How to Write an AI Cover Letter That Doesn't Sound Like AI

You paste the job ad into ChatGPT, ask for a cover letter, and get something that is grammatically flawless, professionally formatted, and instantly recognisable as machine-written to anyone who reads applications for a living.

The problem is not that AI writes badly. It writes averagely, on purpose — and average is exactly what a cover letter cannot be when three hundred people applied for the same role.

This guide covers the specific patterns that give it away and how to strip each one. None of it requires paying for anything.

Why the default output reads as AI

Two mechanical reasons, worth understanding because they tell you what to fix.

It is trained on the average of every cover letter ever published online. That average is bland by construction. When you ask for “a cover letter”, you are asking for the centre of that distribution.

It only knows the job ad. If you paste a job description and nothing else, it has no material about you — so it produces confident, unfalsifiable claims about a candidate it has invented. “I am passionate about delivering results in fast-paced environments” is what a letter says when it knows nothing.

Both problems have the same fix: supply your material, not the job ad.

The seven tells

Read anything AI wrote for you and hunt these specifically.

1. The throat-clearing opening

I am writing to express my keen interest in the [role] position at [company].

Recruiters have read this ten thousand times. It contains no information — they know which role you applied for.

Fix: open with the most interesting true thing about your fit. “I spent two years rebuilding a booking system that handled 40,000 users, which is the problem your job ad describes in paragraph three.”

2. Tricolons everywhere

AI loves three-item lists. “Passionate, driven and results-oriented.” “Planning, execution and delivery.” One tricolon is rhetoric; four is a tell.

Fix: keep at most one. Turn the rest into single specifics.

3. Adjectives doing the work of evidence

“Highly motivated”, “proven track record”, “extensive experience”, “passionate about”. These are claims with nothing behind them, and everyone makes them.

Fix: replace each with a fact. Not “proven track record of improving processes” but “cut our monthly close from nine days to four.”

4. Uniform sentence rhythm

AI writes sentences of similar length with similar structure. Human writing varies — short sentence, then a longer one that develops the idea, then a fragment. Sometimes deliberately.

Fix: read it aloud. Where you hear a metronome, break a sentence in half.

5. Restating the job ad back at them

As outlined in your job description, this role requires strong communication skills, which I have developed throughout my career.

You have used a sentence to tell them what they wrote.

Fix: delete. Every sentence should add information they do not have.

6. The empty enthusiasm paragraph

I am particularly drawn to [company]‘s commitment to innovation and excellence.

If you could paste another company’s name in without changing anything else, it says nothing.

Fix: name something specific and verifiable — a product decision, a public post, something in their documentation. If you cannot find one, cut the paragraph entirely rather than faking it.

7. Length

AI defaults to four full paragraphs. Most good cover letters are three tight ones, under 300 words.

Fix: cut 30%. It almost always improves.

A prompt that starts from you

The single biggest change: stop pasting the job ad first.

Try this instead, in ChatGPT or Claude:

I am applying for [role] at [company]. Before you write anything, interview me. Ask me one question at a time about my actual experience — what I have built, what went wrong, what I fixed, what I am proud of. Ask at least six questions. Do not write the letter until I say stop.

Answer honestly, including the failures. Then:

Now write a cover letter under 280 words using only what I told you. No adjectives that I did not earn with a specific example. Do not restate the job description. Open with the most interesting true thing.

The difference is substantial, because now the letter is made of your material rather than the model’s assumptions.

Claude tends to produce less generic prose here and is better at keeping your voice when you ask it to edit rather than rewrite — we go into the difference in ChatGPT vs Claude for writing. Both free tiers handle this comfortably; there is no reason to pay for a cover letter.

The specificity test

One question settles whether your letter is finished:

Could I send this to a different company by changing only the name?

If yes, it is not a cover letter. It is a form letter, and it will be read as one.

Every letter should contain at least two things that could only be true of you applying to them: something specific you did, and something specific about them.

The hard cases

Cover letters are hardest when something needs explaining. AI is genuinely useful here, provided you steer it.

A career change. The instinct is to apologise for the gap in relevant experience. Do the opposite — make the transferable thing concrete. Prompt:

I am moving from [field] to [field]. Here is what I did: [specifics]. Write two sentences connecting that to [specific requirement in the ad]. Do not say “transferable skills”. Be concrete about the actual overlap.

An employment gap. State it briefly, factually, without apology, and move on. One sentence. AI will usually over-explain — cut it back. Most employers care far less than candidates assume, and a defensive paragraph draws more attention than a plain clause.

Being underqualified on paper. Address the biggest gap directly rather than hoping nobody notices. “You have asked for five years; I have three, and here is what I did in them” reads as confident. Silence reads as hoping.

Redundancy. Neutral, one clause, no bitterness. “My role was made redundant in a restructure” is complete.

A short stint. Do not explain it in the letter at all. It is an interview conversation, not a cover-letter one.

The pattern across all of these: one honest sentence beats a defensive paragraph. AI defaults to the paragraph. Cut it.

Sector differences worth knowing

Start-ups and small companies read for specificity and directness. Shorter is better and personality helps.

Large corporates and public sector often screen against the person specification first. Mirror their language for the essential criteria — not by restating the ad, but by making sure the words they will search for genuinely appear in your evidence.

Academic and research roles want detail and precision over brevity, and the conventions are different enough that a generic AI letter reads badly. Ask for the draft, then rewrite it against a letter from someone in your field.

Creative roles treat the letter as a writing sample. If it reads as machine-written here, that is disqualifying in a way it is not elsewhere.

What never to let AI write

Two things.

The reason you want this job. It does not know, and what it invents will be generic and slightly false — which is precisely what you will be asked about in the interview.

Anything you cannot defend. If the letter claims you “led a team through a complex migration”, you had better be able to describe that migration in detail. The most damaging use of AI in applications is not detection. It is arriving at an interview unable to substantiate your own letter.

Before and after

Before, from a job ad alone:

I am writing to express my keen interest in the Operations Coordinator position. With extensive experience in fast-paced environments and a proven track record of delivering results, I am confident I would be a valuable addition to your team. I am particularly drawn to your company’s commitment to innovation and excellence.

Four sentences, no information.

After, from the interview prompt:

I spent eighteen months as the only coordinator for a team of forty, which meant I built the scheduling process from scratch after the previous system collapsed during a merger. Your ad mentions consolidating two operations teams — I have done exactly that, badly the first time and considerably better the second.

Same length. One is a person.

The last pass

Before you send:

1. Read it aloud. Anything you would not say out loud, rewrite.
2. Delete the first sentence. Check whether the letter is better without it. It usually is.
3. Count the adjectives describing you. More than two, cut.
4. Run the specificity test.
5. Check the spelling of the company name. AI gets this wrong more often than you would believe, particularly with unusual capitalisation.

For a final proofread, Grammarly’s free tier catches mechanical errors without the tone suggestions that would flatten the voice you just worked to put in. Accept the spelling fixes; ignore the rest.

If you want more prompt patterns like the interview technique above, our ChatGPT prompts that save time collects several in the same spirit.

The same evidence-first method applies to CVs: see best AI for resume writing for a full before/after bullet example and the Claude + ChatGPT + Grammarly stack.