Can you use ChatGPT in an AI interview? What each platform's own policy says

There is no single rule, because there is no single AI interview. Each company that runs one has written its own policy, and they differ more than the forums suggest. This is what each one says on its own pages, read again on 28 September 2026.

Swipe the table sideways to see every column.

Career1 is not affiliated with any of these companies. Every line about them comes from their own pages, linked below and in the sources at the end. If your invitation includes its own instructions, those win: micro1 and Karat both say the rule can depend on the interview you are given.

Mercor has the most detailed policy, and it is worth reading in full because it is more permissive than people expect. Its LLM Usage Policy says you may use large language models such as ChatGPT to "ensure the accuracy of your grammar and diction" and to "Refine the phrasing and tone subtly, provided that the fundamental ideas and structure remain your own."

The same page lists what you may not do: use them for "Assessing or evaluating the quality of responses generated by models", for "Providing justifications or explanations for writing purposes", or for "Predicting code behavior or runtime results". It also warns against asking a model to "entirely revise your responses", and ends with the line most people quote: "Submissions that are evidently produced by artificial intelligence will be identified and flagged accordingly."

Mercor's AI interview help page repeats the rule for the interview itself: "Conduct grammar and wording checks as necessary, but refrain from utilizing large language models to compose your responses, rate model responses, generate rationales or explanations, or evaluate code."

Two things the policy does not say. It does not say how a submission is identified, and it does not say what happens after a flag. So neither does this article. Mercor also runs a free practice interview you can take as many times as you like, which is a far better use of your nerves than any tool running beside the real one. The Mercor AI interview guide covers the rest of its published rules.

micro1

micro1's candidate terms, last updated 21 July 2026, say "All answers or responses that you provide during the Interview will be your original work product, and you will answer all interview questions without any third-party assistance, except where explicitly permitted by your interviewer or the interview instructions." The same terms add that micro1 "may also report any violation of these Terms to prospective employers". Its candidate privacy policy lists "information collected from our proctoring tools to assess the integrity of your interviews" among the data it holds, without saying which signals those tools use. Our micro1 AI interview guide covers the interview itself.

Sapia.ai

Sapia's interview is typed, not spoken, which is why its rule reads differently. Its candidate explainer says: "Don't copy answers! Chat Interview checks for plagiarism from various internet sources, and can detect if answers come from content generators like ChatGPT." The same page asks you to "Be sure to check your spelling and grammar" and says the interview "is not timed", so the honest route is to take your time and write in your own words.

Karat

Karat's interview is run by a person, "a professional software developer", according to its candidate page, so it is not an AI interview at all. It is in this list because people ask. The FAQ on Karat's interview page says "Karat has an extensive library of Karat Interviews that both allow and do not allow the use of ChatGPT." In other words, it depends on the interview, and your interviewer will tell you which kind you are in.

HireVue's candidate FAQ does not have a ChatGPT rule. What it does say, in answer to "Can I trick or 'game' a Hirevue Assessment?", is: "Beyond practicing like you would for any other type of interview, nope", and that "there's no course you can take that will help you game an AI-driven job assessment from Hirevue."

Notes are a different matter. HireVue's own interview tips suggest a list of short stories about your past work and say "You can even keep your notes next to you in case you need a reminder during the interview." Prepared notes about your own experience are not the same thing as text generated during the interview, and HireVue treats them differently.

On the employer side, HireVue's cheating and fraud page lists "configurable candidate conduct policies, browser-focus tracking, controls that limit copy-and-paste behavior, and image capture during eligible OnDemand hiring experiences". These are settings an employer chooses, and the page says the checks should be explained "before the interview begins". It also answers the question candidates fear most: "A single behavior should not automatically be treated as proof of fraud."

Some platforms say they can, and none of the pages above publishes how well. Sapia says its interview "can detect" generated text; Mercor says evident AI submissions "will be identified and flagged"; HireVue offers employers "automated flags". Not one of them states an accuracy figure on those pages, so this article does not guess at one, and you should distrust any guide that does.

A flag is also not the same as a rejection. HireVue says plainly that one signal is not proof. Mercor and Sapia do not say what happens next. Where a platform is silent, "we do not know" is the accurate answer.

The more useful point is that detection software is rarely what gives a generated answer away. Three ordinary things do:

• The follow-up question. A conversational interviewer asks what you did, why, and what you would change. A generated answer has no answer to the second question, because nothing happened to you.
• The recording. Someone at the employer can watch it. Reading aloud looks and sounds like reading aloud.
• The sameness. Generated answers are fluent and unspecific in the same way, and anyone reading a stack of transcripts notices.

None of that needs a detector, which is why the question "will I get caught" is less useful than "will this help me". For a spoken interview, it does not.

Here is exactly what Career1 does and does not do, from the product itself rather than a promise about it.

• It does not watch your screen, count your windows or track tab switches. For a vetting interview or a job application, the interview room records your camera, both voices and a transcript. Nothing about your screen or your other windows is recorded, and Career1 does not claim to detect a second device.
• It follows up. The interviewer is told to ask follow-up questions whenever an answer is "vague, impressive, or suspicious", to dig into the projects on your resume by name, and to ask how and why rather than for definitions.
• It reports only what you showed. The report quotes the transcript as its evidence. On the vetting report, each skill on your resume is marked verified only when you discussed it with real detail.
• People can watch the recording. A hiring company sees the recording of an application interview. For the vetting interview, companies on a paid plan or a trial can see it too, unless you hide your profile.
• Someone else may not take it for you. Career1's terms forbid impersonating others, "including by having someone else take your interview", and accounts that do can be suspended.

So the position is simple. Use AI as much as you like to prepare. In the interview itself, answer in your own words, because the one thing the format measures is whether the experience on your resume is yours. The vetting interview is one attempt with no retakes, and a report built on read-aloud answers is a report you then have to live with. The FAQ answers the rest of the common questions about how Career1's interview works.

Everything that makes AI tempting during an interview is more useful the week before it. These uses break no platform's rules that we could find:

1. Predict the questions. Paste the job description and your resume into a chat model and ask what an interviewer would most want to test. Then compare it with the AI interview questions guide.
2. Check your resume against the job. Career1's free resume checker does this with no account, and says what an interviewer would push on.
3. Find the thin spots in your stories. Write four lines about each project, then ask a model to play a sceptical interviewer and ask "what exactly did you do?" three times. Answer out loud, not in the chat box.
4. Review a transcript of yourself. Record a practice answer, transcribe it, and ask what is vague. That is close to what a transcript-based scorer reads.
5. Practise with something that talks back. Career1's free AI mock interview is a private practice interview built from your resume: once a week, no video, no score, written feedback, and never shared with a company.

For the full routine, read how to prepare for a mock interview, and for the interview itself, how to pass an AI interview.

Every source below was opened and re-read on 28 September 2026. Policies change; if one of these pages now says something different, the page wins.

• Mercor, LLM Usage Policy
• Mercor, AI Interview help page
• micro1, Candidate Terms and Conditions, last updated 21 July 2026
• micro1, Candidate Privacy Policy, last updated 21 July 2026
• Sapia.ai, Candidate Explainer
• Karat, Karat Interview and candidate experience
• HireVue, candidate FAQ, interview tips and cheating and fraud mitigation