LunaPrompts
To assess AI fluency in a non-technical role, give the candidate a realistic task, let them use an approved AI assistant, and ask them to check and explain the result. Look for how they define the work, handle missing information, verify claims, and improve an imperfect draft.
Asking “Which AI tools do you use?” tells you what someone has tried. A small work sample gives you something more useful to discuss. For a recruiter, that could mean turning an incomplete hiring brief into a clear assessment invitation. For an operations colleague, it could mean checking an onboarding plan against an approved checklist.
Why this belongs in the 2026 hiring conversation
LinkedIn’s 2026 Future of Talent report describes talent acquisition expanding into skills visibility, development, and workforce planning. Its survey covered 771 HR and talent leaders at organizations with more than 1,000 employees, fielded in June 2026. That is evidence about a particular leadership population, not proof that every employer has adopted the same approach. LinkedIn’s report and methodology.
For HR, a practical implication is to describe the capability you need before adding “AI skills” to a job description. The exercise below is our suggested starting point. It is an illustrative work sample, not a validated employment test or a universal pass standard.
What should an AI-fluent colleague be able to show?
Four observable behaviors are useful across many office roles:
- Frame the task: explain the intended audience, source material, constraints, and usable outcome.
- Guide the assistant: give it enough context and correct a misunderstanding.
- Check the work: distinguish supported facts from invented or uncertain details.
- Own the result: explain what changed and what still needs a person’s decision.
Evaluate these behaviors through work relevant to the role. A payroll specialist and a campus recruiter should not receive the same exercise simply because both use AI.
A 30-minute example for a recruiting role
Give every candidate the same fictional intake brief, tool access, and instructions. Tell them explicitly that AI use is permitted and that the review will include their reasoning and edits.
The fictional manager wants to hire a support engineer. The brief names the product, three recurring tasks, and the interview stages. It does not specify the salary range, final assessment deadline, or whether external AI tools are permitted during the assessment. Those omissions are deliberate test material, disclosed to reviewers so they know what is absent.
The candidate’s task is to prepare an assessment invitation and a list of questions to resolve before sending it. Ask for a draft only. No real candidate information or outbound message is needed.
| Stage | Suggested time | What to collect |
|---|---|---|
| Read the brief | 5 minutes | Questions and assumptions |
| Work with the assistant | 10 minutes | Relevant prompts and first draft |
| Review and revise | 10 minutes | Final draft and corrections |
| Explain the result | 5 minutes | Remaining gaps and next steps |
These timings are starting assumptions. Pilot the exercise, consider accommodations, and adjust it to the complexity of the role before using it in hiring.
What makes one response stronger than another?
A weaker response may sound polished but insert a salary, invent a deadline, and claim that AI tools are prohibited without support in the brief. Asking an assistant for a second version does not repair those mistakes unless the candidate recognizes them.
A stronger response leaves unresolved terms as placeholders, asks the hiring manager specific questions, and improves the invitation’s clarity. It might notice that the candidate needs a support contact and a realistic time expectation. The reviewer should be able to trace the proposed wording to the brief or to a clearly labeled question.
Do not assume that the longest prompt is the best prompt. A concise instruction followed by careful review may produce a more useful result than a complicated template the candidate cannot explain.
A draft rubric the panel can calibrate
| Criterion | Suggested weight | Observable evidence |
|---|---|---|
| Task understanding | 25% | Identifies audience, deliverable, and missing requirements |
| Effective use of AI | 20% | Supplies relevant context and corrects misunderstandings |
| Verification | 35% | Finds unsupported claims and preserves approved facts |
| Explanation and handoff | 20% | Explains edits, unresolved issues, and the next owner |
The weights are an example. Agree the criteria with people who perform the job, then have reviewers independently examine sample responses. Discuss disagreements before evaluating candidates. Keep “not observed” separate from “performed poorly,” and do not invent a universal cutoff from this table.
For a starting worksheet you can edit in the browser, use the LunaPrompts interview scorecard builder.
Follow up on a decision, not on terminology
Choose one sentence from the candidate’s draft and ask what supports it. Then change one condition: the assessment deadline moves, the role location is still undecided, or a policy contradicts a suggested sentence. Ask what the candidate would revise.
Useful questions include:
- Which detail did the assistant add that you could not verify?
- What information would stop you from sending this invitation?
- Which part would you ask a colleague to check?
- What would you change if the audience were an internal candidate?
A candidate may give a strong explanation while producing a flawed artifact, or the reverse. Record both. The purpose is to understand the evidence, not to reward confidence or familiarity with AI vocabulary.
Give candidates a consistent opportunity
Provide the same approved tool access and a short orientation. Do not require a personal paid subscription. Explain what work will be collected and which sources are allowed. Test the exercise using the equipment and connection conditions candidates are likely to have.
Keep the task related to essential job activities. Do not score accents, eye contact, personality impressions, or the speed of speech as proxies for AI fluency. Where an accommodation or alternative format is needed, involve the responsible team before the session.
Where LunaPrompts fits
LunaPrompts offers practical AI skills assessments and AI Viva follow-ups that help teams inspect work and explore a candidate’s reasoning. Bring your role brief and proposed exercise to a pilot discussion, and confirm the task format, permitted tools, evidence capture, and report you need.
For engineering roles, use the more specialized AI engineering assessment templates. For HR colleagues learning the workflow themselves, the 15 ChatGPT prompts for HR provide practice material. This article’s non-technical exercise is a proposed task to calibrate, not a claim that a prevalidated version is available for every role.
Questions hiring teams ask
Can we assess AI fluency without a technical interviewer?
A reviewer who understands the role can inspect a role-relevant draft, check its facts, and ask about decisions. Technical tasks still need reviewers with the corresponding expertise.
Should AI be allowed in the assessment?
If the capability being tested is working effectively with AI, provide approved access and clear rules. If you are testing unaided knowledge, communicate that different condition explicitly.
Does a good result prove the candidate will perform well at work?
It provides evidence from one task under stated conditions. Combine it with other job-relevant evidence, and review the assessment’s usefulness over time rather than treating one result as a complete prediction.