If an Interviewer Asks Whether You Used AI

Sooner or later somebody asks it out loud: did you use AI to write this? It is a question the whole online conversation about AI-written applications manages to skip, because that conversation is about whether anyone can tell. Being asked directly is a different situation, and it needs an answer rather than a theory of detection.

The short version: say what you actually did, briefly, and then talk about the work. Almost every bad outcome here comes from treating a normal question as an accusation.

It can land at any stage, including the first screening call, where it tends to arrive as small talk while somebody reads your resume back to you.

Why it gets asked

Usually one of three reasons, and only one of them is adversarial.

Curiosity about tools. Interviewers are working people who also use these things, and “did you draft this with AI?” is often the same kind of question as “what do you use for notes?” A great many roles now involve using these tools, and how you use them is relevant information rather than a confession.

A check that you can discuss your own material. The interviewer noticed that the document reads smoothly and generically and wants to know whether the person matches the page. This is the common case, and the question behind the question is not “did you use a tool” but “is any of this yours”.

A stated policy. The employer said not to, somewhere in the instructions, and is now checking. Rarer, and a different conversation — see below.

Nobody asks this to trap you into a lie. But answering evasively converts a neutral question into the interesting one.

The answer is the true one, said briefly

Name what the tool did and what you did. Two sentences, then move to substance. As a shape — these are illustrative lines, not a script to memorise:

“I wrote my notes and history myself, then used a model to tighten the phrasing. Every number and date came from my own records.”

“I used it to draft a first version of the summary and then rewrote most of it, because the draft claimed a scope I do not have.”

“For the cover letter, I used it to cut my draft down. The content is mine — I could not have got what I know about your product out of a model.”

Each of these is short, specific about the division of labour, and immediately followed by something about the work. That last part matters: it moves the conversation from your tooling to your record, which is where you want it.

What is being assessed is not tool use. It is whether the person in the room and the document on the table are the same candidate — which is close to what employers who ask about this are usually after.

Hedging costs more than the fact

The bad answers are all the same shape: a long qualification, a defensive tone, or a flat denial that the following ten minutes will undermine. A denial is especially expensive, because it converts an ordinary process question into an integrity question, and integrity questions do not get resolved in your favour by argument.

There is also no need to volunteer a confession. If nobody asks, do not explain your workflow unprompted — it is not an ethical disclosure obligation, it is a tool.

If you cannot describe your own resume, that is the actual problem

The question only becomes dangerous when the honest answer is that you do not know what your resume says. That happens when a tool supplied a specific you never verified: a figure it estimated from a placeholder, a scope word it upgraded, a responsibility it inferred. Generated lines drift in a small number of predictable places, and you will not spot them under pressure because they read as familiar.

The fix is entirely upstream. Before any interview, read every line you did not write from scratch and confirm you can describe it as your own work — the same pass as checking a bullet against your record, done cold. If a line fails, the fix is to change the line, not to prepare a better answer about it.

If the employer said not to

Then the subject is not AI. It is whether you followed a stated condition of the application, and that is a straightforward matter of fact.

If you did follow it, say so plainly and specifically: what you did instead, and what your own process was. If you did not — because you skimmed the instructions, or because you interpreted “your own words” loosely — the recoverable answer is the accurate one, given without elaborate justification. It may end the process. A denial that unravels later ends it worse and closes the employer permanently.

This is why it is worth reading the instructions on each stage before starting: the rules genuinely differ between the resume, the letter, and an exercise, and not every part of an application carries the same expectations.

Do not argue about detection

A tempting response is to challenge the premise — that no tool can reliably identify AI-written text, that a classifier’s output is not evidence. It happens to be a reasonable position, and it is the wrong thing to say in an interview. It sounds like a lawyer’s answer, it implies you have thought about how not to be caught, and it makes the tool the topic for another five minutes.

If a classifier score is actually put in front of you, ask what it is being used for, state what you did, and leave the methodology alone. What a checking step could even consist of is a genuinely unsettled question, but you do not want to be the one relitigating it while being evaluated.

The reverse question is fair

You can ask them. What is your policy on candidates using these tools, and do people here use them in the work? The answer tells you something concrete about the team — whether it has thought about the question at all — and it reframes the exchange as two working people discussing tools, which is what it usually was.

If the interview then moves on to your actual history, you have handled it correctly. Have the questions your resume invites ready, because the next few minutes will be exactly those.