Learn · Guide
How to build an AI literacy training programme with AI
Published 31 October 2026
Short answer. Map who in the company uses which AI tool, and for which tasks. Decide what each group needs to know. Let an AI assistant draft short material for each group from your own rules and examples, and have a person check every page before it is used. Add short exercises on real tasks, run the training, and keep a short record of what was done and when. Repeat the parts that change when a tool or a rule changes.
AI literacy means the skills, knowledge and understanding that let people use AI with judgement. Why a company that uses AI must support it is explained in What does the EU AI Act ask of a company that uses AI? This guide is the how.
What you need before you start
- A list of the AI tools in use. Include the ones people use on their own, and any AI agents that work in your systems.
- The company’s AI rules. The written rules for what may and may not go into an AI tool. If you have none yet, step 3 shows how to write them.
- An AI assistant the company has approved for internal documents. You will use it to draft the material.
- A named person who checks every page of the material, and whom people can ask when in doubt.
Step 1: Map who uses which AI tool, for what
Ask people which AI tools they use, and for which tasks. Then sort them into groups by how they use AI, not by job title. Two people with different titles who both use a chat assistant to draft emails need the same training.
| Group | Tool | Tasks | What data the tool sees | Who checks the result |
|---|---|---|---|---|
| [for example: people who draft text with an AI assistant] | [tool name] | [emails, summaries, translations] | [internal documents, no personal data] | [the person themselves] |
| [for example: people who check an AI agent’s entries] | [agent] | [approve draft entries] | [the accounting system] | [the operator] |
The last column matters most: it shows who has to be able to spot a mistake.
Example prompt
Write a short questionnaire for the staff of a [size] company in [line of business]. It should find out which AI tools each person uses, for which tasks, what kind of data they put into them, and who checks the result. Use plain questions and no more than eight of them. Do not ask for names of customers or other personal data.
Step 2: Decide what each group needs to know
Everyone who uses an AI tool needs five things, fitted to the tool they use:
- What the tool can do. The tasks it is approved for in your company.
- Where it goes wrong. For example, a language model can present a wrong or invented answer with confidence, which is called a hallucination.
- What may go in. Which data may be put into the tool, and which may not.
- Who to ask. The named person, and what to do when something looks wrong.
- The company’s rules. The written AI rules, and where to find them.
Some groups need more:
| Group | What they need in addition |
|---|---|
| The operator, who checks and signs off an AI agent’s work | What the agent may do on its own and what it must ask about; how to check a result against its source; how to stop or correct the agent; what the logbook shows. The European Commission says the person in the loop needs skills targeted to the system they use. |
| The people who decide which AI tools the company uses | What the company’s duties are; how to approve a new tool; when the programme must be updated |
Step 3: Start from your own rules and examples
Training built on general examples teaches general lessons. Build it from your own material instead: the company’s AI rules, the tasks from the map in step 1, and a few real examples of good and bad results.
If the company has no written AI rules yet, write them first. An AI assistant can draft them, but the company decides them: a person goes through every rule, changes what does not fit, and approves the result.
Example prompt
Draft a one-page set of rules for using AI tools at a [size] company in [line of business]. Our approved tools are [list]. Cover: which tasks the tools may be used for, which data must never go in (for example personal data, customer data, passwords), how to check a result before using it, and whom to ask. Write in plain language, one rule per line. Mark every rule I need to decide on with [DECIDE].
Step 4: Let an AI assistant draft the material, and check every page
Give the approved AI assistant the rules from step 3 and the five points from step 2, and ask it for one short page per group. Use only material that may go into that tool under your own rules. The European Commission notes that, in many cases, relying only on a tool’s instructions for use, or asking staff to read them, might be ineffective.
Then a person reads every page before anyone else sees it. They check two things: that the page matches the company’s rules, and that it matches what the tool really does in your company. Anything the assistant got wrong is corrected by hand.
AI drafts, a person decides. Nothing in the programme is used until the named person has read it and approved it.
Example prompt
Here are our AI rules: [paste rules]. Here is what people in the group “[group name]” use AI for: [tasks from step 1]. Write a one-page guide for this group. Cover what the tool can do for them, where it can go wrong in their tasks, what they may and may not put in, whom to ask, and where to find the rules. Use our rules only; do not add rules of your own. List anything you were unsure of at the end.
Step 5: Make short exercises on real tasks
People learn where a tool goes wrong by seeing it happen in their own work. Make each exercise from a real task in the group’s map, with data that is not sensitive or that you make up.
- Check against the source. Ask the tool for a figure or a fact from a document, then find it in the document yourself.
- Find the mistake. Look at an answer that contains an invented figure or reference, and find it.
- Decide what may go in. Look at a few pieces of text and decide, with the rules, which may be put into the tool.
- For the operator: approve or reject. Go through a few prepared results from the agent, some right and some wrong, and approve or reject each one.
Example prompt
Make three short exercises for people who use an AI assistant to [task]. Use made-up data only. Exercise 1: an answer with one invented figure for them to find. Exercise 2: five short texts, where they decide with these rules [paste rules] which may go into the tool. Exercise 3: a question whose answer they must check against a short source text you provide. Give the answers separately.
Check the exercises as you checked the material: a person tries each one before it is used.
Step 6: Run it on people’s own tools
Run the training in short sessions, with people at their own tools. Go through the page for their group, then let everyone do the exercises. Questions that come up go to the named person, and the answers go back into the material.
Step 7: Keep a short record
Write down what was done, so the company can show its measures. No certificate is needed: the European Commission says a company can keep an internal record of its trainings.
| Date | Group | Who took part | Material and version | Tools covered | Checked by |
|---|---|---|---|---|---|
| [date] | [group] | [names or number of people] | [page title, version] | [tools] | [named person] |
Keep the material itself with the record, in the version that was used.
Step 8: Repeat when something changes
Update the material and run the changed part again when:
- the company takes a new AI tool or AI agent into use, or a tool changes what it can do
- the company’s AI rules change
- a new person starts using AI tools, or someone moves to a new group
- a mistake shows something people did not know
Add a line to the record each time.
Look for the Commission’s practical examples
The amended AI Act asks the European Commission to publish practical examples of how companies can meet the AI literacy duty. Check for them when you build or update your programme, and keep your own programme fitted to your own tools and tasks.