If AI-assisted job hunting has a danger zone, it is the cover letter.
Your CV is structured data. Bullet points, dates, keywords. AI-shaped phrasing blends in because everyone’s CV reads a little mechanical. A cover letter is different. It is the one document in the application that is supposed to sound like a person, which makes it the one document where generated text stands out most, and the one recruiters have learnt to smell from the opening line.
The recruitment data covered earlier in this series makes the position clear: detection software is widespread, recruiters have read thousands of identikit AI letters, and the market now penalises anything that reads generated. So Phase 3 of the Evidence-First approach comes with a rule attached. You can use AI to build your cover letter, but the letter does not leave your desk until it passes the detection test. All five checks. Every application.
What a cover letter is actually for
Part of the reason generated letters fail is that people ask AI to write the wrong document. A cover letter is not a prose version of your CV, and it is not 300 words of enthusiasm. It exists to answer three questions the CV cannot: why this role, why this organisation, and what specifically you would bring that the CV’s bullet points do not fully convey.
Every one of those answers has to come from somewhere real. The “why this organisation” comes from your research and the decoded advert. The “what you would bring” comes from your evidence bank. AI’s job, exactly as it was with your CV, is to select and sharpen that material, not to invent it. The no-invention constraint does not relax because the format is prose. If anything it tightens, because a fabricated claim in a cover letter is the one you will be asked about, in your own words, at interview.
The workflow
The inputs are the same three assets you have been building across this series: the decoded advert, your evidence bank, and the tailored CV. The prompt does the assembly:
You are helping me write a cover letter. Use only the material I provide; do not invent, embellish or assume anything that is not stated. Below are the decoded requirements for the role, my evidence bank entries relevant to this application, and my tailored CV.
Write a cover letter of no more than 250 words that: answers why this role and this organisation using only the research notes provided; selects the two strongest pieces of evidence that the CV cannot fully convey and presents them specifically; and sounds like a person writing to a person, in plain British English, no clichés.
If any of the three questions cannot be answered from the material provided, tell me what is missing rather than filling the gap.
That last line matters. A model that reports gaps is doing its job. A model that fills them is writing fiction with your name on it.
The five-check detection test
Before any letter goes out, run it through these five checks.
1. The swap test. Could this letter be sent to a different company by changing the name in the first line? If yes, it fails. A letter that survives the swap test contains detail that only applies to this role at this organisation.
2. The specificity test. Count the verifiable specifics: named systems, numbers, outcomes, dates. A letter with none is a letter the model wrote from atmosphere. Two or three concrete, checkable claims is the floor.
3. The read-aloud test. Read it out loud. Anywhere you would not say the sentence in a phone call, rewrite it. “I am excited by the opportunity to leverage my skills” does not survive being spoken by an actual human. This single check removes most of the phrasing that detection software and tired recruiters flag.
4. The interview test. Every claim in the letter is a question you might be asked. Go line by line: could you talk about this for two minutes without preparation? If not, either it came from your evidence bank and needs rehearsing, or it did not, and it needs deleting.
5. The detector run. Finally, run the letter through an AI-detection tool. Treat the result as a smoke alarm, not a verdict: these tools produce false positives and false negatives, and a pass is not proof of anything. But if a detector flags your letter and checks one to four were marginal, that is your signal to rewrite in your own voice before a recruiter’s tooling reaches the same conclusion.
Notice the order. The detector is last and least. A letter built from real evidence, specific to one organisation, in language you would actually speak, passes checks one to four by construction, and rarely troubles check five. The test is not really about beating detection software. It is about producing the letter the detection software was built to find the absence of.
Before and after
Generated, from a pasted advert: “I am a results-driven professional passionate about delivering innovative data solutions in a dynamic environment.”
Evidence-first, from the bank: “Last year I migrated a 4TB SQL Server estate to Azure for a housing association, cutting their licensing spend by a third; your advert’s emphasis on cost-controlled cloud adoption is exactly that work.”
The first could open any letter to any company, fails four checks out of five, and reads generated because it is. The second survives all five, because every word of it happened.
The line to hold
A cover letter passes the detection test when you could defend every sentence of it, out loud, in the interview it is trying to win. That is the standard, and it is the same standard running through this whole series: AI amplifies your evidence, it does not replace it.
Next in the series: Phase 4, using AI for mock interviews that actually push back, and preparing for the interviews where the first interviewer is not human.
Get the full framework
This post is part of a series drawn from my whitepaper, The Evidence-First Job Search: a complete framework for using AI across job search, applications and interviews, with a 15-prompt library, an evidence bank template, and a 30-day plan. It is free to download.
0 Comments