AI Literacy: Definition, Skills and 6 Exercises to Try

In brief

AI literacy is the skills, knowledge and understanding to use AI knowingly and to judge what it produces. The EU AI Act defines it in Article 3(56), UNESCO's 2024 student framework lists 12 competencies, and the US Labor Department's February 2026 framework names five content areas. EU providers and deployers must still support it, as of November 2026.

What is AI literacy?

The EU's AI Act gives the only legal definition. Article 3(56) says "AI literacy" means "skills, knowledge and understanding that allow providers, deployers and affected persons" to make "an informed deployment of AI systems" and to "gain awareness about the opportunities and risks of AI and possible harm it can cause". Plainer: you know enough to use AI on purpose, and enough to see when it's wrong.

An open notebook with blank pages on a black plinth, a magnifying glass resting across it, with a small terracotta Claude starburst, a brushed silver OpenAI knot and a rainbow glass Gemini star on their own plinths behind it, in a charcoal studio under white rim light.

Several groups have each written their own version. The table lists who issued each one, when, for whom, and how it's built, as read from each issuer's own page on 21 November 2026. A cell that the page didn't state says "not stated".

FrameworkIssued by and whenAudienceBinding?Structure
EU AI Act, Article 3(56) and Article 4EU; Article 4 applied from 2 February 2025, replaced by Regulation (EU) 2026/1744 (the Digital Omnibus on AI), in force 27 July 2026Providers and deployers of AI systems, and their staffYesA duty to take measures; no guaranteed level for any person
UNESCO AI competency framework for studentsUNESCO, 8 August 2024School studentsNot stated; a global reference for curricula12 competencies, 4 dimensions, 3 levels: Understand, Apply, Create
UNESCO AI competency framework for teachersUNESCO, 8 August 2024TeachersNot stated; a global reference for national frameworks15 competencies, 5 dimensions, 3 levels: Acquire, Deepen, Create
Stanford Teaching Commons, Understanding AI LiteracyStanford; date not statedInstructors and studentsNo4 domains (functional, ethical, rhetorical, pedagogical) across novice, intermediate and advanced
US Department of Labor AI Literacy FrameworkEmployment and Training Administration, notice dated 13 February 2026Workers, employers, training providers, teachersNo, voluntary guidance5 content areas and 7 delivery principles
Digital Promise AI Literacy FrameworkDigital Promise, 2024Learners, educators and school leadersNot stated3 components: Understand, Evaluate, Use
AILit Framework for primary and secondary educationEuropean Commission and OECD; draft May 2025, final version announced June 2026Primary and secondary schoolsNot stated; the Omnibus recital names it as an example the Commission and Member States could take into account4 domains and 22 competences, per the Commission's page

No one definition has won. The Labor Department defines AI literacy as "a foundational set of competencies that enable individuals to use and evaluate AI technologies responsibly, with a primary focus on generative AI". Digital Promise says it is the knowledge and skills that let people "critically understand, evaluate, and use AI systems and tools". Stanford never gives a one line definition. It asks four questions instead, and the rest of this page uses them.

What skills does AI literacy include?

The frameworks overlap on a short list: know how it works, use it well, check it, and use it responsibly. They differ in how they split it up. Stanford asks four questions, quoted from its page:

  1. Functional: "How does AI work?"
  2. Ethical: "How do we navigate the ethical issues of AI?"
  3. Rhetorical: "How do we use natural and AI-generated language to achieve our goals?"
  4. Pedagogical: "How do we use AI to enhance teaching and learning?"

The Labor Department builds for the workplace instead. Its five areas are Understand AI Principles, Explore AI Uses, Direct AI Effectively, Evaluate AI Outputs and Use AI Responsibly. Under "understand", it lists pattern recognition and probabilistic outputs, so the same input can give different answers, and hallucinations and accuracy limits. Under "direct", it names contextual framing, prompting techniques and iterating on outputs. Under "evaluate", it names verifying factual accuracy and spotting gaps or logical errors. Under "responsibly", it names protecting sensitive information and following workplace policies.

The same document sets seven delivery principles for training: enable experiential learning, embed learning in context, build complementary human skills, address prerequisites to AI literacy, create pathways for continued learning, prepare enabling roles, and design for agility. The first one explains the exercises below, because it says AI literacy grows by hands on use, not by reading about it.

UNESCO adds a mindset. Its student framework's four dimensions are a human centred mindset, ethics of AI, AI techniques and applications, and AI system design. Its teacher framework swaps the last two for AI foundations and applications, AI pedagogy and AI for professional learning. For school, that means asking what the tool is for before learning how to use it.

Digital Promise keeps it to three: understand what AI can do and how it works, evaluate its benefits and costs with human judgment at the centre, and use it for distinct contexts and purposes. Stanford's ethical domain adds the issues to weigh, including academic integrity, reliability, bias, equity, accessibility, privacy and sustainability.

Checking outputs shows up in every list above, which is why most of the exercises below practice it.

What does the EU AI Act require for AI literacy?

It requires providers and deployers of AI systems to take measures that support AI literacy among their staff and the people using AI on their behalf. As first adopted, Article 4 said they had to ensure "a sufficient level of AI literacy" to their best extent. Regulation (EU) 2026/1744, the Digital Omnibus on AI, was adopted on 8 July 2026, published 24 July and in force 27 July 2026. It replaced Article 4 and dropped that wording.

The new text says providers and deployers "shall take measures to support the development of AI literacy" and adds that the obligation "does not require providers or deployers to guarantee any specific level of AI literacy of any individual". The Commission must publish practical examples of how to comply, and the AI Board must adopt recommendations that take European competence frameworks into account. Lewis Silkin's summary adds that in practice this means documenting training and awareness programmes, not certifying anyone, and that deployers of high risk systems still face the more specific duty in Article 26(2). Most pages on this topic still describe the old wording, so check the date before you quote one.

The scope is wide. The Act's definition covers "affected persons" as well as providers and deployers, though the Article 4 duty itself is aimed at the staff and others who run AI systems for an organisation. A shop using a chatbot to answer customers is a deployer. So is a school that uses an AI marking tool. Neither needs to be an AI company.

Note what the law leaves out. It doesn't name a course, a number of hours or a test. A short written plan, a dated session and a list of who attended is what a small firm can show.

Three real cases of AI literacy at work and in class

  1. A stressed student asking about a chatbot. Stanford's guide gives three scenarios, one of them a family member who is an undergraduate, stressed about an essay, asking if a chatbot would help. A teacher or parent can use the four domains as questions: do you know how the tool produces text (functional), who owns the output (ethical), what does the assignment ask you to say in your own voice (rhetorical), and what does the student learn by doing it themselves (pedagogical)?
  1. A colleague's literature review. Another Stanford scenario has a colleague using AI for one. The evaluation skill has a measured answer here: in a 2023 study, Walters and Wilder asked GPT-3.5 and GPT-4 for short literature reviews on 42 topics and checked the 636 citations. 55% of GPT-3.5's citations and 18% of GPT-4's were fabricated. Those were older models, but the check they imply hasn't changed: open every source.
  1. A firm that must show its work. Under Article 4, a deployer using a chatbot for customer replies can't just say staff know how to use it. A reasonable record is one session on prompts and limits, a rule about what never gets pasted in, and a sign in sheet. That is the "use responsibly" area of the Labor framework turned into evidence, and Article 4 asks for measures rather than proof of results.

What AI literacy exercises can you do today?

Each exercise trains one skill from the lists above, and each takes under 15 minutes in a free chatbot. Teachers can run them as a class activity. Managers can run them as a team session and keep the notes as a record.

  1. Two prompts, one task (direct). Write the task in one line, then again with your role, audience, format and length. Compare. A student sees why context matters, and a team sees why a template helps.
  1. Fabrication hunt (evaluate). Ask for five sources on a niche topic, then open all five. Count how many exist and say what they actually claim.
  1. Disagree on purpose (evaluate). Ask two different models the same factual question and mark every line where they differ. Check those lines first.
  1. The never paste list (use responsibly). List ten things you would never put in a chatbot, such as customer names, passwords or unpublished grades. Check them against your school or company policy.
  1. Teach it back (functional). Explain in four sentences why a chatbot can sound sure and be wrong. Then ask the model to mark where your explanation is off.
  1. Spot the missing voice (ethical). Ask for advice on a topic and note whose view isn't in the answer. Rewrite the prompt to ask for it.

You don't need a paid account for any of these. Prompt engineering for beginners covers the first one in depth, and how to fact check AI answers covers the second.

Questions to ask an AI about AI literacy

Use these to turn the model into a study partner. Paste them into Claude, ChatGPT, Gemini or Whizi, and fill the brackets.

  1. "Quiz me on AI literacy for a [teacher or office worker]. Ask five questions, one at a time, from these areas: how AI works, how to prompt it, how to check it, and how to use it safely. Wait for my answer each time."
  1. "Here is my AI use policy: [paste]. Which of these five areas does it skip: understanding AI, exploring uses, directing AI, evaluating outputs, using it responsibly? Quote the line that covers each one."
  1. "Write a 30 minute AI literacy session for [group, size] who use AI for [task]. Include one exercise on checking sources, one on what never gets pasted in, and a one page attendance record."
  1. "I think this claim about AI is true: [claim]. Argue against it, then tell me the one fact I should check first and where to find it."
  1. "Explain [concept, such as hallucination] to a [grade or role] in six sentences. Then give me three examples where it would cause a real problem at [workplace or school]."
  1. "Turn this assignment into one that tests AI literacy: [paste]. Students may use AI, but they must show what they asked, what was wrong and what they fixed."

Prompt 2 works best when you paste your real policy. If you haven't got one, prompt 3 will draft a start.

How do teachers teach AI literacy?

Teachers start with the four Stanford questions, then pick one skill per lesson. Stanford's framework scales each domain across novice, intermediate and advanced objectives, so you can pick the level that matches your class.

The Khan Academy lesson that ranks second for this query is a 3 minute and 18 second video in a Grade 9 course, with no formal definition on the page. It introduces the idea, and exercises one to six above give students something to do afterward. AI for teachers covers grading, lesson planning and policy in more depth.

Grade the process, not the polish. A simple rubric has four lines, one per Stanford domain. Did the student explain how the tool produced the text? Did they say who owns it and what they shouldn't share? Is the final voice theirs? Did they learn something by doing the step themselves? Each line gets a yes, partly or no, and a one sentence reason. It's original to this page, so adapt it to your course. Stanford itself gives no answer key for its scenarios.

Keep the human centred line from UNESCO. A student who can say what the tool is for, and when not to use it, is more literate than one who knows ten prompt tricks.

What does AI literacy for students look like?

For students it's a progression from understanding AI, to applying it, to creating with it. UNESCO's student framework spans exactly those three levels across 12 competencies, and its page stresses critical judgement of AI solutions and awareness of citizenship responsibilities in the AI era.

The European Commission and OECD's AILit Framework, final version announced in June 2026, is written for primary and secondary schools. Its own site lists four domains: Engage with AI, Create with AI, Manage AI and Shape AI. It adds learner expectations at basic, intermediate and advanced levels and low tech classroom activities that teach AI ideas without direct access to AI tools. The Commission's page counts 22 competences.

The Khan Academy video that ranks second for this query covers the same ground at an introductory level: what AI is, how it works, its limits and the biases it can carry. Exercises 1, 2 and 5 above turn those ideas into something a student does.

How does a workplace run a 30 minute AI literacy session?

A workplace session needs an agenda, a task people actually do, and a record. This one maps to the Labor framework's five areas and fits a team of any size. It's an original outline for this page, not a template from any framework.

  1. Five minutes, understand: show one wrong answer from a chatbot on a question the team knows, and ask why it sounded sure.
  2. Five minutes, explore: each person names one task from their week where AI could help and one where it shouldn't.
  3. Ten minutes, direct and evaluate: pairs write a prompt for their task, run it, and check three claims against a source.
  4. Five minutes, use responsibly: agree the never paste list and who to ask when unsure.
  5. Five minutes, record: write the date, the topics and the names present.

For a class, swap step 2 for exercise 6 and replace the record with a short reflection on what each student fixed. Keep it small. A session people finish beats a workshop they skip.

What mistakes ruin AI literacy training?

Training goes wrong in four familiar ways, and each has a fix.

  1. Teaching tools, not judgment. The fix: spend half the session on checking outputs, as the Labor framework's fourth area asks.
  2. One course, then nothing. The fix: repeat a short exercise each month, because the tools change faster than a slide deck.
  3. Treating a US or UNESCO framework as the law. The fix: in the EU, the legal text is Article 4 as amended; the others are voluntary.
  4. Skipping the record. The fix: write down the date, the content and the attendees. That's what documenting training means in practice.

Where Whizi fits

Whizi is one app where you can run exercise 3 yourself. Pro carries Claude, GPT and Gemini, its first week is $1.99, and side by side Compare puts two models on one prompt so you can read where they disagree. Compare is Pro and Powerhouse.

The free plan gives you 10 messages a day on Whizi AI, Auto and Claude Haiku, with no card, and web search is on by default in each chat. That covers exercises 1, 2, 4, 5 and 6. Teachers and teams can also upload a PDF or Word policy and ask the quiz prompt about it.

What are the limits of AI literacy?

No single definition has won, and literacy isn't a score. Stanford's page has no formal definition and gives no answer key for its scenarios. Coursera's explainer lists a "4 C's" model of AI literacy (conscientious, collaborative, critical and creative) without naming where it comes from.

A framework also can't make anyone check a source. The Walters and Wilder study shows the habit matters: fewer fabricated citations from the newer model is progress, not safety. The Labor Department calls its own framework a starting point that will evolve with AI capabilities, so treat any framework as a starting list and test it on your own tasks.

Measuring literacy in one person is the open part. UNESCO and the Labor Department describe what to teach. Treat any single score you see online as that seller's own measure, not a standard.

Workflow checklist
  • Name the framework you use and the date you read it
  • Run one exercise per skill: direct, evaluate, use responsibly
  • Open the source behind every claim you plan to repeat
  • Write down who joined each session and what it covered
  • Repeat a short exercise every month as the tools change
Common questions

Frequently asked questions

What is AI literacy in simple words?

AI literacy is knowing enough about AI to use it on purpose and to spot when it's wrong. The EU AI Act's Article 3(56) words it as skills, knowledge and understanding for an informed deployment and an awareness of opportunities, risks and possible harm. In practice that means prompting clearly and checking what comes back.

Is AI literacy required by law?

In the EU, yes for providers and deployers of AI systems. Article 4 has applied since 2 February 2025 and was softened by the Digital Omnibus on AI, in force 27 July 2026, so it asks for measures that support literacy, not a guaranteed level. The US Labor Department's February 2026 framework is voluntary, as of November 2026.

What are the four domains of AI literacy?

Stanford's Teaching Commons names functional, ethical, rhetorical and pedagogical literacy. Each has a guiding question, such as "How does AI work?" for functional. Other frameworks split it differently, so the four domains are one popular model, not a universal standard.

How do you measure AI literacy?

There is no single accepted test. Stanford's guide offers a self assessment form for instructors, and the EU doesn't require certification of individuals. A practical measure is a dated exercise: can the person rewrite a prompt, check three claims against sources and name what they would never paste in?