An AI literacy assessment should test decisions people make at work. Definitions and multiple-choice recall can support a baseline, but they do not show whether someone can use AI safely, review an output, or redesign a workflow.
Use the following scenario areas to create a practical assessment.
1. Tool and task choice
Present several tasks and ask which are appropriate for the approved tool. Include a low-risk draft, a sensitive-data task, a consequential recommendation, and a task where the output cannot be meaningfully checked.
Look for: fit-for-purpose judgment and awareness of account or environment differences.
2. Data handling
Show sample inputs with public, internal, confidential, personal, and restricted data. Ask what may be entered, what must be removed, and which environment is required.
Look for: application of the company's actual classification and policy.
3. Briefing and prompting
Give a vague request and ask the employee to improve it. A strong response should define the objective, context, audience, constraints, source material, output format, and quality standard.
Look for: clear thinking, not a memorized acronym.
4. Output review
Provide a plausible response containing a factual error, unsupported claim, missing caveat, or confidentiality issue. Ask the employee to identify the problems and explain how to verify or correct them.
Look for: healthy skepticism and an appropriate verification method.
5. Accountability and escalation
Describe a situation in which AI can technically perform an action, but policy or consequence requires approval. Ask who remains accountable and where the issue should go.
Look for: recognition that capability does not grant permission.
6. Workflow application
Ask the employee to map a recurring process and select one step for AI assistance. Require a baseline, input, output, review point, exception path, and measure.
Look for: practical workflow design rather than isolated prompting.
Manager extension
Managers should also respond to an employee who discloses unsanctioned use, choose a team workflow, create practice time, set review expectations, and interpret adoption evidence. Score whether the manager creates a safe, useful operating environment.
Scoring
Use four levels: unaware, foundational, applied, and able to coach. Score each capability separately. A person may be strong at prompting and weak at data handling. Set role-specific minimums and direct learning to the gaps.
Administration
Assess before training, shortly after practice, and again after workplace application. Combine scenario scores with observed behavior, manager feedback, and workflow results. Do not use the assessment as a hidden performance-management tool; explain its purpose and data handling.
The goal is not a perfect test score. It is dependable judgment when AI enters real work.