Candidates using AI in interviews: how to set a fair policy

A fair policy on candidates using AI in interviews says three things in plain words: which uses are fine (preparing, researching, polishing a resume), which are not allowed during the interview itself (reading answers generated live, or someone else answering), and what interviewers do when they suspect it: ask sharper follow-up questions and score only what the candidate can explain, rather than accuse. Tell candidates the rule before the interview, apply it the same way to everyone, and keep assistive technology for accommodations outside it.

Candidates are already using these tools. A Gartner survey of 3,290 job candidates in the fourth quarter of 2024 found that 39% used AI during the application process, most often to generate résumé text (54% of those users) and cover letters (50%), according to Gartner's July 2025 release. In LinkedIn research published in January 2026, 81% of people said they have used or plan to use AI in their job search, and 48% said AI tools boost their interview confidence. More figures, each linked to the original source, are in AI in recruiting statistics.

Why do you need a written policy at all?

Without one, each interviewer decides alone. One hiring manager marks a candidate down for a polished, rehearsed answer; another praises the same answer. One recruiter ends a call because the candidate seemed to be reading; another never notices. The candidate experience and the hiring decision both depend on which interviewer they happened to get.

A written policy fixes three things:

• Consistency. Every candidate is held to the same rule, which is the basic test of a fair process.
• Notice. Candidates can only follow a rule they were told about. Gartner's own advice in the release above is for employers to explain how they define acceptable use of AI.
• A response plan. Interviewers know what to do in the moment, so a suspicion leads to a better question rather than an awkward accusation.

What should the policy allow, and what should it ban?

Draw the line by stage. The question at each stage is whether the AI use hides what you are trying to assess.

For many roles, AI use on a take-home task is part of the job. If you want to see how a candidate works with AI, allow it and ask them to show the prompts and explain what they changed. That is often a better test than banning it.

Identity fraud is a different problem with a different response. Gartner's same release reports that 6% of 3,000 candidates surveyed in the second quarter of 2025 admitted to interview fraud, meaning posing as someone else or having someone pose as them. For proxy interviewers and deepfakes, follow how to spot candidate identity fraud rather than this page.

How do you tell candidates about the rule?

Put it in the interview invitation and say it again in the first minute of the interview. Keep the tone neutral: most candidates have no intention of breaking it.

IN THE INVITATION

A note on AI tools: you are welcome to use AI to prepare, research
[company] and polish your application. During the interview itself,
please answer in your own words without live help from AI tools or
other people. If you need an accommodation, including assistive
technology, contact [name, email] before the interview and we will
arrange it.

SPOKEN, AT THE START OF THE INTERVIEW

"Before we start: we'd like to hear your own answers today, so please
don't use AI tools or outside help while we talk. Notes you made
beforehand are fine. Any questions before we begin?"

If your own team uses AI during the interview, such as a note taker or transcription, candidates should hear about that too, and it is only fair to tell them in the same message. The AI disclosure to candidates templates cover that side.

How can interviewers tell if an answer is AI-generated or read out?

They usually cannot be certain, and the policy should say so. No detector can prove an answer was written by AI, and several of the signs below also describe a nervous candidate, a non-native speaker, or someone who simply prepared well.

• A pause, then a fluent, well-organized answer that never mentions a specific person, number or date.
• Answers that cover the textbook points but not the question you actually asked.
• Details that do not match the resume, or change between answers.
• Strong first answers that cannot survive a second follow-up.
• Eyes moving as if reading, with long gaps before each answer.

Interview design does more than detection. Questions about the candidate's own history are hard to outsource: "On your resume you list the warehouse system migration. What broke in the first week?" Then follow up twice: "What did you do next?" and "What would you do differently?" Ask for numbers, ask who else was involved, and change a detail of a hypothetical halfway through. Evidence that holds up under that kind of probing is evidence, whoever helped the candidate prepare.

Some interview tools now flag answers privately for the interviewer. Interview Signal, for example, has a private answer check that flags answers that read as AI-written or read out and suggests a question to try; it never changes a score and never appears in the submittal. Treat any flag like that as a prompt to ask a better follow-up, not as a finding.

What should an interviewer do when they suspect it?

1. Do not accuse. You may be wrong, and an accusation ends the interview either way.
2. Probe. Ask for specifics on the answer in question: what they did, in what order, with what result.
3. Score the evidence, not the suspicion. If the candidate cannot explain their own example, that answer earns a low rating on its merits. The method is in evidence-based interview feedback.
4. Write down observations, not conclusions. "Could not describe what happened after the first week of the migration" is a note. "Was cheating" is not.
5. Escalate as the policy says. Usually to the recruiter, who decides whether a follow-up conversation is needed.
6. Offer a follow-up, not a verdict. A short second conversation, in person where practical, settles most cases fairly. Gartner's 2025 release also found 62% of candidates said they were more likely to apply where in-person interviews are required, so an in-person step need not cost you candidates.

How do you keep the policy fair?

• Separate accommodations. Screen readers, captioning or speech-to-text requested for a disability go through your accommodation process. See ADA accommodations in interviews, and tell interviewers in advance when an approved tool will be in use.
• Do not penalize pauses. Thinking time, accents and second-language phrasing are not evidence of AI use.
• Apply it to every candidate, not only the ones who seem suspicious.
• Keep it out of the notes unless it matters. A suspicion that came to nothing should not follow the candidate into the ATS. What not to write in interview notes covers why.

The policy text to copy

CANDIDATE USE OF AI IN OUR HIRING PROCESS      Version [x], [date]

1. Allowed at any time
   Using AI to research us, prepare for interviews, and draft or polish
   a resume or cover letter. Everything in an application must be true.

2. Not allowed during live interviews
   Live help from AI tools or other people while answering questions,
   and any altered voice or video. Notes prepared in advance are [allowed
   / not allowed].

3. Work samples and assessments
   Follow the instructions for each task. Where AI use is allowed, tell
   us how you used it.

4. Accommodations
   Assistive technology requested as an accommodation is arranged
   through [contact] and is not covered by this policy.

5. How we tell candidates
   This policy is linked in every interview invitation and summarized
   at the start of each interview.

6. What interviewers do if they suspect live AI help
   Ask follow-up questions, score only evidence the candidate can
   explain, record factual observations, and inform the recruiter. No
   interviewer ends an interview or rejects a candidate on suspicion
   alone.

7. Follow-up
   The recruiter may offer a further conversation [in person / on
   camera]. Decisions are based on the evidence from all conversations.

8. Identity
   Suspected impersonation is handled under [identity verification
   procedure], not this policy.

Owner: [name]     Review: every [6/12] months

This policy covers what candidates do. For the rules your interviewers follow with their own AI tools, including note takers, consent and data, use the AI use policy for interviews template.