How to Answer "How Do You Use AI in Your Work?" in an Interview

12 min read

Published August 22, 2026

Interviewers score the AI question on judgment, not enthusiasm: the four things they listen for, a 90-second answer formula, and scripts for all seven versions.

"How do you use AI in your work?" is scored on judgment, not enthusiasm. Interviewers are listening for four things: when you reach for AI, how you direct it, how you verify its output, and whether you stay accountable for what ships. A strong answer runs about 90 seconds and contains six elements — named tools by task, one specific workflow, a quantified result, an iteration detail, a failure you caught plus the fix, and a boundary where you don't use AI. Both extremes fail: "I don't really use it" reads as incurious, and "I use it for everything" reads as a judgment risk.

That's the answer in compressed form. The rest of this page is the evidence behind it and the execution: what the question is really testing (with the employer-side research most guides never cite), a weak answer and a strong answer diagnosed line by line, scripts for all seven versions of the question, the play for people who barely use AI, and the two situations where AI use gets candidates disqualified rather than hired. The whole thing also exists as a free 10-page PDF guide you can prep from the night before.

Why does the AI question show up unannounced?

The AI question is the new "greatest weakness": decision-relevant, nearly universal, and almost never on the formal script — which is exactly why nobody has a rehearsed answer for it.

The demand side is unambiguous. In Microsoft and LinkedIn's Work Trend Index, a survey of 31,000 workers and leaders, 66 percent of leaders said they would not hire someone without AI skills, and 71 percent said they would hire a less experienced candidate with AI skills over a more experienced one without. PwC's AI Jobs Barometer, built on close to a billion job ads, found workers with AI skills command a wage premium averaging 56 percent. Public company memos pushed the expectation into hiring: Shopify CEO Tobi Lütke's April 2025 memo, posted publicly on X, was titled "Reflexive AI usage is now a baseline expectation at Shopify" and tied AI use to performance reviews and headcount requests. Similar directives followed — not all of which held: Duolingo's CEO later backed off evaluating employees on AI usage after public backlash. The hiring-side expectation stuck anyway.

But the screening side is informal. A Harvard Business Review analysis of 6,380 recorded first-round interviews, run with BrightHire and Harvard Business School's Managing the Future of Work project, found that only 2.2 percent of 2025 interviews included explicit AI questions. The question arrives conversationally — one probe dropped into an ordinary screen, or hidden inside "walk me through a process you improved" while the interviewer listens for whether AI shows up.

Those two facts together define the opportunity. The expectation is everywhere; the formal test is nowhere; so the whole evaluation compresses into one or two unrehearsed conversational moments. Job seekers describe the same blindside in public forum threads — asked about AI in interviews for roles that have nothing to do with AI, in marketing, operations, HR, finance, support. A prepared 90-second story wins the entire exchange, and preparing one takes an evening. It often lands in the recruiter screen, before you've met the hiring manager.

What are interviewers scoring when they ask about AI?

Interviewers aren't screening for AI enthusiasm — they're screening against risk, and the risk has a price tag. Research by BetterUp Labs and Stanford's Social Media Lab, published in HBR, named the failure mode "workslop": AI output polished enough to pass and wrong enough to cost. In their survey of 1,150 US desk workers, 40 percent had received it from colleagues, each incident took about two hours to resolve, and the invisible tax worked out to roughly $186 per employee per month. KPMG and the University of Melbourne, surveying 48,000 people across 47 countries, found 66 percent of employees who use AI have relied on its output without evaluating it — and 56 percent have made work mistakes because of it.

Every interviewer asking the AI question has seen some version of that damage. So the question behind the question is a four-part rubric:

The third row is the differentiator. Nearly every candidate can now say they use AI; almost none can describe how they verify it. One concrete verification habit — "I check every number and citation before anything leaves my desk" — separates you from the pile, because it answers the exact fear the workslop research quantified.

Why do both extremes get rejected?

The trap in the AI question is a double bind: the two most natural answers both fail, for opposite reasons.

"I don't really use it" reads as an adaptability red flag. In a market where 71 percent of surveyed leaders prefer AI skills over experience, "not really" sounds like "not curious" — even when it's honest.

"I use ChatGPT for everything!" reads as a judgment red flag. Unqualified enthusiasm signals exactly the unverified-output risk the question exists to screen out — and enthusiasm without specifics is indistinguishable from faking, a pattern interviewers complain about openly: candidate after candidate giving nearly the same generic AI answer.

The skeptical hot take fails too. A dismissive framing — AI as overhyped autocomplete, delivered as a verdict — converts a defensible judgment into an attitude problem. Skepticism survives interviews only when it's specific: a limit you found, in a task you ran, with the check you now use.

The needle you're threading is enthusiastic, specific, and in control: you use AI deliberately, you can prove it with details, and you're clearly the one steering. A few phrasings work against you often enough to treat as banned:

• "I use it for everything."
• "I've been experimenting lately…" — a vague timeline signals no real usage
• Claiming months of use while having no saved or refined prompt to show
• Being unable to name a single time AI got something wrong for you
• Philosophy about the future of AI in place of a workflow you actually run

What does a strong 90-second answer contain?

Interview Drills calls the format the 90-Second AI Answer: six elements, in roughly this order, delivered as one specific story rather than a tool list.

1. Named tools, by task. Not "I use AI" — which tools for which jobs.
2. One specific workflow. A real process you run, step by step.
3. A number. Time saved, turnaround cut, output improved.
4. An iteration detail. Proof you refined it: "took five or six versions before it was reliable."
5. A failure plus your fix. The credibility element almost everyone omits.
6. A judgment boundary. Where you deliberately don't use AI, and why the accountability stays with you.

Here is the format assembled, with the elements marked — adapt the structure, not the words:

"I keep two tools open for different jobs — Claude for drafting and analysis, ChatGPT for quick research summaries (1). My main workflow is client proposals: I feed in the brief and past examples, generate a structured first draft, then rewrite the strategy section myself — that part has to be mine (2). It cut my turnaround from about three days to one (3). The prompt took me five or six iterations before the drafts stopped being generic (4). It has burned me — early on it invented a statistic, so now I verify every number and citation before anything goes out (5). And I don't use it for final pricing or anything touching confidential client data. It's a power tool — I'm still the one accountable for what ships (6)."

Read aloud, that's about 90 seconds: long enough to be substantive, short enough to invite follow-up on your terms. The failure story is the element that earns the most trust per second, because it can't be faked by someone who started using AI yesterday — it converts "I use AI" into "I have judgment about AI," which is the thing actually being scored.

The formula travels across roles by swapping the workflow, never the structure:

• Operations or PM: meeting synthesis and first-draft status reports — "Friday reporting went from three hours to thirty minutes of editing; the summaries missed action-item owners at first, so I added a check for that."
• Finance or analyst: first-pass variance narratives or formula drafting — "every figure still ties back to the source workbook before anything leaves my desk."
• Customer-facing: response drafting and call summaries — "anything going to an upset customer I write myself; AI drafts the routine tier."

Seniority shifts the boundary element, not the formula: individual contributors should anchor on their own verification habit, while managers score better anchoring element six on the team — where you allow AI on your team, what review you require before AI-assisted work ships, and who stays accountable.

What separates a weak AI answer from a strong one?

The fastest way to see the 90-Second AI Answer's value is a diagnosis. Here is the answer most candidates give:

"Yeah, I'm a big believer in AI. I use ChatGPT all the time — emails, research, brainstorming, you name it. Honestly it's like having an extra assistant. I'm really excited about where the technology is going, and I think companies that don't adopt it are going to fall behind."

It sounds fine in the room, and it fails on every scored dimension:

• No task mapping — "all the time" and "you name it" are the opposite of element 1; heavy use with no selectivity is the workslop profile.
• No workflow, no number — nothing here is checkable or memorable; it's a vibe, not evidence.
• No failure, no verification — the differentiator row is completely absent, so the answer never addresses the interviewer's actual fear.
• No boundary — "like having an extra assistant" hands off accountability instead of keeping it.
• The closing opinion adds risk — "companies that don't adopt it will fall behind" is a hot take, and this candidate is one follow-up ("walk me through your prompt") from having nothing to show.

The strong version of the same candidate is the model answer above: identical enthusiasm, but routed through one workflow, one number, one failure, one boundary. Nothing about the candidate changed — only the narration did. That's worth internalizing, because the AI question is a narration test: people who genuinely use AI every day still fail it by never having said their workflow out loud.

What are the seven versions of the AI question?

One rehearsed core answer covers nearly every AI question you'll face — what changes is which element leads. This is the full taxonomy:

Then there's the layer most guides skip: strong answers get probed. "Show me the prompt." "What did it get wrong?" "Walk me through how you used it yesterday." This is the Follow-Up Test — the same probing that exposes memorized behavioral answers — and it's where invented AI experience dies, because a claimed months-long habit with no saved prompt and no named failure collapses in one exchange. If you have a refined prompt saved anywhere, screenshot it before the interview: it's the single most convincing artifact you can reference, and almost nobody has one ready.

How do you answer if you barely use AI?

You don't need months of AI history — you need true stories, and true stories can be manufactured deliberately in one week. Interview Drills calls this the One-Week True-Story Plan:

1. Tonight, 15 minutes: pick three real tasks from your actual job — things you did this month. A report, an analysis, an email sequence, a plan, a spreadsheet.
2. This week, 30 minutes per task: run each through an AI tool. Keep a simple log: what you asked, what worked, what it got wrong, how you caught it. That log is elements 2, 4, and 5 — workflow, iteration, failure-and-fix.
3. This weekend, 10 minutes: assemble your best task into the six-element format and say it out loud once.

You're not faking experience — you're compressing it. And the honest framing scores fine: "I've been deliberately building AI into my workflow recently — here's what I've learned so far, including what it gets wrong." A recent-but-real answer with a verification habit beats a fake veteran every time, because the fake veteran fails the Follow-Up Test and the honest beginner doesn't.

If your employer bans AI at work, name the policy and pivot: "My current company restricts AI tools, so I've been building fluency on personal projects — here's one." Policy awareness is itself evidence of the responsible-use instinct the rubric is scoring.

When does AI use get you disqualified instead of hired?

The same skill that wins the AI question ends candidacies when deployed in the wrong place. Three rules cover every case:

AI for preparation is safe and increasingly expected. Researching the company, drafting stories, generating practice questions, rehearsing answers — universally accepted, and legitimately part of your answer to the AI question itself. (If you want the tool landscape for this, see the best AI interview prep tools.)

Covert AI use during a live interview is disqualifying. Amazon warns candidates that generative AI use in interviews risks disqualification, and Goldman Sachs prohibits ChatGPT and external sources in its campus-recruiting interviews. Beyond the policy risk, interviewers have learned to probe past teleprompter-style answers — and the moment a follow-up leaves the script, the whisper-bot user freezes. Real-time answer-feeding doesn't just risk the offer; it prevents you from ever building the fluency the question is testing.

For take-home assignments, read the instructions and disclose. Policies now split by company — Anthropic, notably, reversed its blanket ban in 2025 and now allows AI in applications and take-homes when instructed, while keeping live interviews AI-free. When use is allowed or ambiguous, disclose cleanly:

"I used AI to draft the initial structure; the analysis, judgment calls, and final version are mine. Happy to walk through exactly how I used it."

That sentence converts a gray area into a demonstration of accountability — the fourth row of the rubric, out loud.

One absolute, in either direction: never feed a current or former employer's confidential data into public AI tools, and never imply you have. It's an instant disqualifier — and its mirror image, "I don't put confidential data into public models," is one of the strongest boundary statements you can make.

How do you make the answer automatic under pressure?

Knowing your answer is not the same as saying it under pressure — and the AI question arrives unannounced, mid-conversation, with follow-ups. People who use AI daily still bomb it, not from ignorance but because they've never narrated their workflow out loud. That failure mode — "I knew the answer, I just couldn't say it" — is the same retrieval-under-pressure gap that makes minds go blank in interviews, and reading about the formula exercises neither retrieval nor delivery.

The one-evening protocol:

1. Fill in the six elements with real details from your own work.
2. Say the 90-second answer out loud, twice — rehearsing alone works if you do it aloud, not in your head.
3. Screenshot your best refined prompt — your Follow-Up Test insurance.
4. Pick your lead element for the judgment version: boundary plus verification.
5. Run the disclosure script once so it's available if the integrity question comes.

The printable version of this protocol — with the worksheet, the cheat sheet, and all seven scripts — is the free 10-page AI Question guide (PDF), no signup required.

Interview Drills is an AI-powered interview training platform designed for people who have a real interview coming up. It diagnoses interview weaknesses with a free 3-minute quiz, prescribes a 14-day training track of practice drills, scores spoken answers on six dimensions in under 60 seconds, and personalizes scoring and feedback to the candidate's resume and target job.

Speak your 90-second AI answer into a drill rep this week and listen for the weak-answer patterns above in your own delivery — a tool list, a missing number, no failure story. A natural place to run it: your AI workflow story doubles as evidence in the "What are your strengths?" drill, where a workflow, a number, and a verification habit are precisely what a strong answer needs. If you want the diagnosis first, the free 3-minute quiz (25 questions, no account needed) identifies your interview persona and blind spot.

Where to go next: if the AI question will arrive as a behavioral probe, structure the story with the STAR method and file it in your interview story bank; if you're prepping for the screen where it usually first appears, start with the recruiter screen guide.

Frequently asked questions

What is a good answer to "How do you use AI in your work?"

A strong answer runs about 90 seconds and contains six elements: named tools matched to tasks, one specific workflow you actually run, a quantified result, an iteration detail showing you refined the process, a failure you caught plus the fix, and a boundary where you deliberately don't use AI. Example spine: which tools for which jobs, the workflow, the time saved, how many prompt versions it took, the error you now check for, and where you stay fully manual and accountable.

Is it OK to say you don't use AI in an interview?

A flat "I don't really use it" is risky: 71 percent of leaders in Microsoft and LinkedIn's 31,000-person Work Trend Index said they'd hire a less experienced candidate with AI skills over a more experienced one without. If your usage is genuinely light, say what you have deliberately tried recently, what it got wrong, and how you verified it — an honest recent-adopter answer with a verification habit scores far better than either denial or faked expertise.

Should I admit I used ChatGPT to write my resume or cover letter?

Disclose honestly and keep the ownership: "I used AI to draft the structure; the content, judgment calls, and final version are mine — happy to walk through how I used it." That framing turns an integrity check into a demonstration of accountability. Denying it when your materials read as AI-polished is the losing move, because interviewers increasingly probe with follow-ups you can't script.

Can I use AI during a live job interview?

No — covert AI use during a live interview is a disqualifier at a growing list of companies. Amazon warns candidates that generative AI use in interviews risks disqualification, and Goldman Sachs prohibits ChatGPT and other external sources in its campus-recruiting interviews. Using AI to prepare beforehand is safe and increasingly expected; using it live and covertly is the one move that ends candidacies.

How do I answer AI questions if my employer bans AI at work?

Name the policy, then pivot to deliberate personal use: "My current company restricts AI tools, so I've been building fluency on personal projects — here's an example." Naming the policy is a point in your favor: it shows the responsible-use instinct interviewers are screening for. Then give one real example with a verification step, the same way you would for a work task.

How long should my answer to the AI question be?

About 90 seconds spoken — long enough to include a real workflow, a number, a failure you caught, and a boundary; short enough to invite follow-up on your terms. One-sentence answers read as evasive, and answers past two and a half minutes usually mean you're listing tools instead of telling one specific story.

What AI questions do interviewers actually ask?

Seven recurring types: workflow ("How do you use AI in your work?"), judgment ("When would you not use it?" / "How do you validate output?"), impact ("Tell me about a time AI improved your work"), opinion ("What do you think about AI in our field?"), tool inventory ("What AI tools do you know?"), integrity ("Did you use AI to write this?"), and currency ("What's new in AI?"). One rehearsed core answer covers most of them — you change which element leads.