In this article
The short answer
Measure AI literacy with four methods: a short knowledge check on the basics and your policy, scenario questions that test judgement, a work sample where people use AI on a real task and show how they checked it, and observed behaviour over time. Record the results for each person against your framework, and report a few simple metrics such as coverage, levels reached and incidents.
If someone asks "how do you know your staff are AI literate?", you want a better answer than "they all watched the video". Good measurement is also good training: the act of being asked a realistic question is often where the learning sticks.
Why measure AI literacy at all?
- To target training. A baseline shows where the real gaps are, so you do not pay to teach people what they already know.
- To evidence it. If you are in scope of the EU AI Act literacy duty, or a client or regulator asks, records of how understanding was checked are far stronger than attendance.
- To show value. Leaders fund what they can see working. Pair literacy measures with the business measures in how to measure the ROI of AI training.
What are the best ways to assess AI literacy?
| Method | What it tests | Effort | Strength of evidence |
|---|---|---|---|
| Knowledge check | Basics and your policy: approved tools, data rules, disclosure | 10 minutes, online | Low on its own |
| Scenario questions | Judgement in realistic situations | 15 minutes, online or in session | Medium |
| Work sample | Using AI on a real task and checking it | 30 to 45 minutes, reviewed by a person | High |
| Observed behaviour | Habits in normal work over time | Ongoing, through managers and champions | High |
For most businesses, a knowledge check plus scenarios for everyone, and a work sample for regular users and high-risk roles, is the right balance.
What do good AI literacy assessment questions look like?
Good questions are about decisions, not definitions. Three examples:
Scenario: the client spreadsheet
A colleague asks you to use AI to summarise a spreadsheet of client names, addresses and account balances by this afternoon. What do you do?
What a strong answer includes
It uses only the approved business tool, removes or pseudonymises identifiers if the policy requires it, checks the totals against the source and does not use a personal account.
Scenario: the confident statistic
AI gives you a striking statistic for a proposal, with a source name but no link. What do you do?
What a strong answer includes
It finds and reads the original source, and drops the figure if it cannot be verified.
Scenario: the shortlist
Your recruitment tool now offers to rank applicants automatically. What should happen before anyone relies on it?
What a strong answer includes
It escalates to the policy owner, asks about bias testing and keeps the decision with a person.
Create 10 multiple-choice scenario questions to assess AI literacy for [role] at a UK [type of business]. Each question should describe a realistic workplace situation involving an AI tool and test judgement, not definitions. Cover: data protection, checking accuracy, approved tools, bias, disclosure and when not to use AI. For each question give four options, mark the best answer, and explain in one sentence why it is best with reference to our policy below. Do not invent rules; if our policy does not cover a situation, flag it. POLICY: [paste policy text]
Check it: Have a manager and your policy owner review the answers before you use them. Scenario banks need checking like any other AI output.
What should an AI literacy training record include?
| Field | Example |
|---|---|
| Name and role | Priya Shah, Recruitment Consultant |
| AI tools used | Microsoft 365 Copilot, ATS screening feature |
| Target level | Confident on competencies 5 to 8 |
| Training received | Foundation (half day), High-risk roles (half day) |
| Date and provider | 14 October 2026, The Oxford AI School |
| How understanding was checked | Scenario check 9 out of 10; work sample reviewed by manager |
| Level reached | Confident |
| Next review | October 2027 or on tool change |
Keep it in your HR or learning system alongside other mandatory training. A spreadsheet is fine for a small business.
Which AI literacy metrics should we report?
Keep the board report to a handful of numbers:
- Coverage: the share of staff who have completed the foundation tier, and of each target group who have completed their tier.
- Level reached: the share at or above their target level, by team.
- Shadow AI: whether use of unapproved tools is falling, from surveys or IT data.
- Incidents and near misses involving AI, which should be reported more at first as awareness rises, then fall.
- Benefit: time saved or quality improved on two or three tracked workflows.
Measuring AI literacy: questions people ask
How do you assess AI literacy in employees?
Combine a short knowledge check on the basics and your policy, scenario questions that test judgement, a work sample where people use AI on a real task and explain how they checked it, and ongoing observation by managers.
Is there a standard AI literacy test?
There is no single standard workplace AI literacy test in the UK. The most useful assessments are built around your own tools, policy and roles.
What evidence of AI literacy training should we keep?
For each person: the AI tools they use, their target level, the training received with dates and provider, how understanding was checked, the level reached and the next review date.
How often should AI literacy be reassessed?
Reassess at least annually, and when someone changes role or your tools or AI policy change significantly.
Can AI literacy be measured with a survey?
Self-assessment surveys are useful for a quick baseline and for spotting shadow AI, but people often over- or under-rate themselves. Back them up with scenarios or work samples.
Keep reading
Training that ends with evidence, not just a certificate
Every Oxford AI School literacy programme includes a scenario check and a record of who reached which level, ready for your HR system.
- Basic AI Training Hands-on AI literacy for every employee.
- AI Policy Workshop Write the rules your training will teach.
- AI Capability Assessment Baseline your team before you train.
Harry Lang
Harry Lang founded The Oxford AI School in Witney, Oxfordshire. He trains UK businesses and teams to use AI well, writes the Marketing Unfiltered newsletter, and is the author of Learn To Fly With AI.