AI Cover Letter Prompts That Don't Sound Like AI (and Don't Invent Experience)

You've pasted the job description into a chatbot and typed "write me a cover letter." What comes back opens with "I am writing to express my keen interest" and promises "a proven track record of driving results." It could have been sent by anyone, to anyone. Worse, a sentence or two describes things you've never done, and you only notice if you read it slowly.

An AI cover letter that doesn't sound like AI comes from changing the input, not the instruction. Give the model a fact sheet of things that really happened, forbid it from adding to the sheet, then edit the draft against the sheet. This post gives you all three prompts, a list of phrases to ban, and a checklist for the last read.

Why does a generic prompt give a generic letter?

Because it has nothing to work with except the job description. Faced with a role and no evidence, a language model writes the average of every cover letter it has seen. That average is the "passionate team player with a proven track record" letter.

It also fills gaps. If the posting asks for stakeholder management and your CV doesn't mention it, a model asked to write a persuasive letter will often supply a plausible sentence about stakeholders. That's the dangerous failure, because the sentence reads fine.

So the fix has two halves. Supply specifics (the giving context lesson is the general principle), and make "do not add facts" an explicit rule.

Step 1: Build a fact sheet before you ask for anything

Spend ten minutes on this once per application. It's the part a model can't do for you, and it's what makes the letter yours. Use this structure and write in plain, messy sentences. Nobody sees it but the model.

ROLE I'M APPLYING FOR: [title] at [company]
WHAT THEY SEEM TO CARE ABOUT MOST (my reading of the posting, top 3):
1. 
2. 
3. 

MY EVIDENCE FOR EACH (real things only):
1. [What I did. My specific part. Numbers if I have them. What happened.]
2. 
3. 

ONE THING ABOUT THIS COMPANY I ACTUALLY KNOW (product I've used, post I read,
a decision they made I have an opinion on):

GAPS: requirements I don't fully meet, and the nearest thing I have done:

HOW I TALK: [two or three sentences I wrote naturally, e.g. from an email
or LinkedIn post]

THINGS I WILL NOT CLAIM: [anything that would be a stretch]

Two fields do most of the work. "One thing about this company I actually know" stops the model from writing "I admire your innovative culture." And "How I talk" gives it a style to imitate that isn't the default. The post on prompting AI to sound human covers the style side in more depth, including how to fingerprint your own writing.

If you can't fill in evidence for a requirement, that's information. Either you're not a fit for that part of the role, or you've done it and forgotten. Ask yourself the question that an interviewer will ask in round one: "Tell me about a time you did that."

Step 2: The drafting prompt

This is an untested template: I haven't run it against a specific model for this post, so check the first output against your fact sheet. The structure (facts in, rules, format) is the part to keep.

You are helping me write a cover letter. You are an editor, not an author.
The only source of facts is the FACT SHEET below. 

Rules:
1. Use only facts from the fact sheet. If a sentence needs a fact I didn't
   give you, write [NEED: what's missing] instead of inventing it.
2. Do not claim skills, tools, years of experience, team sizes, or results
   that are not in the fact sheet. Do not round numbers up.
3. 220 to 280 words. Three short paragraphs:
   - First: the role, and my single strongest piece of evidence for it, in
     the first two sentences. No "I am writing to".
   - Second: one or two more pieces of evidence, each as: what I did, what
     happened. Tie each to something the posting asks for.
   - Third: the one thing I know about the company, then a plain close with
     no "I look forward to the opportunity to discuss how my skills".
4. Write in my voice (see HOW I TALK). Contractions are fine. Plain words.
5. Never use these: passionate, proven track record, dynamic, fast-paced,
   results-driven, team player, synergy, leverage, unique opportunity,
   thrilled, excited to apply, I am confident that, in today's.
6. For each gap listed, do not mention it unless a truthful bridge exists in
   the fact sheet. If none, say nothing about it.
7. After the letter, list every factual claim in it and the line of the fact
   sheet that supports it.

JOB DESCRIPTION:
[paste]

FACT SHEET:
[paste]

Rule 7 is the cheap insurance. A model that must name the supporting line for each claim will often expose its own inventions, because it has nothing to point at. It's not perfect. Treat the list as a prompt to check, not a guarantee.

Step 3: The edit pass catches what the draft prompt misses

Run this in the same chat. It turns the model into a critic of its own draft, and critics are easier to keep honest than authors.

Now audit the letter you wrote.

1. Quote every sentence that could appear unchanged in a letter to a different
   company. Suggest a specific replacement from the fact sheet, or delete it.
2. Quote every claim that goes beyond the fact sheet (a stronger verb, a
   bigger number, an implied responsibility). Show the fact sheet line next
   to it.
3. Find sentences that start with "I" three times in a row and merge them.
4. List any phrase from the banned list or any similar corporate filler.
5. Return the corrected letter. Keep [NEED: ...] markers in place.

Point 2 is where "led" quietly replaces "contributed to" and "improved performance" replaces "reduced a page load from 4.1 to 2.8 seconds." If your fact sheet says "contributed," the letter says "contributed."

Which phrases give a letter away?

You don't need a detector to find AI-sounding text. These are the patterns that make a reader's eyes slide off the page. Ban them in the prompt and search for them in your final draft.

One more tell is rhythm: every sentence about the same length, each paragraph ending with a neat, quotable line. Read your draft aloud. If you'd never say a sentence to the hiring manager across a table, rewrite it.

What does a before and after look like?

This example is illustrative, not captured output. The applicant is invented.

Generic draft:

I am writing to express my keen interest in the Support Analyst role. With a proven track record of delivering exceptional customer service, I am confident my skills align perfectly with your needs.

Draft from a fact sheet (where the fact sheet said: handled about 40 tickets a day on a billing queue; wrote a macro for refund-status replies; first-response time went from 6 hours to under 2):

On my current billing queue I handle around 40 tickets a day, and the one thing I'm proudest of is a refund-status macro I wrote: it took our first response time from about six hours to under two. Your posting mentions reducing repeat contacts, which is the problem that macro was built for.

The second version is shorter, specific, and checkable. It also took about two minutes more of input. That's the trade.

How do I handle a requirement I don't meet?

Say the posting wants Salesforce experience and you've used HubSpot. Wrong moves: claim Salesforce, or ignore it and hope. Better is a single honest sentence naming the nearest real thing. Add a gap line to the fact sheet and use this follow-up:

The posting asks for [REQUIREMENT]. My nearest real experience is: [what you
actually have]. Write one sentence, in my voice, that is honest about the
difference and shows how I'd close it. If you can't do that without
exaggerating, say so and suggest leaving it out.

The last sentence matters. You want the model to be allowed to say "leave it out." Without that permission, it tends to produce something.

For the interview that follows, rehearse the gap questions out loud. The mock interviewer prompt is built for that, and the same "ask about the gap" rule works there too.

Will recruiters actually mind AI?

It depends who you ask, and the evidence is thin. The most-quoted figure is from a CV Genius poll of 625 UK hiring managers in 2024, which reported 80 percent expressing distrust toward AI-generated content and 74 percent saying they can spot it (summary at Digit). That's one survey from a company that sells CV services, in one country. A TopResume survey reported a lower figure, about one in five recruiters who would reject a candidate for an AI-generated CV or cover letter (TopResume). I haven't audited either method.

Read both the same way: nobody is defending the generic letter. What gets penalised is sameness and claims that don't hold up, which your fact sheet and edit pass address. Also check each employer's application page. Some say whether AI use should be declared, and you should follow what they say.

Before you send: a last read

• Every number and tool name is in my fact sheet or my CV.
• The first two sentences contain my best evidence, not a greeting.
• Nothing in it could go to a different company unchanged.
• The company-specific line is something I really know.
• No banned phrases, no "[NEED: ...]" left over.
• I could talk for two minutes about any sentence in it.
• Spelling of the company name, the role title, and the recipient's name is right.

That sixth line is the real test. If you couldn't defend a sentence in an interview, it shouldn't be in the letter.

Privacy and what not to paste

Don't paste your phone number, home address, ID numbers or referee details into a chatbot to get a cover letter, because the model doesn't need them. Leave out confidential figures from a current employer. Check your tool's data-controls setting: some products use conversations for training unless you opt out, and the setting differs by product and plan.

Once you've got a letter, your LinkedIn profile should tell the same story. The LinkedIn content prompts post covers that, and the prompt library has other reusable templates when you want them.