Using AI to Write Your Application: Where the Line Is

Almost everyone applying for a job now has access to a tool that will write a cover letter in twenty seconds. Almost everyone is unsure whether using it is acceptable. The workplace concept of moving mental tasks into external systems is explained in cognitive offloading.

The uncertainty is reasonable, employers have not been consistent, and there is a workable line. It is not about whether a model was involved. For an independent public reference, consult UK ICO artificial-intelligence guidance.

We make a product that does this. That gives us an interest, and it is also why we would rather set out the limits than pretend there are none.

The test that actually holds

Everything in your application must be true, and you must be able to discuss all of it.

That is the whole rule, and it resolves nearly every case.

Using a model to phrase your real experience more clearly: fine. You are not being assessed on your prose speed.

Using a model to produce experience you do not have: fabrication. It stops being a question about AI at that point; a fabricated CV is a fabricated CV whoever typed it.

Using a model to write something you cannot then explain: the practical failure, and the most common one. A candidate whose cover letter describes a sophisticated project they cannot discuss for two minutes has created a problem for themselves in the interview rather than solved one.

Where it is straightforwardly fine

Structure and phrasing. Turning your notes into a coherent paragraph.

Tightening. Cutting a rambling description to something readable.

Grammar and register, which matters enormously for candidates applying in a second language and is one of the more equalising uses of these tools.

Getting started. A first draft you then rewrite is not the same as a submitted output.

Tailoring. Adjusting emphasis for a role, from material that is already true. See tailoring a CV.

Preparation. Practising interview answers, anticipating questions, rehearsing.

Research. Understanding a company, a role, a technical area.

Where it is a problem

Inventing experience, projects, results or figures. Models will produce plausible specifics when asked to write impressively. Check every number and every claim in anything you generate — this is the most common way an honest candidate ends up with a dishonest application.

Generating a cover letter you do not read. They contain errors: the wrong company name, an inference about the role that is wrong, an enthusiasm you do not have.

Answering during a live interview. Increasingly detected, and where discovered it ends the process.

Take-home tasks where the instructions prohibit it. If the task says no AI assistance, the task is testing something specific, and ignoring the instruction is a straightforward integrity matter. If it does not say, ask. See take-home tasks.

Assessments designed to measure the thing you are outsourcing. A writing test is a writing test.

What recruiters notice

Not "was AI used" — that is mostly undetectable and increasingly beside the point. What they notice is the result.

Generic content. A letter that could go to any company, because it did. This is the commonest tell and it was a tell long before these tools existed.

Register mismatch. Polished prose in the letter, and a different person in the interview.

Unverifiable specifics. Impressive claims that collapse on one follow-up question.

Volume behaviour. Fifty applications from one candidate in a day, all lightly varied.

And errors nobody read. The wrong company name is the classic, and it is fatal for the reason you would expect: it demonstrates that the application was not read by the person who sent it.

See do recruiters notice AI-written applications.

Ask, where you can

The simplest resolution and it is under-used.

Many employers now state a position in the job posting or the application form. Read it.

Where the instructions permit it with disclosure, disclose. It costs nothing and it removes the question.

Where they prohibit it, do not. Even where you disagree with the policy.

Where nothing is said, the test at the top applies: true, and defensible in conversation.

And if a take-home is ambiguous, ask the recruiter. Candidates worry this looks bad. It reads as someone who takes the instructions seriously, which is not a poor signal.

The practical method

If you use these tools, this is the way that does not create problems.

Start from your own material. Your actual history, your actual results, in your own rough words.

Have the model organise and tighten, not invent. Prompt with facts rather than asking it to be impressive.

Verify every specific. Numbers, dates, titles, technologies, outcomes.

Rewrite in your voice. If you would not say it, do not send it.

Read it aloud before sending. This catches the register problem and the wrong company name in the same pass.

Then prepare to discuss it. Take each claim and rehearse two minutes on it. If you cannot, either the claim needs removing or you need to know more about your own experience than you currently do.

That last step is where the value is. A candidate who has done it walks into the interview more prepared than one who wrote the letter by hand and never revisited it.

The uncomfortable part

Everyone is using these tools, which means the application no longer differentiates on writing quality. The bar has moved to what is behind the writing.

Which is a good outcome for people who have the substance and struggle to present it, and a poor one for people who were relying on presentation.

And it means preparation matters more, not less. The interview is now the filter that the written application used to be, and it is the part these tools help with most legitimately.

The short version

Everything true, and everything defensible in conversation. That resolves nearly every case.

Phrasing, structure, tightening and translation are ordinary assistance. Inventing experience is fabrication regardless of what typed it.

Verify every specific. Generated impressiveness produces plausible numbers that are not yours.

Read the instructions, and ask when they are ambiguous — that reads well, not badly.

And rehearse every claim you send, because the interview is where the application gets checked.