This is a 20-question AI literacy test scored against the four domains of the OECD/EC AILit framework — Engage, Create, Manage and Shape. It takes about eight minutes. Nothing is submitted anywhere and no sign-up is required; scoring happens in your browser.
The questions test judgement rather than vocabulary. Most people score well on Engage and Create and noticeably lower on Manage — which is the gap that matters, because Manage is where the professional risk sits.
All 20 questions and answers
Take the test first if you want a real score. This section is here so the questions are readable without JavaScript, and so teachers can use them offline. Each answer includes the reasoning, which is the part worth discussing in a classroom.
Domain 1 — Engage with AI
1. An AI gives you a journal citation that turns out not to exist. Why does this happen?
It predicts plausible text rather than retrieving real records. The model was never looking anything up. It generates what a citation for that topic would most plausibly look like — right author style, believable journal, convincing page number. Usually plausible and true overlap. When they diverge you get a confident invention.
2. Which best describes what a large language model actually does?
Predicts the most plausible next piece of text from learned patterns. Not a search engine, not a database lookup, not a rule engine. This single idea explains most AI behaviour that otherwise seems mysterious.
3. A video platform recommends something you end up liking. The most likely reason is:
Patterns from people whose behaviour resembles yours. This is collaborative filtering, and it is the AI most students meet daily without recognising it as AI.
4. “Training data” means:
The examples the model learned its patterns from. Its biases, gaps and errors become the model’s biases, gaps and errors — which is why bias is a data question before it is an ethics question.
5. Which of these is NOT machine learning?
A calculator applying fixed arithmetic rules. Rules written by a programmer are not learning. The other three all derive their behaviour from data.
Domain 2 — Create with AI
6. Your AI output was vague and unhelpful. The most reliable fix is to:
Add audience, constraints, format and one example of what good looks like. Vague brief, vague output. Politeness changes nothing; specificity changes everything.
7. Before asking AI to draft a report, the best first step is to:
Write a brief stating role, audience, constraints and success criteria. This is the same skill that makes someone good at delegating to a junior colleague.
8. An AI draft is about 80% of what you wanted. The best move is to:
Give specific corrections and iterate, keeping authorship of the result. Restarting throws away the 80%. Accepting it makes you a publisher of something you did not write.
9. Which is the most under-used high-value way to apply AI to your own work?
Having it critique your draft and name the three weakest points. AI as a critic is consistently more useful than AI as a producer, and almost nobody does it.
10. Which task is an AI assistant most likely to do badly?
Precise arithmetic across a long list of figures. Language models predict text; long exact calculation is the classic failure mode.
Domain 3 — Manage AI
11. Which elements of an AI answer should you verify as a standing rule?
Numbers, names, dates, citations and anything you are accountable for. Verification triage — you cannot check everything, so know what always gets checked.
12. An AI answers with total confidence. What does that confidence tell you?
Nothing reliable about accuracy. Fluency and correctness are unrelated in AI output. The most dangerous answers are the most confident wrong ones.
13. A teacher wants AI help writing report-card comments. What must NOT be pasted in?
Students’ names attached to their marks or behaviour notes. That is disclosure of a minor’s personal data into a system the school does not control. The topic and a sample comment are fine.
14. You use AI for a question in a field you know nothing about. The core risk is:
You have no way of recognising a wrong answer. AI is safest where you could have caught the error yourself, and most dangerous where you could not.
15. Which is the strongest reason to deliberately NOT use AI on a task?
The struggle involved is the actual point of the task. A student using AI to write an essay has outsourced exactly the cognitive work the essay existed to produce.
Domain 4 — Shape AI
16. Bias in an AI system most often originates from:
Patterns and gaps in the training data. Usually history showing through rather than anyone’s intent. This is what CBSE teaches from Class 7 under data and fairness.
17. When should you disclose that AI was involved in work you produced?
When a reader’s judgement of the work would change if they knew. A teacher generating practice questions needs no disclaimer. A student submitting an AI-drafted assignment does.
18. A hiring model appears to favour one group of candidates. The best first response is to:
Examine the training data and compare outcomes across groups. Diagnose before you patch. Forcing equal rates without understanding the cause hides the problem rather than fixing it.
19. An organisation buys AI tools and sees no measurable return. The most common cause is:
The existing process was accelerated but never redesigned. In 2026 data, 17% of organisations reported no positive ROI at all — falling to 11% among those with mature literacy programmes.
20. Which represents the highest level of AI literacy?
Influencing how AI is used and governed by others. That is the Shape domain. Using tools quickly is the Engage domain — the entry level, not the summit.
How this test is scored
Five questions per domain, drawn from the four-domain structure of the AILit framework. Your overall score maps to a level, but the domain breakdown is the more useful output — it tells you which of the eight core skills to work on next.
| Score | Level | What it means |
|---|---|---|
| 0–9 | Emerging | You are still forming the mental model. Everything else depends on this. |
| 10–14 | Engaging | Comfortable using AI, thin on judgement. The commonest place to be in 2026. |
| 15–17 | Managing | You catch most errors and protect your data. Bias and redesign are the remaining gaps. |
| 18–20 | Shaping | Equipped to set AI policy for others rather than just follow it. |
An honest caveat: a 20-question test measures knowledge of good practice, not whether you apply it under time pressure. The genuinely diagnostic assessment is performance-based — hand someone an AI-generated document with three planted errors and count how many they catch. Use this as a starting map, not a certificate.
Frequently asked questions
Is this AI literacy test free?
Yes, and there is no sign-up. Scoring runs entirely in your browser; nothing is transmitted or stored.
Can I use this with my students or staff?
Yes, freely, with attribution to this page. It works well as a before-and-after measure around a training programme — take it at the start, then again after four to six weeks.
What is a good score?
Most people who use AI daily land between 10 and 14, typically strong on Engage and Create and weaker on Manage. Scoring 15 or above on Manage specifically is the marker that matters professionally.
Does this give a certificate?
No — deliberately. A certificate awarded on a multiple-choice quiz signals very little. For credentialled training with performance-based assessment, see AI training for teachers and staff.
Built by Piyush Wairale (IIT Madras), founder of PiyushAI Edtech. Related: complete AI literacy guide · 30-day roadmap · AI glossary
