How to assess AI skills in employees
Most organisations know how many employees have access to AI tools. Some track active users, completed courses or the number of prompts submitted. Those figures describe adoption, but they do not show whether people can use AI well.
An employee can use ChatGPT every day and still miss a confident answer built on a false assumption. Someone else might use it less often but know how to frame the task, challenge the result and check the information that matters.
If you want to understand AI capability across your workforce, you need to assess what people can do with AI in the context of their work.
What is an AI skills assessment?
An AI skills assessment measures whether someone can use AI effectively, responsibly and with appropriate judgement.
A useful assessment looks beyond familiarity with tools or knowledge of AI terminology. It examines whether an employee can choose a suitable task, provide enough context, evaluate the response and verify the result before it affects real work.
This is increasingly important because AI use is spreading faster than formal training. The US Department of Labor's 2026 AI Literacy Framework advises employers to begin with the workflows where AI is already appearing, including report writing, data analysis and customer communication. It also treats AI literacy as a combination of technical understanding, critical thinking and responsible use, rather than a single tool skill.[1]
Research published in 2025 makes a similar case for task-oriented assessment. The authors found that practical, contextualised scenarios were more useful for measuring applied workplace AI literacy than tests focused mainly on abstract technical knowledge.[2]
Why AI usage is a weak measure of capability
Usage data is useful when you need to understand adoption. It can tell you who has opened a tool, how often it is used and which teams are experimenting.
It cannot tell you whether the work is reliable.
High usage may mean that someone has found valuable applications for AI. It may also mean that they are repeating weak prompts, correcting poor answers or using an expensive model for simple tasks. The number itself does not explain the quality of the interaction.
Course completion has the same limitation. It confirms that someone reached the end of a learning programme. It does not prove that they can apply the material to a forecast, customer response, risk review or management decision.
Self-assessment adds another partial view. It helps you understand confidence, attitudes and perceived development needs. People are not always good at judging their own proficiency, particularly when the standard for competent AI use is still taking shape.
Each of these signals can contribute to an assessment. None should carry it alone.
What should an employee AI skills assessment measure?
An effective assessment should cover the whole interaction between the employee, the task and the AI system. Five areas give you a practical starting point.
Task selection and judgement
The employee should understand where AI is likely to help and where its involvement creates unnecessary risk.
This includes recognising:
- tasks that can be accelerated safely;
- tasks that need close human review;
- information that should not be entered into a public tool;
- decisions that require accountable human judgement;
- situations where a simpler tool would be more suitable.
You can assess this with realistic choices rather than definitions. Present several tasks from the person's role and ask how AI should be used in each one.
Context and prompting
Good prompting depends on clear thinking about the work.
A capable employee can explain the objective, provide relevant background, define constraints and specify what a useful answer should contain. They should also be able to refine the instruction when the first result is weak.
The assessment should test whether the employee can:
- describe the task clearly;
- supply the necessary context;
- distinguish facts from assumptions;
- specify the desired format;
- state important limits;
- improve a prompt after reviewing the output.
This is more useful than asking whether they know a particular prompt template. Models and interfaces will change. The ability to structure a task will remain useful.
Output evaluation
AI systems can produce clear, persuasive work that contains mistakes. Employees need to recognise the difference between a polished answer and a dependable one.
A practical test can include an AI-generated report, analysis or recommendation with deliberately placed problems. Ask the employee to identify:
- unsupported claims;
- missing evidence;
- incorrect assumptions;
- inconsistent figures;
- invented or unsuitable sources;
- important information the response has ignored.
The errors should reflect the person's work. A finance employee might review KPI commentary that overstates the meaning of a variance. A marketer might inspect an audience analysis based on incomplete data. An operations manager might review a recommendation that overlooks a process constraint.
Verification
Recognising that an answer may be wrong is useful. The employee must also know how to check it.
Verification methods vary by task. They may include checking an original source, reconciling the result with company data, repeating a calculation, testing code or comparing the answer with an approved policy.
A good assessment looks at whether the employee can decide:
- which claims need verification;
- which source should be treated as authoritative;
- how much checking is proportionate to the risk;
- what evidence should be retained;
- whether the final result is safe to use.
This avoids two common problems: trusting everything because it sounds credible, or checking every minor detail until AI no longer saves any time.
Workflow application
Individual prompting skill creates limited value if it never becomes part of a reliable process.
Employees should be able to place AI within a workflow where inputs, review points and responsibility are clear. Another person should be able to understand what happened and repeat the process where appropriate.
For an advanced assessment, ask the employee to design or improve a small AI-assisted workflow. Evaluate the quality of the final work, how they used AI and how they handled verification.
Why the assessment should be role-specific
A single general test can measure basic AI literacy across a company. It becomes less useful as employees move into practical work.
A finance manager, HR business partner and marketing analyst may all need sound prompting and verification skills. The evidence of proficiency will look different in each role.
Role-specific assessment also makes the result easier to act on. It tells you what someone can do in their current work, rather than how well they performed on a generic AI quiz.
A practical AI proficiency model
An assessment needs clear levels so employees and managers can understand the result. Four levels are usually enough to show meaningful progress without creating false precision.
Level 1: Aware
The employee understands basic uses and risks but needs guidance to apply AI to work.
Level 2: Supported user
The employee can complete defined tasks using examples, approved tools and an established review process.
Level 3: Independent practitioner
The employee can choose suitable use cases, produce useful work and verify important outputs without continuous support.
Level 4: Workflow builder
The employee can design repeatable AI-assisted workflows, define controls and help colleagues develop their practice.
A person may sit at different levels across the assessment. Strong prompting does not guarantee strong verification. A single total score is helpful for comparison, but a capability profile is more useful for development.
How to run an AI skills assessment
Start with the work your teams are expected to improve. Identify the tasks where AI is already used or likely to be introduced, then define what competent performance looks like.
Use several forms of evidence:
- a short questionnaire to understand experience and confidence;
- realistic scenarios that test judgement;
- practical tasks based on the employee's role;
- review exercises containing plausible AI errors;
- completed work that shows how the employee applied and verified AI.
Keep the initial assessment manageable. A short baseline can identify broad gaps. More detailed task-based assessments can then be used for roles where AI has greater value or risk.
Tell employees what is being measured and why. The assessment should help them understand where they are and what to learn next. If it feels like hidden performance monitoring, the answers and behaviour you observe will be less useful.
Turn the result into a development plan
The final score is only a starting point.
Assessment data should help you decide:
- which foundational skills need attention;
- which employees are ready for independent use;
- where stronger verification controls are required;
- which role-specific projects would build useful experience;
- who may be able to support colleagues;
- where capability gaps exist across a team or department.
The learning that follows should use the same context as the assessment. If an employee struggled to verify AI-generated financial commentary, a generic prompt engineering course will not close the gap. They need guided practice using financial tasks, data and standards that resemble their work.
Repeated assessment then shows whether capability is improving. The evidence can include better prompts, fewer unsupported claims, higher-quality deliverables and stronger judgement about where AI should be used.
Measure capability through real work
AI skills will keep changing as tools improve. A useful assessment should therefore focus on abilities that transfer between systems: task judgement, context, evaluation, verification and workflow design.
Etiq builds these skills through personalised learning pathways and applied projects based on the work people already do. Built-in verification checks prompts and outputs, while skills matrices give leaders a view of capability across employees, teams and departments.
If you want to understand where your workforce stands, talk to Etiq about a role-based AI capability assessment.
