The Line Isn't Whether AI Touched It — It's What You Signed Up To

The usual framing of this question — is it acceptable to use a chatbot on your resume — cannot be answered, because “using AI” covers everything from fixing a comma to inventing a job. What can be answered is the question underneath it: a resume is a set of claims you are agreeing to be held to, and the only thing that matters is whether you can stand behind every sentence of it in a conversation with no notes in front of you. If yes, it does not matter what typed it. If no, it does not matter that you typed it yourself.

That reframing does more work than a pros-and-cons list, because it tells you which specific edits are fine and which are not, instead of giving you a verdict on a tool.

What a resume actually is

A resume is not a writing sample and it is not an essay. It is closer to a statement of facts submitted in support of an application, and it is treated that way: an employer will check some of it, ask about more of it, and in many places require you to confirm its accuracy in writing before you start.

Two consequences follow, and they explain most of the confusion around this topic.

First, nobody promised to write it unaided. There is no convention, and almost never a rule, that a resume must be composed without help. People have used templates, books, friends, mentors, career services, and paid writers for decades, entirely openly. A drafting tool joins a long list. There was no unaided-authorship norm for it to violate.

Second, you are the one who has to answer for the content. That obligation does not transfer to anything. This is why the ownership question is the whole question — not because using a tool is suspect, but because the consequences of what the document says land on you exclusively.

Where the line falls in practice

Sort the edits by whether they change what the document claims.

Not a claim, so not an issue. Fixing grammar and punctuation. Cutting a sentence that ran long. Suggesting a clearer word for something you already said. Making tense and formatting consistent. Reordering bullets so the strongest comes first. Turning three rambling notes into one tidy line whose content you supplied. None of this alters what you are asserting; it alters how legibly you assert it.

A claim, so yours. Every number. Every verb describing your role. Every scope statement. Every skill in the skills list. Every date. Whether an outcome was yours or your team’s. Whether a qualification is complete. Keeping that boundary explicit as you work is the practical form of the whole thing, and the reason the boundary needs stating is that generated prose crosses it constantly without signalling that it has.

The failure mode is not deliberate. You ask for a bullet to be made “stronger,” and what comes back is stronger because it claims more — a bigger verb, an implied sole ownership, a plausible improvement figure you never measured. Drafted bullets drift in consistent directions, and the drift is upward. The document did not become dishonest because a tool was involved; it became dishonest because a claim changed and nobody checked.

The three cases where the answer is genuinely no

Most of this topic is permissive. Three situations are not, and they are worth separating out because they get lost in the general reassurance.

The document asserts something you did not do. This is not about tooling at all — a fabricated employer, a degree you did not finish written as finished, a title you never held, a system you have never used listed as a skill. It is the same problem whether it arrived from a tool or from wishful thinking, and it is the only version of this question with serious consequences. Employment is routinely ended over it, years later, when it surfaces.

Unaided work was the point. If an employer sets a written exercise, a take-home task, or an assessment, the thing being measured is you. Using a tool there is not a phrasing shortcut, it is answering a different question than the one asked. This holds even where nobody said “no AI,” because the purpose is evident.

A stated rule says otherwise. If an application form asks whether AI was used, or a portal’s terms state a policy, answer it accurately. This case is easy and people make it hard: an honest yes to “did you use AI tools in preparing this application” is very rarely disqualifying, and a false no is a written misstatement on an application form, which is the worse of the two by a wide margin.

Disclosure, when nobody asked

Should you volunteer it? Generally there is nothing to volunteer, for the same reason you would not announce that you used a spell-checker or that a friend read it over. The document is a joint product of your history and whatever help you used, as it has always been.

Where it becomes worth mentioning is when a specific claim on the page came from somewhere other than your own recollection. In practice that should not happen, because such claims should be deleted rather than disclosed — but if you find yourself thinking “I would want to explain how that line came about,” the line should not be on the page.

If the question does come up directly, answer about the content, not the tool: what you did, how you know, who could confirm it. That is the ground you can defend, and it is available to you as long as the claims trace to something real.

The workflow that makes the question stop coming up

Write down your history first, in plain language, before opening any tool — real dates, real titles, what you actually did, in whatever words come out. That document is the input, and having it changes everything downstream, because from then on the tool is editing your account rather than inventing one.

Then, when a draft comes back, read it once against that record with a single question per line: is this still what I said? Anything that grew, delete or narrow. Anything with a number, check the source. Anything claiming sole ownership of shared work, restore the distinction between your contribution and the team’s.

Finally, before sending, do the sceptic’s pass — line by line, how do I know this, who would say the same. Twenty minutes, and it also happens to be most of your interview preparation.

Do that and the ethics question dissolves. You are holding a document whose every claim is yours, which you can talk through cold, and which would be exactly the same document if you had written it longhand. The tool changed the typing. It did not change what you are answering for. Applied to the wider application — forms, letters, assessments — the same principle scales, but the strictness varies by artefact.