What is AI literacy? Definition, levels and skills
AI literacy: the EU AI Act definition, three practical levels for the workplace, the skills to train and a sample programme for every employee.
AI literacy is the ability to use artificial intelligence tools with your eyes open: knowing what they do well, where they go wrong, what you can hand them and how to check what they produce. It is not a technical skill reserved for data scientists. It is a set of habits for anyone who uses AI at work.
How is AI literacy defined?
The EU AI Act gives a legal definition in Article 3, point 56: the "skills, knowledge and understanding that allow providers, deployers and affected persons […] to make an informed deployment of AI systems, as well as to gain awareness about the opportunities and risks of AI and possible harm it can cause".
Three ideas stand out:
- skills, knowledge and understanding: knowing how, knowing what, and knowing why;
- informed: the goal is not to use AI more or less, but to make sound decisions about it;
- opportunities and risks: AI literacy includes what AI makes possible, not only what to fear.
Since 2 February 2025, Article 4 of the same regulation has required organisations that use AI systems to support the development of this literacy among their staff (see Article 4 of the AI Act).
What levels should an organisation aim for?
Not everyone needs the same depth. A simple three-level grid is enough to organise a programme:
| Level | Who | What the person can do |
|---|---|---|
| 1. Aware user | every employee | recognise a task suited to AI, never enter personal or confidential data, review before sharing, raise a concern |
| 2. Regular user | people who use AI every week | write a structured prompt, check an answer against a source, spot a hallucination or bias, choose a script when the task is mechanical |
| 3. Champion | managers, AI champions, high-risk teams (HR, legal, finance, health) | assess a new use case, explain internal rules, decide on human oversight, escalate incidents |
This grid is not an official framework. It is a practical way to size the effort, in the spirit of Article 4 (knowledge, experience, context of use).
Which skills should you train?
The following skills appear in most programmes. Each one matches a skill measured by a Maketools game, so progress shows up in the reports.
Knowing when to use AI
- Knowing when AI fits: language work, rewording, summarising a non-sensitive text.
- Choosing a script over a model: a calculation, a sort or a format conversion is better, and free, done by a formula or a program.
- Keeping a human accountable: decisions with consequences (hiring, legal, medical, financial) are not delegated to a model.
- Protecting personal data: no names, no health data, no identifiable customer record in an unapproved tool.
- Reviewing before publishing: nothing goes to a customer or the public without review.
Related game: AI or Not AI?, which puts dozens of realistic requests in front of the player and asks them to apply these five rules.
Knowing how to ask
- Giving context: role, audience, goal.
- Setting constraints: length, tone, output format, what must not be made up.
- Using examples: a sample of the expected answer is often worth more than a long instruction.
Related game: Prompt Quest, where players build a prompt from blocks and watch the answer change with its quality.
Knowing how to check
- Spotting hallucinations: an invented fact, a source that does not exist, a made-up figure.
- Detecting bias: a stereotype, a sweeping generalisation, a single viewpoint presented as neutral.
- Knowing when to verify: a plausible answer you cannot check is not a reliable answer.
Related game: Hallucination Hunter, where players sort an assistant's answers into reliable, hallucinated, biased or to be checked.
What does a sample programme look like?
An effective programme is short, repeated and close to people's daily work. For example:
- Kick-off (30 minutes): why the organisation is doing this, which tools are approved, the one-page internal policy.
- Core practice (one 5 to 10 minute game a week for a month): one game per skill family, in the order use, ask, check.
- Role-specific depth: high-risk teams replay a game whose content has been adapted to their own cases (HR, legal, customer service).
- Follow-up: skill-level reports, reminders for people who have not played, content updates when tools change.
The 30-day AI literacy plan sets this out week by week.
AI literacy and data literacy
AI literacy rests on data literacy: a model learns from data, and dirty data produces wrong results delivered with the same confidence as right ones. A complete programme covers both.
Get started
Play a game of AI or Not AI? from the catalogue, no account needed: it is the fastest way to check your own habits. Terms used here are defined in the glossary.