How to Use AI on a Cover Letter Without It Showing

Let’s dispense with the moralising. Using an AI assistant to help write a cover letter is normal, widespread, and not cheating. Nobody objects to spellcheck, and nobody should object to a tool that helps you restructure a clumsy paragraph.

The practical problem is different and worth taking seriously: a letter written entirely by a model reads like one, and the thing that makes it obvious is also the thing that made the letter worth sending.

Why unedited output fails

Ask an assistant for a cover letter and you get something fluent, well-organised, correctly punctuated, and completely empty. It will say you’re excited about the opportunity. It will say your skills align well with the requirements. It will say you’d bring value to the team.

Every one of those is a claim without evidence, and the reason is structural: the model doesn’t know your history. It knows the posting and it knows what cover letters sound like. So it produces the average of all cover letters, which is exactly the register readers have learned to skip.

The failure isn’t stylistic. It’s that the letter contains no information. Recall what a cover letter is actually for — the one job is connecting your specific experience to their specific problem. That connection requires knowing which of your experiences matters here, and only you have that.

You can hand the model that knowledge. Most people don’t, which is why most AI-drafted letters are bad.

The tells

If you’ve read a lot of these, the texture is recognisable. It’s rarely one smoking gun — it’s the accumulation.

Abstract praise with no referent. “Your commitment to innovation in the space.” Which innovation? The model doesn’t know, so it gestures.

Perfectly balanced sentences. Real writing is lumpy. Model prose has an even, slightly rehearsed rhythm — every paragraph roughly the same length, every sentence roughly the same shape.

Enthusiasm adjectives doing structural work. Thrilled, excited, passionate, delighted. One is fine. Four in three paragraphs is a signature.

The triad habit. “Collaboration, communication, and continuous improvement.” Three abstract nouns in a row, repeatedly.

No numbers, or implausibly round ones. A model that doesn’t know your outcomes will either omit them or invent something like “increased efficiency by 30%.” The second is worse than the first — see below.

Nothing surprising. The clearest tell of all. Not a single sentence a reader couldn’t have predicted from the job title.

Readers can’t prove any of this, and they generally don’t try. They just find the letter unpersuasive and move on, which has the same effect as being caught without the drama.

A workflow that works

The useful principle: you supply the substance, the model helps with the shape. Never the reverse.

1. Write the raw material yourself, badly. Five minutes, no structure, bullet points, bad grammar, whatever comes out:

  • posting wants regional reporting consolidated across teams
  • did exactly this at northwind, 3 regions, took ~6 months
  • hard part wasn’t the tech, took 6 wks. hard part was the regions each defined “completed delivery” differently and nobody had noticed
  • monthly close 4 days → 1 day
  • want to ask whether their regions have own analysts, changes whether it’s a build or coordination job

This is the whole letter. Everything in it is true, specific, and unavailable to any tool. It took five minutes because you’re not performing, just remembering.

2. Ask for structure, not content. Give it your notes and the posting, and be explicit about the constraint:

Here are my rough notes and the job posting. Organise these into a four-paragraph cover letter of about 300 words. Use only the facts in my notes — do not add claims, achievements, or enthusiasm I haven’t written. Plain direct register, no “I am writing to express.”

The prohibition matters. Without it, the model fills gaps, and the gaps are exactly where fabrication happens.

3. Read every sentence and delete anything you didn’t supply. This is the step people skip and it’s the entire point. If a sentence contains a fact you didn’t provide, it’s either invented or generic. Either way, cut it.

4. Rewrite the first and last sentences yourself. Those get read most closely and they’re where model prose is most recognisable. Write them in your own words even if the middle stays as drafted.

5. Read it aloud. If it doesn’t sound like something you’d say, change the words until it does. Contractions help. So does a sentence fragment. So does an aside.

Steps 3 to 5 take about as long as step 2 saved. That’s fine — the model didn’t save you time, it saved you the blank page, which was the actual problem.

Other genuinely useful asks

Beyond drafting, some prompts are reliably good because they exploit what the model is actually good at:

  • “Which requirement in this posting is most central, and which is boilerplate?” Useful triage, and you’ll often disagree — which is itself informative.
  • “This paragraph is 140 words. Cut it to 80 without losing any specific fact.” Compression is a genuine strength.
  • “What would a hiring manager be sceptical about after reading this?” Frequently surfaces the hedge you were hoping nobody would notice.
  • “Rewrite this sentence three ways, all plainer.” Then pick one and edit it.

Notice all four operate on material you already wrote.

The line you don’t cross

Never let it invent an achievement or a number. This is the one that actually damages you. A model asked to make a letter more impressive will produce “reduced processing time by 40%” without hesitation, and it looks completely at home. Then someone asks about it in an interview and you have to either explain a figure you can’t source or admit you don’t know where it came from. Neither is survivable.

The test: for every number and claim in your letter, could you answer how did you measure that in two sentences? If not, it doesn’t go in — regardless of whether you or a tool wrote it.

Don’t let it invent enthusiasm either. “I’ve long admired your work in this space” is a small lie that a single interview question dismantles. If you don’t have feelings about the company, write about the work; that’s a real reason and a good one.

Should you disclose it?

No, and nobody expects you to. Using a writing tool to draft a document is not a disclosable act any more than using a thesaurus is. Employers who genuinely care ask you to write something live, which is a reasonable thing to ask.

What you are accountable for is the content. Every fact in that letter is a claim you’re making, and “the AI wrote it” is not a defence anyone will accept. Sign it, own it.

The short version

Use the tool for structure, compression, and getting past the blank page. Supply every fact yourself. Delete anything you didn’t supply. Rewrite the opening and closing in your own voice. Read it aloud.

Do that and nobody will wonder whether you used AI, because the letter will contain things no model could have known — which was the only thing that made it worth sending.