lloyds-essential-digital-skills-2025 - Cebr for FutureDotNow, supported by DSIT, *The Economic Impact of Closing the Work Essential Digital Skills Gap*, March 2025 — https://futuredotnow.uk/about-us/the-essential-digital-skills-gap/ - Ofcom, *Adults' Media Use and Attitudes 2026*, 2 April 2026 — https://www.ofcom.org.uk/media-use-and-attitudes/media-habits-adults/adults-media-use-and-attitudes/ - King's College London Policy Institute and King's Institute for AI, *Public attitudes to AI*, Published 19 May 2026 — https://www.kcl.ac.uk/news/one-in-five-britons-think-ai-will-create-civil-unrest-study-finds - Department for Education, *Essential digital skills standards: 2026*, 15 July 2026 — https://www.gov.uk/government/publications/essential-digital-skills-standards-2026 - Skills England, *AI foundation skills for work*, 28 January 2026 — https://www.gov.uk/government/publications/ai-foundation-skills-for-work - Ministry of Housing, Communities and Local Government, *Social Tenant Access to Information Requirements: policy statement*, Checked 8 August 2026 — https://www.gov.uk/government/consultations/social-tenant-access-to-information-requirements-consultation/outcome/social-tenant-access-to-information-requirements-policy-statement - Department for Education, *Employer Skills Survey 2024*, Published 24 July 2025 — https://explore-education-statistics.service.gov.uk/find-statistics/employer-skills-survey - Department for Science, Innovation and Technology, *AI Skills Boost expansion*, 28 January 2026 — https://www.gov.uk/government/news/free-ai-training-for-all-as-government-and-industry-programme-expands-to-provide-10-million-workers-with-key-ai-skills-by-2030 --- # For organisations 35% of UK businesses with ten or more staff use AI. 11% say more than half their workforce has had any training on it. The teaching here is the gap between those two numbers. Source: Office for National Statistics, *Artificial intelligence in UK businesses: 2023 to 2026*, 20 July 2026 — https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026 ## Three things people ring about **"The board asked how tenants who aren't online will get information under the new requirements. There wasn't a good answer."** The Social Tenant Access to Information Requirements take effect for private registered providers from 1 October 2026, with individual request rights from 1 April 2027. Here is the plain version of what that means for tenants who cannot get online, and what can be done in the time left. Free, no form. [Read it](/organisations/stairs/) **"We bought the licences a year ago. The usage reports are embarrassing and finance has noticed."** The most common call. Before buying any training, here are eleven questions worth asking and what each answer tells you. Anybody who can answer all eleven does not need help. [The questions](/organisations/unused-licences/) **"Half the team are frightened of it and the other half are already using it without telling us."** Sixty-nine per cent of UK workers are worried about what AI does to the economy. A manager who says nothing is not being neutral — the silence gets filled. Here is what to say when you do not know yet. [Read it](/learn/horizon/) ## What is on offer ### One day, one team — £1,800 fixed up to 20 people · on site A single day. Afterwards the team can decide whether a task is safe to give an AI tool, write a request that gets something useful back, and check an answer before it goes to anybody. - Written summary for people who could not attend - Practice cards for everybody, printable, yours to copy - Four weeks' notice ### AI at work, six weeks — £3,600 fixed up to 12 people · North East or online Six sessions of ninety minutes. Nine hours of contact time. What is being measured is agreed at the start and reported against at the end, rather than against attendance. - £300 a head at full group size - No day rates, no expenses within the North East - Four weeks' notice ## What is not on offer Not an accredited qualification. Not an awarding organisation. No e-learning licences, no platform, no hundred-person webinars. General users and the people who support them, not specialists or developers. --- # Your landlord has a deadline on 1 October From 1 October 2026 tenants of private registered providers get a new right to see information about how their home is managed. From 1 April 2027 they can request specific information. To use either they will almost certainly need to be online. Not everybody is. Source: Ministry of Housing, Communities and Local Government, *Social Tenant Access to Information Requirements: policy statement*, Checked 8 August 2026 — https://www.gov.uk/government/consultations/social-tenant-access-to-information-requirements-consultation/outcome/social-tenant-access-to-information-requirements-policy-statement ## What the requirement actually is A publication scheme from 1 October 2026 — information published proactively. Individual information requests from 1 April 2027. Tenant is defined broadly, including licensees and shared owners holding under 100% equity. ## The bit that is nobody's job yet 8 million UK adults cannot do all eight basic-level tasks. The sharpest predictor is qualifications, not age: 53% attainment among adults with no formal qualifications against 93% of graduates, and 71% among adults not in paid work. Those groups are heavily represented in social housing. Source: Ipsos for Lloyds Banking Group, *Essential Digital Skills 2025*, Fieldwork 5 March – 2 April 2025 — https://www.ipsos.com/en-uk/lloyds-essential-digital-skills-2025 ## Four things a landlord can do before October 1. Find out how many tenants have logged into your portal this year — not how many are registered. 2. Decide what the non-digital route is, write it down, and publish it alongside the scheme. 3. Check the published information is readable by somebody online but not confident. 20% of online UK adults depend exclusively on a smartphone. 4. Brief whoever answers the phone, before October rather than after the first request. --- # Eleven questions before you buy any AI training Published because anybody who can answer all eleven does not need help. 1. How many people have a licence, and how many opened it last month? Not registered — opened. 2. Who told them it had arrived, and what did that message say? 3. What is the first useful thing somebody in your busiest role could do with it, in under ten minutes? 4. Has anybody written down what staff are allowed to put in it? 5. If somebody pasted a customer's details into it this morning, would you know? 6. Who signed off the tool, and what did they actually sign off — the software, or the way it is used? 7. What did the last training you bought change? Name the behaviour, not the attendance. 8. Which team is quietly using something you did not buy, because the thing you bought is harder? 9. When something it wrote goes out with a mistake in it, whose name is on the letter? 10. What would you measure in six weeks to know this worked? 11. If the answer to question ten is 'usage', what will you do when usage goes up and nothing else does? ## What the answers usually reveal Nobody was told what it was for. There are no rules, so people made their own. Nobody agreed what success looks like, so the only available measure is usage — and usage going up is not the same as work getting better. --- # For teachers and helpers For library staff, Digital Champions, union learning reps, Citizens Advice volunteers and anybody running a drop-in. Free to copy, print and hand out under CC BY-NC-SA 4.0. ## Which sheet for which situation - **Somebody has been scammed, or thinks they have** → [Tell a real message from a fake one](/practice/tell-a-real-message-from-a-fake-one/) - **Somebody has lost a document and is upset about it** → [Find a file you saved and cannot find again](/practice/find-a-file-you-have-lost/) - **Somebody is frightened of AI and has never touched it** → [Ask it something daft and see what happens](/practice/ask-it-something-daft/) - **Somebody has been told to use AI at work and is worried about the rules** → [Is it safe to put this in?](/practice/is-it-safe-to-put-this-in/) --- # Practice cards One page of A4 per lesson. No email address needed. CC BY-NC-SA 4.0. - [What AI at work actually means](/practice/what-ai-at-work-actually-means/) - [Try one for the first time without giving anything away](/practice/try-a-chatbot-for-the-first-time/) - [Is it safe to put this in?](/practice/is-it-safe-to-put-this-in/) - [Asking for something and getting it first time](/practice/asking-for-something-and-getting-it-first-time/) - [Checking the answer before you send it](/practice/checking-the-answer-before-you-send-it/) - [Find a file you saved and cannot find again](/practice/find-a-file-you-have-lost/) - [Tell a real message from a fake one](/practice/tell-a-real-message-from-a-fake-one/) - [Ask it something daft and see what happens](/practice/ask-it-something-daft/) --- # Working notes Notes on building and running this. Corrections, reversals, decisions and things that went wrong. - [A figure we could not trace, and why it is not on this site](/blog/the-figure-i-could-not-trace/) — 2026-08-08. One statistic, one afternoon, and a decision that shapes everything else here. - [Nothing here has passed the person-checks yet](/blog/nothing-here-is-signed-yet/) — 2026-08-06. Eight lessons are published. Every one of them says, at the top, that it is a draft. Here is what that admission actually means. - [Twelve checks. Six of them a computer cannot do.](/blog/twelve-checks/) — 2026-08-04. The whole quality bar, published, including the bits that are currently failing. --- # A figure we could not trace, and why it is not on this site 2026-08-08 · 4 minutes *One statistic, one afternoon, and a decision that shapes everything else here.* There is a figure in circulation that says only 21% of UK workers feel confident using AI at work. It appears in a government press release from January. It is quoted widely. It is a good figure for anybody selling training, which is what this site does on the paid side. It is not used here, and this is the page that says why. ## What the search turned up The figure is associated with a research programme published on 28 January 2026. The report exists. The programme exists. What could not be found was the figure itself in a published data table — the thing you would point at if somebody in a meeting asked where it came from. Then the fieldwork date. The research behind that programme completed in March 2025. By the time the figure was being quoted as current, the answers behind it were sixteen months old, in the fastest-moving area anybody teaches. ## What is used instead The Office for National Statistics reported in July 2026 that 35% of UK businesses with ten or more employees use at least one AI technology, and that 11% say more than half their workforce has had AI training. Same release, same base, published with the survey wave and the response rate attached. Two numbers, six days old at the time of writing, and you can click through to the table. That makes the same argument and it survives being challenged. ## Why this matters more than the statistic Everybody quotes digital skills numbers. Almost nobody says where they came from, and a striking number of the most-repeated figures turn out to be a misreading of a line in a report nobody went back to. So the rule here is: every figure carries an organisation, a publication, a date and a link. Anything that cannot be traced gets cut, and the notable ones get listed with the reason. That list is on the evidence page, and it is the part of this site most worth reading if you are deciding whether to trust the rest of it. --- # Nothing here has passed the person-checks yet 2026-08-06 · 5 minutes *Eight lessons are published. Every one of them says, at the top, that it is a draft. Here is what that admission actually means.* There are eight lessons on this site and not one of them has passed the six checks that need a person. Every lesson page says so, at the top, in a box you cannot miss. The time estimate says PLANNED, NOT YET MEASURED. The check block says which boxes are unticked. That is deliberate and it is not modesty. ## What the claim actually commits to "Checked by a person" means five things, and it is worth being exact because a claim that turns out to be soft is worse than no claim at all. - Somebody has performed every step described, on a real machine, and watched what happened. - Somebody has checked every factual claim against a primary source, and every source is listed. - Somebody has timed the lesson with a stopwatch and published the number they got. - Somebody is accountable — there is an address, a person reads it, and corrections get published rather than edited in quietly. - Somebody's judgement decided what went in and what was cut. None of the eight has met all five. The steps were drafted rather than performed. The times are planned rather than measured. The failures in the 'if it goes wrong' sections are the likely ones rather than the ones that were actually hit. ## Why publish at all, then Two options were available. Hold everything back until it is all done, and have nothing to show for months. Or publish with the gap marked, so anybody reading can see exactly what state it is in. The second one is more useful and it is also more honest. A draft lesson that says it is a draft is worth more than a finished-looking one that quietly is not. ## What happens next Each lesson gets performed end to end, on a normal machine, at the pace of somebody meeting it for the first time. The planned minutes get replaced with the measured ones. The real failures replace the imagined ones. Then the draft block comes off, one lesson at a time. This post stays up afterwards, at this address, with a note. It is a better record than deleting it would be. --- # Twelve checks. Six of them a computer cannot do. 2026-08-04 · 6 minutes *The whole quality bar, published, including the bits that are currently failing.* A lesson goes live here when twelve checks pass. Six of them run in about four seconds. The other six take an afternoon and cannot be automated, which is the reason this site will always be small. ## The six a computer does Required blocks present. Reading grade under 13.0, warning above 11.0. No banned words and no voice tics. Every statistic resolves to a source that exists. Every link alive, and a warning when a lesson is past its stale window. Brand rules — two primary buttons on a page, a rounded corner, a video without a transcript, all fail the build. None of those checks is clever. They are all things that quietly rot over three months if nobody is watching, which is exactly what an automated check is for. ## The six that need somebody - Is the outcome observable, or does it say "understand"? - Is the title a moment somebody could overhear in a break room, or a topic? - Are the entry criteria honest, or written so nobody would ever skip? - Was the time measured with a stopwatch, or estimated? - Does every step say what you should see afterwards? - Are the failures real ones, hit while doing it, or imagined ones? These are the six that erode first. They are also the six a better-funded competitor cannot buy their way past, because each of them requires somebody to have actually done the thing. ## Today's build The quality bar page shows the current build output, including the warnings. Right now that means two warnings and four unticked person-checks on every lesson. Showing the failing ones is worth more than showing the passing ones. --- # Accessibility WCAG 2.1 AA in both themes, tested separately. ## What has been done - Semantic HTML, one h1 per page, no skipped heading levels. - Skip-to-content link. Focus visible everywhere and never removed. - Colour is never the only signal. - Text resizes to 200%, reflows at 320px. - Real alt text on content images. Every video requires a transcript. - Works with JavaScript off. Respects prefers-reduced-motion. No cookie banner. ## What is known not to conform yet - No manual screen reader pass on the lesson template. - No text-to-speech. - Reading age measured automatically but not checked by an independent person. - English only. Tell us: hello@howtobedigital.com --- # Privacy No cookies. One third-party script: Google Analytics in its cookieless mode — consent is denied by default and never granted, so no cookie is set and nothing identifies you. It produces anonymous page counts. No tag manager, no session recording, no social widgets. No cookie banner, because there is still nothing stored on your device. Video is facade-loaded — nothing is requested from YouTube until you press play, and the embed is youtube-nocookie. The only personal data collected is what you type into an email if you choose to send one. There is no form on this site that stores anything. hello@howtobedigital.com --- # Licence CC BY-NC-SA 4.0 unless a page says otherwise. Copy it, print it, adapt it, share it. Say where it came from, do not sell it, keep the same licence. The practice cards are meant to be copied. The method is published under the same terms. --- --- ## Find your starting point Twelve statements about your circumstances, not a test. Tick the one that is true for you today. You get the area to start with and the next three lessons, in order, with the time each one takes. Nothing is sent and nothing is stored. The result lives in the page's web address, so you can bookmark it or close the tab and lose it. The check runs in the browser and needs JavaScript. Nothing else on this site does. If you would rather not turn it on, start at [all lessons](/learn/) or take one of the three [paths](/paths/). ## The four areas - **Everyday digital** — finding a file, sharing a document, passwords. - **AI Fundamentals** — what these tools are, how to ask, how to check what comes back. - **Copilot and Agents** — what your employer bought, what it can see, what an agent is. - **Safety and judgement** — a message that wants something urgently, what is safe to paste in, what your organisation's rules actually say. Each area is scored 0 to 3, matching the lesson levels used across the site. The lowest area is where you start. Two more questions — how much time you have in a week, and whether you can practise on your own device — change what is suggested, not how you are described. ## For a team The same page runs a second check, six questions about an organisation rather than a person, and returns one of five bands: Unaware, Aware, Trying, Embedded, Leading. The lowest answer pulls the band down, because an organisation is only as far on as the person who cannot take part. It is a conversation starter, not a score to report upwards. See [for organisations](/organisations/) for what a programme includes and costs. Take the check at [howtobedigital.com/start/](/start/). --- ## What could an agent do here? A thinking board, not a lesson. Four piles of tiles and one sentence frame: > When **[a situation from your week]**, an agent could **[do something]** > using **[something you already have]**, so that **[something changes]**. Most conversations about agents start with the technology and work backwards to a problem, which is how organisations end up owning software nobody uses. This does it the other way round. Read the sentence out loud. Some will sound daft — a sentence that sounds daft in a meeting was never a plan. Works best with a team, out loud, in about twenty minutes. Nothing is sent anywhere and nothing is stored; print the sheet to keep it. ## The four piles **When this happens** — twelve situations from ordinary UK working weeks: a form that sits for three days, a policy where the answer is on page 41, a rota only one person understands, somebody checking every single one just in case. **An agent could** — ten things worth asking for: read it and pull out what matters, draft a first version for a person to check, flag only the ones that need a human, fill in the parts it can and stop. **Using** — ten things an organisation already has: SharePoint documents, the shared mailbox, a Teams channel, the policy library, the forms people already fill in. **So that** — eight outcomes: nobody waits three days, the answer is the same whoever asks, a person only sees the exceptions, one person stops being the only one who knows. ## Before anybody buys anything A sentence you like is not a business case. Three things decide whether it is worth doing, and none of them are on the board: - How often does it actually happen? Count it for a week before you guess. - What happens when it gets it wrong? If somebody is refused a repair, a person checks every one, and the saving is smaller than the pitch said. - Who is left out? If the person doing the job today cannot use the thing that replaces it, the work has moved, not gone. The useful next step is usually not a pilot. It is finding out where the team actually is and reading the floor rather than the average — see [find your starting point](/start/#team). If the word "agent" means nothing yet, start with [What AI at work actually means](/learn/ai/what-ai-at-work-actually-means/), eight minutes. Use the board at [howtobedigital.com/agents/](/agents/). --- --- --- ## Explainers Twenty-six picture-led pages. One idea per screen, a big picture on every screen, under a hundred and twenty words on the whole page. These are not lessons. There is nothing to follow and nothing to complete, just one small thing to do at the end. They are for the moment before a lesson, when you need to see what the thing actually is. Each is a single self-contained page. Save it, print it, or open it on a phone with no signal. Nothing is stored and nothing is sent. ## Digital skills foundation The five areas the European Commission named in DigComp 3.0, one page each. Start here if the phrase “digital skills” has never been pinned down for you. - [What digital skills means](/explain/what-digital-skills-means/index.md) — The five areas of DigComp 2.2 drawn once, so the reader can see which one they are worried about. 7 screens. - [Finding things, and telling what is true](/explain/finding-things-and-telling-what-is-true/index.md) — Information and data literacy: searching, judging a source, keeping what you found. 7 screens. - [Working with other people online](/explain/working-with-other-people-online/index.md) — Communication and collaboration: which channel, who can see it, what you leave behind. 7 screens. - [Making something](/explain/making-something/index.md) — Digital content creation: making, changing, and who owns it. 7 screens. - [Keeping yourself safe](/explain/keeping-yourself-safe/index.md) — Safety: devices, personal data, wellbeing, and the environment. 7 screens. - [Getting unstuck](/explain/getting-unstuck/index.md) — Problem solving: what to do when the thing does not work, and knowing what you do not know. 7 screens. ## Microsoft 365, apps and services What the subscription actually is, where your files really go, and what your employer can and cannot look at. - [What Microsoft 365 is](/explain/what-microsoft-365-is/index.md) — A subscription, not a program. What arrives with it and what does not. 7 screens. - [Where your files really live](/explain/where-your-files-live/index.md) — OneDrive, SharePoint and Teams are one place wearing three coats. 7 screens. - [A chat is not an email](/explain/a-chat-is-not-an-email/index.md) — Teams and Outlook: what each is for, what is kept, who can find it later. 7 screens. - [What your organisation can see](/explain/what-your-organisation-can-see/index.md) — What an admin can and cannot look at, said plainly, without frightening anybody. 7 screens. ## Microsoft's AI stack How the pieces sit together, which Copilot you have, and why the same question gives a different answer at work. - [The Microsoft AI stack](/explain/the-microsoft-ai-stack/index.md) — One picture: the model underneath, Copilot on top, agents beside it, your data behind it. 7 screens. - [Which Copilot have you got?](/explain/which-copilot-have-you-got/index.md) — The free one, the paid one, and how to tell which is in front of you. 7 screens. - [What grounded means](/explain/what-grounded-means/index.md) — Why the same question gives a different answer at work than at home. 7 screens. ## Copilot The one included with Microsoft 365, and how to ask it for something and get it. - [Copilot Chat](/explain/copilot-chat/index.md) — What it is, where it is, what it can see, what it cannot. 7 screens. - [Asking for something](/explain/asking-for-something/index.md) — Task, audience, length, example. Four parts of a request, drawn. 7 screens. ## Microsoft 365 Copilot The licensed one: in the apps, what it can see, and what to check before you send anything. - [Copilot inside the apps](/explain/copilot-inside-the-apps/index.md) — The same assistant in three places, doing three different jobs. 7 screens. - [What Copilot can see](/explain/what-copilot-can-see/index.md) — It sees what you can already see, and no more. Permissions in one picture. 7 screens. - [Checking what it gives you](/explain/checking-what-it-gives-you/index.md) — Names, dates, figures. Three checks before anything leaves your hands. 7 screens. ## Microsoft pre-built agents The agents Microsoft ships: what one is, and where to find them. - [What a ready-made agent is](/explain/what-a-ready-made-agent-is/index.md) — Somebody else already made it, gave it a job, and pointed it at some knowledge. 7 screens. - [Where the ready-made ones are](/explain/where-the-ready-made-agents-are/index.md) — Where to find the agents Microsoft ships, and how to tell one from a chatbot. Re-check names before writing. 7 screens. ## Copilot in SharePoint Asking a single site a question, and why your colleague gets a different answer. - [Asking a site a question](/explain/asking-a-site-a-question/index.md) — Copilot in SharePoint: point at a site, ask, get an answer with a link back. 7 screens. - [Why two people get different answers](/explain/why-two-people-get-different-answers/index.md) — Permissions decide the answer. The same question, two people, two results. 7 screens. ## Agent Builder agents Making your own without code, and telling it what to refuse. - [Making one yourself](/explain/making-an-agent-yourself/index.md) — Agent Builder: a name, a job description, some knowledge, a test. 7 screens. - [Knowledge, and edges](/explain/knowledge-and-edges/index.md) — What you point it at, and telling it what to refuse. 7 screens. ## SharePoint agents An agent that is a file, and what happens when you share it. - [A SharePoint agent is a file](/explain/a-sharepoint-agent-is-a-file/index.md) — It sits in Site Assets as a .agent file, and file permissions decide everything. 7 screens. - [Who can use the one you made](/explain/who-can-use-the-one-you-made/index.md) — Sharing an agent is sharing a file, and the answers still follow the reader's own permissions. 7 screens. ## How these are made Every picture is drawn from six shapes: a box, a line, an arrow, a square dot, a dashed frame and a label. No photographs, no icons, no stock. Green marks the thing being explained; amber marks the thing you do. A page cannot publish until it passes a checker that counts the words, the screens and the colours, and refuses anything without a description on every picture. ## Where the facts come from The five digital skills areas are quoted from DigComp 3.0, the European Digital Competence Framework, published by the Joint Research Centre on 11 November 2025: five areas, twenty-one competences. Everything about Microsoft was checked against Microsoft Learn on 22 August 2026. Microsoft renames things often; where a page teaches a name rather than an idea, it says so. Not sure which to read? [Find your starting point](/start/), three minutes. --- ## What digital skills means The five areas of DigComp 2.2 drawn once, so the reader can see which one they are worried about. 7 screens, about three minutes. Pictures, almost no words. ## Two words, nine meanings. *Picture: One large box with the words digital skills inside it.* Everyone uses them. Nobody says which bit. ## Somebody wrote the list down. *Picture: A tall dashed cupboard labelled digital skills, holding five stacked shelves, each named.* The European Commission named five areas. ## Twenty-one things in all. *Picture: The same cupboard with small squares on each shelf: three, six, four, four and four.* You already do some of them every day. ## Nobody is steady on all five. *Picture: The same cupboard with the second shelf drawn in green.* Most people are fine on two and quiet about the rest. ## Start on the quiet one. *Picture: The same cupboard with the fifth shelf in amber and an arrow pointing at it.* Not the one you are best at. ## Do today Say out loud which of the five you would rather not be asked about. --- Read it as a page: [howtobedigital.com/explain/what-digital-skills-means/](/explain/what-digital-skills-means/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## Finding things, and telling what is true Information and data literacy: searching, judging a source, keeping what you found. 7 screens, about three minutes. Pictures, almost no words. ## You need to know something. *Picture: One large box labelled a question.* That is where this area starts. ## Searching is the easy half. *Picture: A box labelled your question with an arrow to a dashed frame holding nine result boxes.* Anyone can get a thousand results. ## Judging it is the hard half. *Picture: Three result boxes side by side, the middle one green, each with a mono label: who wrote it, when, and what for.* Three questions do most of the work. ## Then put it somewhere. *Picture: A box labelled a good answer with an arrow into a dashed frame labelled where you can look again.* A good answer you cannot find again is not an answer. ## Three competences, one habit. *Picture: Three steps in a row with arrows: find, judge, keep. The middle step is amber.* The framework calls this area three things. It is one loop. ## Do today Find one thing you looked up last week and see whether you could find it again. --- Read it as a page: [howtobedigital.com/explain/finding-things-and-telling-what-is-true/](/explain/finding-things-and-telling-what-is-true/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## Working with other people online Communication and collaboration: which channel, who can see it, what you leave behind. 7 screens, about three minutes. Pictures, almost no words. ## Two people, one message. *Picture: Two people drawn as squares, facing each other with a line between them.* The whole area is deciding how. ## The channel is a choice. *Picture: Four boxes in a row labelled a chat, an email, a meeting, a shared document.* Most people use whatever is already open. ## More people see it than you meant. *Picture: One large box labelled a message with a dashed frame around it labelled everyone who can see it.* A channel decides who, and it decides quietly. ## Nothing you send stops existing. *Picture: A row of three boxes labelled sent, stored, searchable, with arrows between them.* It is kept, and findable later. ## Tone does not travel. *Picture: A dashed frame labelled you online holding two boxes: what you meant and what they read.* Short reads as cross. Say the extra sentence. ## Do today Take one message you were about to send in a chat and send it as an email instead, or the other way round. --- Read it as a page: [howtobedigital.com/explain/working-with-other-people-online/](/explain/working-with-other-people-online/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## Making something Digital content creation: making, changing, and who owns it. 7 screens, about three minutes. Pictures, almost no words. ## You make more than you think. *Picture: One large box labelled a thing you made.* A message, a list, a photograph, a form. ## Starting from nothing is rare. *Picture: Two panels side by side. On the left, a blank page. On the right, the same page with three ruled lines and a green border.* Most making is changing something that was already there. ## Somebody made it before you. *Picture: A box labelled somebody else’s work with an arrow into a box labelled your version, and an amber padlock beside the first.* Whether you may reuse it is a rule, not a feeling. ## Three answers, and you can find out which. *Picture: Three boxes in a row labelled anyone may, ask first, and no, with the middle one amber.* The licence is usually written on the thing. ## Put your name on it. *Picture: A dashed frame labelled what you made holding a green box labelled your name.* The same rule protects you. ## Do today Open the last thing you made and check whether your name is anywhere on it. --- Read it as a page: [howtobedigital.com/explain/making-something/](/explain/making-something/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## Keeping yourself safe Safety: devices, personal data, wellbeing, and the environment. 7 screens, about three minutes. Pictures, almost no words. ## Four things, not one. *Picture: One large box labelled you with four small boxes around it.* Safety online is usually taught as passwords only. ## The device. *Picture: A laptop with an amber padlock on it.* Locked, up to date, and yours. ## Your details are worth something. *Picture: A dashed frame labelled things about you holding four boxes: your name, where you live, your birthday, your health.* Somebody wants them. That is why they are asked for. ## How you feel counts too. *Picture: Two panels. On the left a person and a phone close together. On the right the same person and phone apart, the person in green.* The framework says so. Most training leaves it out. ## And it uses something. *Picture: A box labelled one hour of video with an arrow to a box labelled electricity.* Every file you keep sits on a machine somewhere, running. ## Do today Check that the device you are reading this on locks itself when you put it down. --- Read it as a page: [howtobedigital.com/explain/keeping-yourself-safe/](/explain/keeping-yourself-safe/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## Getting unstuck Problem solving: what to do when the thing does not work, and knowing what you do not know. 7 screens, about three minutes. Pictures, almost no words. ## Everybody gets stuck. *Picture: One large box labelled it is not working.* The difference is what happens next. ## There is an order. *Picture: Four boxes in a row: turn it off and on, look at the message, search the message, ask somebody. The first is amber.* Most problems stop at step one or two. ## The message is not noise. *Picture: A cut-away box labelled the error message with three boxes inside: what happened, where, and a number.* It is somebody telling you exactly what went wrong. ## Sometimes nothing is broken. *Picture: Two boxes side by side, one labelled what I need and one labelled what exists, with a line between them.* You just have not met the thing that does it yet. ## Knowing the gap is a skill. *Picture: A dashed frame labelled what I can do, with a green box just outside it labelled the next thing.* It is the last of the twenty-one, and the one that moves you. ## Do today Write down one thing you cannot do yet that would make your week easier. --- Read it as a page: [howtobedigital.com/explain/getting-unstuck/](/explain/getting-unstuck/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## What Microsoft 365 is A subscription, not a program. What arrives with it and what does not. 7 screens, about three minutes. Pictures, almost no words. ## It is not a program. *Picture: One large box with the words Microsoft 365 inside it.* There is nothing on your machine called this. ## You are renting a workshop. *Picture: Two panels. On the left a box labelled buy once. On the right a green box labelled pay monthly.* Stop paying and the tools stop opening. ## Apps, and the space behind them. *Picture: A dashed frame labelled what you rent holding six boxes: Word, Excel, Outlook, Teams, OneDrive, SharePoint.* The storage matters more than the apps. ## The account is the thing. *Picture: A box labelled your organisation with a green box inside it labelled your account.* It decides what opens, and what you can see. ## Not everything is in the price. *Picture: Three boxes in a row labelled included, extra, and not sold, with the middle one amber.* Copilot is the one people assume is included. ## Do today Find out which Microsoft 365 plan your account is on. Somebody in IT will know. --- Read it as a page: [howtobedigital.com/explain/what-microsoft-365-is/](/explain/what-microsoft-365-is/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## Where your files really live OneDrive, SharePoint and Teams are one place wearing three coats. 7 screens, about three minutes. Pictures, almost no words. ## Your file. *Picture: One large box with the words your file inside it.* You saved it somewhere. Where? ## Not on your laptop. *Picture: A laptop on the left, an arrow to the right with a small amber square travelling along it, and a large box labelled one store.* It went to a building you will never visit. ## Three doors, one store. *Picture: Three boxes on the left labelled OneDrive, SharePoint and Teams, each with an arrow into one large box labelled one store.* They look like three places. They are one. ## The door decides who else gets in. *Picture: The same three doors, each with a small grey label underneath: just you, a team, the team.* That is the only real difference. ## Teams keeps nothing. *Picture: The Teams door drawn in amber, with a dashed amber arrow running up from its arrow to the SharePoint one.* A file you put in a channel is in that team’s SharePoint. ## Do today Open a file from a Teams channel, then find the same file in SharePoint. --- Read it as a page: [howtobedigital.com/explain/where-your-files-live/](/explain/where-your-files-live/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## A chat is not an email Teams and Outlook: what each is for, what is kept, who can find it later. 7 screens, about three minutes. Pictures, almost no words. ## They feel the same. *Picture: Two boxes side by side, one labelled a chat and one labelled an email.* You type, you send, it arrives. ## One is a corridor. One is a letter. *Picture: Two panels. On the left two people close together with a short line. On the right two people with a document travelling between them.* A corridor is for now. A letter is for the record. ## Both are kept. *Picture: Three steps in a row: written, kept, and findable, with the last in green.* Chat feels like talking. It is not. ## Somebody can go looking. *Picture: A dashed frame labelled the record holding a green box labelled every message, with a magnifying box beside it labelled anyone who is allowed.* A tribunal, an audit, a freedom of information request. ## Pick on how long it matters. *Picture: Two boxes labelled needs an answer today and needs to be found in a year, the second in green.* Not on which window is already open. ## Do today Find the oldest message in your Teams chat and note the date. --- Read it as a page: [howtobedigital.com/explain/a-chat-is-not-an-email/](/explain/a-chat-is-not-an-email/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## What your organisation can see What an admin can and cannot look at, said plainly, without frightening anybody. 7 screens, about three minutes. Pictures, almost no words. ## It is not your account. *Picture: One large box labelled your work account.* Your employer owns it and pays for it. ## Two machines, two rules. *Picture: A house shape labelled home and a larger box labelled work, with a laptop in each.* One is yours. One is somebody else’s, lent to you. ## There are two lists. *Picture: Two columns, one headed can look and one headed cannot without a reason, with three rows in each.* Most people assume everything is on the first one. ## Looking leaves a mark. *Picture: A box labelled a request with an arrow through a box labelled a reason into a box labelled your files.* It is recorded, and somebody has to justify it. ## Ask to read it. *Picture: A dashed frame labelled written down somewhere holding one green box labelled the policy.* You are allowed. It is usually duller than you fear. ## Do today Search your intranet for the words acceptable use and see what comes back. --- Read it as a page: [howtobedigital.com/explain/what-your-organisation-can-see/](/explain/what-your-organisation-can-see/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## The Microsoft AI stack One picture: the model underneath, Copilot on top, agents beside it, your data behind it. 7 screens, about three minutes. Pictures, almost no words. ## One word doing too much work. *Picture: One large box with the letters A I inside it.* At work it usually means three separate things. ## Three floors. *Picture: Three stacked boxes, from the bottom: the model, Copilot, agents.* Each one sits on the one below it. ## The model has never seen your work. *Picture: The same three boxes with the bottom one, the model, drawn in green.* It read a great deal, none of it yours. ## Copilot is the model plus a door. *Picture: The same boxes with Copilot in green, a tall dashed room to the right labelled your files, with arrows into it from the two upper boxes.* The door leads to what your organisation already holds. ## An agent is Copilot with one job. *Picture: The same picture with the top box, agents, in green and an amber padlock inside the room of files.* Fewer keys, one task, one accountable person. ## Do today Open Copilot at work and ask it which of your files it can see. --- Read it as a page: [howtobedigital.com/explain/the-microsoft-ai-stack/](/explain/the-microsoft-ai-stack/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## Which Copilot have you got? The free one, the paid one, and how to tell which is in front of you. 7 screens, about three minutes. Pictures, almost no words. ## Two doors, one sign. *Picture: Two identical boxes side by side, both labelled Copilot.* Same name, different jobs. ## The included one reads the web. *Picture: A box labelled Copilot Chat with an arrow to a box labelled the web, and a crossed-out line to a box labelled your work.* It comes with Microsoft 365, and stops there. ## The paid one reads both. *Picture: A box labelled Microsoft Copilot with arrows to both the web and your work.* Your files, email, meetings and chats. ## The paid one is in the apps. *Picture: Three boxes in a row labelled inside Word, inside Outlook, inside Teams, all in green.* That is the easiest way to tell which you have. ## Ask it something only work would know. *Picture: A dashed frame labelled how to check holding one amber box labelled ask it about your own file.* If it cannot, you have the included one. ## Do today Ask your Copilot to summarise a document you saved last week, and see whether it can. --- Read it as a page: [howtobedigital.com/explain/which-copilot-have-you-got/](/explain/which-copilot-have-you-got/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## What grounded means Why the same question gives a different answer at work than at home. 7 screens, about three minutes. Pictures, almost no words. ## One question, two answers. *Picture: One large box labelled the same question.* Asked at home and at work, you get different things. ## One of them has read your files. *Picture: Two people, one labelled a stranger and one labelled somebody who has read your filing cabinet.* That is the whole difference. ## It looks in the cabinet first. *Picture: A box labelled your question with an arrow into a dashed frame labelled what you are allowed to open, and an arrow out to an answer.* Only in the drawers your own key opens. ## Two colleagues, two answers. *Picture: Two people side by side asking the same question and getting two different answers.* Not a fault. It is your permissions, working. ## Always look at the source. *Picture: A box labelled the answer with an amber box beside it labelled where it came from.* A grounded answer shows you where it got it. ## Do today Ask your work Copilot a question and click through to one of the files it cites. --- Read it as a page: [howtobedigital.com/explain/what-grounded-means/](/explain/what-grounded-means/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## Copilot Chat What it is, where it is, what it can see, what it cannot. 7 screens, about three minutes. Pictures, almost no words. ## You already have this one. *Picture: One large box labelled Copilot Chat.* It arrives with Microsoft 365. ## A very well-read stranger. *Picture: A person labelled it, drawn beside a very large box labelled the web.* It has read a great deal, none of it yours. ## It cannot see your files. *Picture: A dashed frame labelled your work holding three documents, with a dashed line to the box labelled it, marked cannot.* Not your email, meetings or chats. ## Unless you hand it one. *Picture: A box labelled a file with an amber arrow into the box labelled it.* Upload a file and it reads that file, once. ## Signed in, it is protected. *Picture: A dashed frame labelled protected holding a green box labelled what you type.* Unlike the free one you use at home. ## Do today Open Copilot Chat, ask it to explain a word you met at work this week, and see how far it gets. --- Read it as a page: [howtobedigital.com/explain/copilot-chat/](/explain/copilot-chat/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## Asking for something Task, audience, length, example. Four parts of a request, drawn. 7 screens, about three minutes. Pictures, almost no words. ## You asked. It gave you something else. *Picture: One box labelled what you typed with an arrow to a very large box labelled what came back.* Almost always, the request was the problem. ## Four parts. *Picture: A cut-away box labelled a good request holding four boxes: the task, who for, how long, an example.* Most people send the first one on its own. ## Longer is faster. *Picture: Two panels. On the left one small box labelled write something. On the right four boxes stacked, in green.* Thirty extra words saves three goes at it. ## The first answer is a draft. *Picture: Four steps in a row with arrows: ask, read, say what is wrong, ask again. The third is amber.* Tell it what is wrong. That is the whole trick. ## An example beats an adjective. *Picture: A box labelled an example of what good looks like, in amber, with an arrow into a green box labelled much closer.* Paste something you liked instead of saying make it friendly. ## Do today Take a request you sent this week and add the three parts you left out. --- Read it as a page: [howtobedigital.com/explain/asking-for-something/](/explain/asking-for-something/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## Copilot inside the apps The same assistant in three places, doing three different jobs. 7 screens, about three minutes. Pictures, almost no words. ## One assistant, three desks. *Picture: One box labelled Copilot with three arrows out to boxes labelled Word, Outlook and Teams.* The same thing, sitting in different rooms. ## In Word it writes a first go. *Picture: A page with ruled lines and an amber box beside it labelled draft this.* A draft to argue with, not an answer. ## In Outlook it sorts. *Picture: A dashed frame labelled Outlook holding eight small boxes, with a green box beside it labelled the three that matter.* It reads the pile and says which need you. ## In Teams it catches you up. *Picture: A box labelled a meeting you missed with an arrow to a green box labelled what was decided.* What was decided, and what you were given. ## None of it sends anything. *Picture: Three boxes in a row labelled it drafts, it sorts, it summarises, with a slate box beneath labelled you still send it.* A person presses send. That is on purpose. ## Do today Open a long email thread and ask Copilot what you have been asked to do. --- Read it as a page: [howtobedigital.com/explain/copilot-inside-the-apps/](/explain/copilot-inside-the-apps/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## What Copilot can see It sees what you can already see, and no more. Permissions in one picture. 7 screens, about three minutes. Pictures, almost no words. ## It borrows your keys. *Picture: One large box labelled your keys.* Not a special set. Yours. ## The building is bigger than your keys. *Picture: A dashed frame labelled the building holding four rooms, two of them green and two plain.* Most of it was never yours to open. ## The rule is that simple. *Picture: Two rows of doors. The top row unlocked and green, the bottom row with amber padlocks.* If a door is shut to you, it is shut to it. ## The risk is the other way. *Picture: A box labelled a file shared with everyone, in amber, inside the building frame.* A file left open to everyone was always findable. Now somebody finds it. ## Tidy first. *Picture: A green box labelled tidy the sharing with an arrow into a box labelled before you switch it on.* Every organisation learns this in the same order. ## Do today Open one folder at work and check who it is actually shared with. --- Read it as a page: [howtobedigital.com/explain/what-copilot-can-see/](/explain/what-copilot-can-see/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## Checking what it gives you Names, dates, figures. Three checks before anything leaves your hands. 7 screens, about three minutes. Pictures, almost no words. ## That is the problem. *Picture: One large box labelled it looks right.* Wrong answers do not look wrong. ## Three things go wrong most. *Picture: A cut-away box labelled the answer with three boxes inside: names, dates, figures, the last in green.* Check those three and you catch most of it. ## Every figure needs a source. *Picture: A box labelled a figure with an arrow to a box labelled where it came from, in amber.* If it cannot show you one, do not send it. ## Cut what you cannot check. *Picture: Two panels. On the left a green box labelled you can check. On the right a slate box labelled you cannot.* A sentence you cannot stand behind is not yours. ## Your name goes on it. *Picture: A box labelled your name at the bottom of a document.* Not the tool’s. That is the whole reason to check. ## Do today Take one thing an assistant wrote for you and check every number in it. --- Read it as a page: [howtobedigital.com/explain/checking-what-it-gives-you/](/explain/checking-what-it-gives-you/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## What a ready-made agent is Somebody else already made it, gave it a job, and pointed it at some knowledge. 7 screens, about three minutes. Pictures, almost no words. ## A word doing too much work. *Picture: One large box labelled an agent.* In Microsoft 365 it has a narrow meaning. ## Three things, always. *Picture: A cut-away box labelled an agent holding three boxes: a job, some knowledge, some limits.* Take one away and it is just a chatbot. ## Like a temp who was briefed. *Picture: Two panels. On the left a person labelled a new starter. On the right a person labelled a temp who knows the filing.* Same person, told what the job is first. ## Somebody built it. *Picture: Four boxes in a row labelled built by Microsoft, built by your organisation, built by you, bought in.* Ready-made ones come from Microsoft. ## It is not accountable. You are. *Picture: A box labelled the agent with an arrow to a box labelled a result and an amber box beneath labelled a person is still accountable.* Nothing about an agent moves that. ## Do today Open Copilot at work and look at what is listed under agents or tools. --- Read it as a page: [howtobedigital.com/explain/what-a-ready-made-agent-is/](/explain/what-a-ready-made-agent-is/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## Where the ready-made ones are Where to find the agents Microsoft ships, and how to tell one from a chatbot. Re-check names before writing. 7 screens, about three minutes. Pictures, almost no words. ## They sit inside Copilot. *Picture: A dashed frame labelled Copilot holding a smaller box labelled a list of agents.* Not separate apps. A list you pick from. ## Two do the heavy work. *Picture: Two boxes side by side labelled Researcher and Analyst, both green.* One reads widely and writes up. One does numbers. ## Slower on purpose. *Picture: Two panels. On the left a short arrow labelled quick. On the right a long arrow labelled slow, in green.* It spends longer looking so the answer holds up. ## One sits in meetings. *Picture: A dashed frame labelled Teams holding a box labelled a meeting and a green box labelled notes and actions.* It keeps the notes and the actions while you talk. ## The names keep changing. *Picture: A box labelled the names with an amber arrow to a box labelled they change.* Learn what each does. The label will be different next year. ## Do today Open Copilot and try one ready-made agent on a question you already know the answer to. --- Read it as a page: [howtobedigital.com/explain/where-the-ready-made-agents-are/](/explain/where-the-ready-made-agents-are/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## Asking a site a question Copilot in SharePoint: point at a site, ask, get an answer with a link back. 7 screens, about three minutes. Pictures, almost no words. ## A site is a cupboard of files. *Picture: A dashed frame labelled a site holding six documents.* Somebody put them there over several years. ## You can ask the cupboard. *Picture: A box labelled your question with an arrow into the site frame.* Not the whole building. This one. ## It shows its working. *Picture: A green box labelled the answer with an amber box beneath labelled the file it came from.* Every answer names the file underneath it. ## It replaces opening nine files. *Picture: Two panels. On the left a person opening many documents. On the right a person and one green answer.* That is the whole benefit. It is a real one. ## It is only as good as the cupboard. *Picture: A box labelled a tidy site with an arrow to a green box labelled a good answer, and a box labelled a messy site with an arrow to a slate box labelled a poor one.* Nothing here tidies the files for you. ## Do today Open a SharePoint site you use and ask it one question you would normally ask a colleague. --- Read it as a page: [howtobedigital.com/explain/asking-a-site-a-question/](/explain/asking-a-site-a-question/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## Why two people get different answers Permissions decide the answer. The same question, two people, two results. 7 screens, about three minutes. Pictures, almost no words. ## Two people. One question. *Picture: Two people side by side, both with a box labelled the same question.* They get different answers and both are correct. ## Different keys. *Picture: Two sets of keys, one with three keys and one with one key.* You have been in longer, or you are on more teams. ## Some files are not for everybody. *Picture: A dashed frame labelled the site holding four documents, two of them green.* Decided years ago by somebody who has left. ## Fewer keys, thinner answer. *Picture: Two answers side by side, one long and green, one short and slate.* Nothing is hidden from them. It never was theirs. ## Check before you forward. *Picture: An amber box labelled ask who can see this before an arrow into a green box labelled then share the answer.* An answer you can see may quote a file they cannot. ## Do today Ask a colleague to run your question and compare what the two of you get back. --- Read it as a page: [howtobedigital.com/explain/why-two-people-get-different-answers/](/explain/why-two-people-get-different-answers/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## Making one yourself Agent Builder: a name, a job description, some knowledge, a test. 7 screens, about three minutes. Pictures, almost no words. ## You can make one. *Picture: One large box labelled an agent you made.* It takes about twenty minutes and no code. ## You are writing a job advert. *Picture: A page with ruled lines labelled a job advert.* What it is called, what it does, how it sounds. ## Four steps, in that order. *Picture: Four steps in a row: name it, describe the job, give it knowledge, test it. The last is amber.* Most people skip the last one. ## You point it at things. *Picture: A box labelled the agent with arrows from three boxes labelled a SharePoint site, a Teams channel, a public website.* It answers from those and says where it looked. ## Then you decide who gets it. *Picture: A dashed frame labelled who gets it holding two boxes, just me and my team, the second in amber.* Start with just you. Share it when it stops embarrassing you. ## Do today Open Copilot on a computer, find agent builder, and read the first screen without making anything. --- Read it as a page: [howtobedigital.com/explain/making-an-agent-yourself/](/explain/making-an-agent-yourself/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## Knowledge, and edges What you point it at, and telling it what to refuse. 7 screens, about three minutes. Pictures, almost no words. ## Hand it one drawer. *Picture: A dashed frame labelled everything holding many boxes, with one green box labelled one drawer.* Not the building. One drawer you know is tidy. ## Narrow beats broad. *Picture: Two panels. On the left a wide box labelled anything. On the right a narrow green box labelled this only.* An agent that answers everything answers nothing well. ## Say what it must refuse. *Picture: A cut-away box labelled what you write holding three boxes: what to do, what to refuse, what to say instead.* People forget this half and it is the half that saves you. ## Give it a way out. *Picture: A box labelled a question it cannot answer with an arrow to a green box labelled ask a person.* Tell it who to send people to when it does not know. ## Try to break it before anybody else does. *Picture: Three steps in a row: try to break it, fix the wording, try again. The first is amber.* Ask it the question you are dreading. ## Do today Write, on paper, the one question your agent should always refuse to answer. --- Read it as a page: [howtobedigital.com/explain/knowledge-and-edges/](/explain/knowledge-and-edges/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## A SharePoint agent is a file It sits in Site Assets as a .agent file, and file permissions decide everything. 7 screens, about three minutes. Pictures, almost no words. ## It is a file. *Picture: One page with ruled lines, labelled point agent.* That is the surprising part. ## It lives in a folder. *Picture: A dashed frame labelled the site holding a folder labelled Site Assets, holding one green file.* Same folder as the pictures and the logos. ## File permissions decide. *Picture: A key beside the file, labelled whoever can open the file.* No new setting, no new list to manage. ## Nothing tells anybody it exists. *Picture: A box labelled you make it with an arrow to a box labelled it sits there, and a slate note saying it does not appear anywhere.* You have to send people to it. ## You can go looking for them. *Picture: A search box labelled dot agent with three green files beside it.* Search a site for the words dot agent and see what turns up. ## Do today Search one SharePoint site you use for the file ending .agent and see whether anybody has made one. --- Read it as a page: [howtobedigital.com/explain/a-sharepoint-agent-is-a-file/](/explain/a-sharepoint-agent-is-a-file/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- ## Who can use the one you made Sharing an agent is sharing a file, and the answers still follow the reader's own permissions. 7 screens, about three minutes. Pictures, almost no words. ## You made it. Now what? *Picture: One file in green with two people beside it.* Sharing it is the same as sharing any file. ## Send them the file. *Picture: A box labelled share the file with an arrow to a green box labelled they can use it.* There is nothing else to switch on. ## They still get their own answer. *Picture: Two people asking the same agent and getting two different answers, one longer than the other.* The agent reads with their keys, not yours. ## Two different things. *Picture: An amber box labelled sharing the agent with a slate box beside it labelled is not sharing the files.* The agent moves. The files do not. ## A site can have a main one. *Picture: A dashed frame labelled the site with a green box labelled one agent for the site.* The owner picks it, so nobody chooses between nine. ## Do today Share one agent you made with a single colleague and ask them what answer they got. --- Read it as a page: [howtobedigital.com/explain/who-can-use-the-one-you-made/](/explain/who-can-use-the-one-you-made/) More explainers: [/explain/](/explain/). Lessons: [/learn/](/learn/). --- # Digital Learning Practice Training day is not the same as still being able to do it. Digital Learning Practice turns a one-off session into short, spaced practice that looks like the real decision at work. We measure what people can still do weeks later — not who turned up, and not who clicked Complete. Employers pay for this. That funds the free lessons. Nothing on the free side is paywalled, delayed, or made worse to sell a programme. One behaviour. Practice over weeks. A delayed check. A readout you can defend. Existing teaching offers stay: **One day, one team** (£1,800) and **AI at work, six weeks** (£3,600). Digital Learning Practice sits beside and above those. New DLP prices are hypotheses — see the pricing page. ## Pages in this section - [Overview](/digital-learning-practice/) - [Offers](/digital-learning-practice/offers/) - [How it works](/digital-learning-practice/how-it-works/) - [What good looks like](/digital-learning-practice/what-good-looks-like/) - [Evidence](/digital-learning-practice/evidence/) - [Sample journey](/digital-learning-practice/sample-journey/) - [Pricing hypotheses](/digital-learning-practice/pricing/) - [FAQ](/digital-learning-practice/faq/) --- # Spaced Learning Practice B2B channel section: white-label retention and transfer layer for consultancies, academies, LMS/LXP implementers and certification bodies. Category: Spaced Performance as a Service. Promise: the right practice, for the right person, at the right time, with evidence the capability survived and transferred — not more microlearning. Sits beside organisational teaching programmes and free lessons. Company and product brand names are stripped from public copy; substance kept. Standards bodies and public research (NIST, UNESCO, WEF, EEF, Cepeda, 1EdTech LTI, xAPI) remain attributable. Prices are planning hypotheses in £. ## Pages - [/spaced-learning-practice/](/spaced-learning-practice/) — overview; Prove / Scale / Operate - [/spaced-learning-practice/offers/](/spaced-learning-practice/offers/) — six services and packages - [/spaced-learning-practice/partners/](/spaced-learning-practice/partners/) — white-label channel pitch; 12-week pilot - [/spaced-learning-practice/how-it-works/](/spaced-learning-practice/how-it-works/) — spacing vs retrieval; headless engine - [/spaced-learning-practice/evidence/](/spaced-learning-practice/evidence/) — d=0.54; EEF SMART Spaces; microsurgery RCT - [/spaced-learning-practice/measurement/](/spaced-learning-practice/measurement/) — evaluation stack; go/no-go - [/spaced-learning-practice/build/](/spaced-learning-practice/build/) — 180-day stage gate; automation map - [/spaced-learning-practice/faq/](/spaced-learning-practice/faq/) — LMS, microlearning, surveillance, channel, guarantee