AI fluency interview questions: ask two, then watch one task

AI fluency interview questions test whether a candidate can work well with AI tools: what they hand off, how they ask, whether they catch mistakes, and whether they own the result. AI fluency is the practical skill of using AI to get real work done, with good judgment about when and how to trust it.

Here's the catch. A rehearsed story about using AI tells you the candidate has read about AI fluency. Fifteen minutes watching them use it tells you whether they have it. So ask a couple of good questions, then watch one short task.

Key Takeaways

Our position: you can't ask your way to AI fluency. Ask two questions, then watch the candidate do one short task with a planted flaw.
Use four behaviors as your frame: delegation, description, discernment and diligence. Each one maps to something you can observe.
Self-report AI questions are the easiest round to game, because candidates prepare the answers with the same tools.
Plant one problem in the task (a wrong number, a missing constraint, an invented source). Fluent people catch it.
Score what you can see, not how confident the story sounds, and write the levels before the interview.

What is AI fluency, in interview terms?

What is AI fluency, exactly? Anthropic's AI Fluency course organizes it around four Ds: Delegation, Description, Discernment and Diligence. That's a useful frame because each one turns into a behavior you can watch in an interview. Here's how we'd read each one in hiring terms.

Delegation: knows which parts of a task to hand to AI and which to keep.
Description: says clearly what they want, with context and the shape of the output.
Discernment: checks the output and catches what's wrong, missing or made up.
Diligence: owns the final result, protects sensitive data, and is open about AI use.

When a hiring team says it wants a "GenAI fluency" check, this is usually what it means. When candidates prepare for a genai fluency interview, this is what they should practice.

Why asking about AI use is the easiest round to fake

Most AI skills interview questions are self-report. "Tell me about a time you used AI to save time." The answer is a story, and stories are easy to prepare with the very tools you're asking about.

AI use is now normal in many jobs. In the Stack Overflow 2025 Developer Survey, 84% of respondents said they use or plan to use AI tools in their development process. But more of them actively distrust the accuracy of those tools (46%) than trust it (33%).

That gap is the whole point of an ai fluency interview. Using AI is common. Using it with good judgment is the actual skill, and judgment shows up in what people do, not in what they say they do.

AI fluency interview questions on delegation and description

Use two or three of these rather than all of them, and for each one, listen for the strong signals and watch for the red flag.

"Walk me through the last piece of work where you used AI. Which parts did you keep for yourself, and why?" Strong: a clear split with reasons. Red flag: "I just let it do the whole thing."
"What's a task in this role you'd never hand to AI?" Strong: a specific task and a reason, like judgment calls or sensitive conversations. Red flag: no limits at all.
"Show me a prompt you've written that worked well. What made it work?" Strong: context, constraints and an example of the output wanted. Red flag: a one-line question.
"When the first answer misses, what do you change?" Strong: they add context, tighten the ask, or split the task. Red flag: they just ask again.

Interview questions to assess AI skills in discernment

Discernment is where most candidates fall down, and where the best ones stand out. These interview questions to assess AI skills focus on catching errors.

"Tell me about a time AI gave you something wrong. How did you notice?" Strong: a specific check they ran. Red flag: "It's usually right."
"How do you check a number or a source that an AI tool gives you?" Strong: they go back to the original. Red flag: they trust it if it sounds right.
"What kinds of mistakes do these tools make most in your work?" Strong: a concrete pattern, like invented citations or wrong totals. Red flag: vague worry with no examples.
"How do you know when a draft is good enough to send?" Strong: they name their own standard. Red flag: "When it reads well."

Questions on judgment and diligence: data, disclosure and when not to use AI

These are the questions most guides skip, and they matter most when you think about operational and reputational risk.

"What would you never paste into a public AI tool?" Strong: customer data, personal information, confidential numbers, with a reason. Red flag: they haven't thought about it.
"When would you choose not to use AI at all?" Strong: high-stakes judgment, sensitive messages, or tasks where checking takes longer than doing.
"If AI helped with a piece of work, do you tell people? When?" Strong: a clear habit, matched to the team's norms.
"Who's responsible if an AI-assisted report has an error in it?" Strong: "I am." Anything else is a red flag.

Ask two from this group in every interview, whatever the role, because they're short and the answers reveal a lot about professional judgment.

The 15-minute task that beats any question, by role

This is how to assess AI skills in candidates directly. Give them an AI tool, a short realistic task, and 15 to 20 minutes. Plant one problem they should catch. Watch the process, not just the result.

Sales or marketing: draft a follow-up email using account notes and an AI-generated summary. The planted flaw: the summary gets the customer's renewal date wrong.
Operations or support: use AI to draft a reply to a customer complaint. The planted flaw: the draft promises a refund your policy doesn't allow.
Analyst: ask AI to summarize a small dataset and pull three insights. The planted flaw: one total in the AI's summary doesn't match the data.
Product manager: use AI to turn user feedback into a short prioritized list. The planted flaw: the AI cites a "user survey" that doesn't exist.
Recruiter: use AI to write a job post from intake notes. The planted flaw: the draft adds a requirement nobody asked for. Our post on AI skills for recruiters covers what else that role needs.
Engineer: for coding roles, see our guide to the prompt engineering assessment for a fuller treatment.

Allowing AI in this one round is the whole design. If you're unsure about AI in the rest of the process, our framework on whether to allow AI in job interviews helps you set a clear policy.

How to score AI fluency without being an AI expert

You don't need to understand how language models work to score this. You need a short rubric on observable behaviors, written before the interview.

Delegation: did they split the task sensibly, or hand over everything?
Description: did their prompts give context, constraints and an output shape?
Iteration: when the first answer missed, did they improve the ask or just retry?
Discernment: did they catch the planted flaw? Did they check anything at all?
Ownership: is the final output something they'd stand behind, and did they say what AI did?

Give each a weak, adequate or strong level, with one line describing each. Pair it with consistent behavioral questions, like the ones in our behavioral question bank, so every candidate gets the same structure.

If you also want a scored prompting round at volume, Expert Hire has an AI Prompt Assessment round, and it isn't only for engineers: hiring teams can run it for non-technical roles too, business roles included. Candidates demonstrate AI prompt-engineering skill by writing prompts to an AI tool for real-world tasks, scored on 8 dimensions including prompt quality, context framing and iteration strategy.

It measures prompting skill, so treat it as a complement to the role-specific task above, not a replacement for it. Our guide to choosing a round type shows where it fits, and results land on the report card with no black-box pass/fail.

Frequently asked questions

What is considered AI fluency?

AI fluency is the practical ability to use AI tools well at work. That means knowing what to hand off, asking clearly, catching mistakes in the output, and owning the final result. Anthropic's course frames it as four Ds: delegation, description, discernment and diligence.

What are some good AI interview questions to ask candidates?

Ask what they'd never hand to AI, how they check a number an AI tool gives them, and what they'd never paste into a public model. Then give a short task with a planted error and watch whether they catch it. The task tells you more than any answer.

How do you tell if a candidate is using AI in an interview?

Detection is hard and getting harder, so design around it. Ask follow-up questions that require specific, personal detail, and use an observed task where AI use is allowed and visible. For dedicated rounds, proctoring signals can help, but a human should always review them.

What are the top AI skills to look for in a non-technical hire?

Look for clear instructions, the habit of checking output against the source, good judgment about sensitive data, and honesty about what AI did. These matter more than knowing specific tools. A marketer or recruiter who catches a wrong number is more fluent than one who knows every model name.

Should you let candidates use AI during the interview?

For an AI fluency round, yes, because that's what you're testing. For other rounds, decide in advance and tell candidates clearly. What matters is that the policy is consistent, written down, and the same for every candidate.

The bottom line on AI fluency interviews

The best AI fluency interview questions are short, and there aren't many of them. Ask two from the judgment group, then spend the rest of the time watching a real task with a planted flaw. Score what you see against levels you wrote first.

If you want a scored prompting round in the same pipeline as your interviews, see how the Expert Hire AI interview platform runs a prompt assessment round.

By TK, Growth at Expert Hire. Last updated October 13, 2026. Reviewed by Anand Suresh, CPO at Expert Hire.