Best AI for a Cover Letter? What the Evidence Actually Says

About 29.3% of job seekers now use AI to write or customise a resume or cover letter, up from 17.3% a year earlier, according to iHire's 2025 State of Online Recruiting Report. Employers noticed. The honest answer to which is the best AI for a cover letter starts somewhere stranger than a tool roundup, though. It starts with a study of 5.5 million job applications, which found that the advantage AI hands you fades the moment everyone else has it too.

This piece answers the two questions people actually ask, in the order they ask them. First, whether using AI on a cover letter will quietly cost you the job. Then, which model to open once you have decided it will not. Along the way it flags something about the comforting statistics that circulate on this topic, because several of the most widely repeated ones fall apart when you follow them back to their sources.

The Key Takeaways

• Nobody has actually measured detection: 88% of hiring managers say they can spot AI in an application, but that is a self-reported belief. The one real test, run in March 2023, gave hiring managers a mix of human and ChatGPT-written intros. Only 18% sorted all three correctly.
• Editing is the part that works: across 5.5 million applications, a one standard deviation increase in time spent editing the AI draft was associated with a 0.31 percentage point rise in offer probability.
• The advantage decays: once an AI cover letter tool reached everyone on the platform studied, the link between a tailored letter and a callback fell by 51%. Employers shifted to judging prior work history instead.
• ChatGPT gave the safest draft in our resume testing, staying closest to the source. None of the three invented a statistic, but Gemini wrote around nine job-advert requirements back in as the candidate's experience. That is the failure worth checking for.
• Check the numbers you get quoted: the widely repeated "52% detection accuracy" figure is a misreading of an unrelated statistic, and the "63% find it acceptable" figure does not appear in the survey it is credited to.

Best AI for a Cover Letter: The Short Answer

Draft it in ChatGPT if you want the safest starting point, and in Claude if you want the strongest structure and are willing to edit harder. Then spend real time on that editing, because that is the part with evidence behind it.

That ranking comes from the testing behind our companion piece on the best AI for resume writing, where one resume and one job advert went through all three models. The result cut against expectations. No model invented a statistic and none inflated the candidate's seniority. What they did instead was quieter: they wrote requirements from the job advert back into the document as though the candidate had already done them. Gemini did this around nine times over. ChatGPT did it once, mildly, and stayed almost entirely inside the source. Claude sat in between, bracketing every number as a placeholder rather than inventing one, while still asserting that metrics the candidate never mentioned existed at all.

Two honest caveats before the table. That test was run on a resume, not a cover letter, and a letter is a piece of persuasive writing rather than a structured list, so the failure modes are not identical. And the most thorough published test of cover letter tools, Jobscan's July 2026 review of ten products, deliberately refused to score the underlying models at all. Its testers reported that a newer model and an older, cheaper one "produced drafts we could not tell apart." Treat model choice as a tiebreaker. What you do after the draft appears matters more.

What the Research Actually Found

The most useful study on this subject is not a vendor survey. It is Signaling in the Age of AI: Evidence from Cover Letters, by Jingyi Cui, Gabriel Dias and Justin Ye. It is a preprint, revised in November 2025, and it tracks what happened in 2023 when a large freelancing platform introduced an AI cover letter writing tool. The dataset covers 5,499,707 bids across 106,714 jobs from 264,082 workers, and because the tool rolled out to some users before others, the authors could measure cause rather than correlation.

The advantage was real, and then it faded

Getting access to the tool raised a worker's chance of a callback by 0.43 percentage points against a baseline callback rate of 7.02%, which is a gain of roughly 6%. Among people who actually used it, the effect was much larger. Two caveats the authors state plainly: that callback result clears significance at the 10% level rather than the 5% level, and the effect tapered off after about two months. So AI help worked, for a while.

Then the interesting part. After the tool became widely available, the correlation between how well a cover letter matched the job post and whether it earned a callback fell by 51%. When everyone can produce a perfectly tailored letter in seconds, a perfectly tailored letter stops proving anything. Employers responded by leaning on signals AI cannot manufacture, chiefly a candidate's prior work history. Your letter is being read by people who have already adjusted to the fact that it was probably written with help.

The part that still works

The same paper found that time spent editing the AI-generated draft was positively associated with hiring success. A one standard deviation increase in editing time, about three and a half minutes, came with a 0.31 percentage point rise in the probability of winning the job. That is around 52% of the average win rate among AI-assisted applications. Workers who were stronger writers to begin with edited more. This one is a correlation rather than a causal estimate, so read it as a strong hint and not a guarantee.

That is the single most actionable finding available on this topic. No cover letter vendor has an incentive to lead with it, because it locates the value in your effort rather than in their generator.

Can Employers Tell You Used AI?

Most of them believe they can. Insight Global's AI in Hiring survey put the question to 1,005 hiring managers at organisations with 100 or more employees. 88% said they can tell when a candidate used AI on an application, a cover letter or a resume.

Read that carefully, because the framing does a lot of work. It records a belief, not a score. The survey never tested whether those managers were right. The fieldwork also ran between 17 and 22 October 2024, despite the report carrying 2025 in its title, so it predates the current generation of models entirely.

The only well-known attempt to test the belief comes from ResumeBuilder.com in March 2023. It showed 1,000 hiring managers three cover letter introductions, a mix of human-written and ChatGPT-written, and asked them to sort one from the other. Just 18% got all three right. That result is three years old, it concerns models that have since been replaced twice over, and the company publishing it sells resume products. It is the best evidence available, which tells you how thin the evidence is.

What about software? Applicant tracking systems parse, file and search your application, and they do not ship AI detection as standard. Jobscan checked the major platforms one by one and found no AI-writing detection feature in Workday, Greenhouse, iCIMS, SAP SuccessFactors, Lever or Oracle Taleo. Dedicated detectors are a separate product, and they are unreliable enough that we have written about how AI detectors actually perform. For an ordinary commercial job, nobody is scanning your cover letter.

Do Recruiters Care?

Some do. In that same Insight Global survey, 54% said they would care if an applicant used AI. But caring about AI use and rejecting for it are different behaviours, and the thing recruiters actually reject is a letter that says nothing specific.

There is also a reason they are short-tempered about it. Robert Half surveyed HR leaders in November 2025 and found 67% said reviewing AI-generated applications had slowed their hiring, with 20% reporting delays of more than two weeks. 84% of HR teams reported feeling overworked as a result, and 65% of hiring managers said the flood of applications had made verifying candidate skills harder. The release specifically calls out AI tools fabricating or embellishing work history and skills.

So the resentment is real, and it is aimed at volume and invention rather than at assistance. A specific, truthful letter that a model helped you write is not what is ruining their week.

Why the Statistics in These Articles Disagree

If you read three articles on this question you will meet three incompatible sets of numbers. We tried to verify the most quoted ones and several did not survive.

The claim that hiring managers detect AI with "52% accuracy, barely better than a coin flip" appears repeatedly. Follow it back and the only 52% in the cited source is a statement that 52% of respondents find using AI for drafting or proofreading acceptable. That is a different claim entirely, and it carries no named source. The reassuring line that "80% of recruiters would not automatically reject an application" almost never arrives with a citation either, though this one does have a real basis. TopResume surveyed 600 recruiters in May 2025 and found that 19.6% would reject a candidate over an AI-generated resume or cover letter, which leaves roughly four in five who would not. The number holds. Its provenance, as usually printed, does not. A "63% of hiring managers find AI-assisted cover letters acceptable" statistic is attributed to a 2026 Resume Genius survey. That company's published cover letter research is a 2023 poll of 625 hiring managers, and it asks nothing about AI at all.

None of this is exotic. Career sites cite each other, a qualifier drops off in transit, and a number arrives at your screen meaning something its authors never said. When a statistic on this topic has no year, no sample size and no link, assume it has been through several hands.

Best AI for Cover Letter Writing, Model by Model

The last row deserves a comment. Paid cover letter generators are mostly wrappers around the same models listed above, with a form in front and a subscription behind. Jobscan's testers flagged auto-renewal billing as a recurring problem in the category. If your budget is zero, the free tiers of Claude, ChatGPT and Gemini do this job, and switching between them to compare drafts is the entire reason a multi-model app like Fello AI exists. Our comparison of Claude and ChatGPT across writing tasks goes deeper on where each one pulls ahead.

How to Use AI for a Cover Letter Properly

Start by pasting in the full job advert, the company's own description of itself, and your real history in whatever rough form you have it. A thin prompt produces exactly the generic output recruiters bin. Then ask for a draft that opens on the employer's problem rather than on your own enthusiasm, and that uses only facts present in what you supplied.

Now do the work the research says pays. Read every sentence against your actual history and delete anything you cannot defend in an interview. Rewrite the opening paragraph entirely in your own words, since that is the part a human reads before deciding whether to continue. Add one detail about this specific employer that no other applicant would have bothered to find. If you want prompt scaffolding to start from, our collection of prompts for personal branding covers the framing.

What not to do is run the finished letter through a humanising tool. Those products exist to defeat detectors that, for ordinary jobs, nobody is running, and they work by degrading your writing into something statistically noisier. We looked at what these tools do to text in our piece on AI humanisers. The goal is a letter that is true and specific, not one that is camouflaged.

The Verdict

Draft it in ChatGPT on the free tier, then spend twenty minutes editing it against your real history, deleting every claim you cannot defend in an interview. That is the whole recommendation, and the second half carries more weight than the first. To find out which claims survive, pressure-test them in a mock interview first.

The question people arrive with, whether AI use will get them caught, turns out to be the wrong worry. Nobody is scanning your letter, most managers who think they can spot AI have never been tested, and the evidence that does exist says assistance raised callback rates. The real risk is subtler and better documented: as everyone gained the same tool, a well-matched letter stopped carrying information, and employers moved on to judging what you have actually done. A model cannot fake that for you. It can only help you describe it, quickly, in language worth reading.

FAQ