AI Resume Builder and Portfolio Maker
Explore templates and pricing
Two studies tracked what happened when an AI tool started writing cover letters. Tailored text stopped standing out, and edited drafts won more jobs.
Canonical: https://www.prezumi.com/blog/ai-cover-letter-what-works
Using AI to draft a cover letter is fine, and on one large job platform it raised the number of replies applicants got. The same research found something less comfortable: once everyone could produce a well-tailored letter in seconds, a well-tailored letter stopped telling employers much. The applicants who did better were the ones who spent time editing the draft.
So the useful question is what to add after the draft exists. This guide goes through the two studies, then through the parts of a letter that an AI tool cannot write for you.
Both studies use data from Freelancer.com, a marketplace where freelancers send a short written proposal for each job. In April 2023 the platform added an AI tool that writes that proposal from the job description and the freelancer's profile. Employers could not see which proposals the tool had written.
| Study | What they found |
|---|---|
| Cui, Dias and Ye, "Signaling in the Age of AI: Evidence from Cover Letters" (2025) | Access to the tool made letters match the job post more closely and raised callback rates. Time spent editing the AI draft went with more hiring success. After the tool arrived, the link between a closely matched letter and a callback fell by 51%, and employers leaned more on work history. |
| Galdin and Silbert, "Making Talk Cheap: Generative AI and Labor Market Signaling" (Dartmouth Tuck and Princeton) | Before the tool, a customised proposal went with getting hired and with higher pay. Afterwards employers were willing to pay far less for it, and it no longer predicted whether the job would be completed well. They relied more on price and past ratings. In the authors' model, the least able fifth of workers were hired 14% more often and the most able fifth 19% less often. |
What this does and does not show. These are short proposals for freelance work on one platform. A cover letter for a salaried job is read by different people for different reasons. The studies show a direction: when good writing becomes free, employers trust it less and look for evidence they can check.
A letter that repeats the job posting back to the employer used to prove you had read it and could write. An AI tool now does that for every applicant, so it no longer sets you apart. Three things still do. The research points at the first two, and the third follows from how these tools work.
Resume Genius surveyed 625 hiring managers: 83% said they read cover letters always or frequently, and 45% read the letter before the resume. They read fast, though. About a third spend under 30 seconds on it, and almost half spend between 30 seconds and two minutes.
That reading time is the real constraint. In the same survey, 41% said the introduction leaves the biggest impression. If the first two sentences could have been sent to any company, the reader has little reason to keep going.
| Task | How the draft does | Why |
|---|---|---|
| Structure and length | Well | It produces three or four paragraphs on one page without being asked twice. |
| Matching your experience to the posting | Well, if it has your resume | It can find which of your roles answers which requirement faster than you can. |
| Grammar and tone | Well | The text is clean. In a second language it is often cleaner than a first attempt by hand. |
| Facts about you | Badly without your resume | Given only the job post, it invents experience to fit the requirements. |
| Why this company | Badly | It writes admiration that would fit any employer. |
| Your voice | Badly | Phrases like "I am writing to express my interest" and "I am confident that my skills" turn up in many drafts. |
This is the pass we would do on any AI draft before sending it. It takes about 20 minutes, and it covers the parts the studies say still count.
Before: I am writing to express my strong interest in the Customer Success Manager position. I am confident that my skills and experience make me an ideal candidate. After: Last year I took over 40 accounts that were due to renew within 90 days and kept 37 of them. Your posting describes the same problem at a larger scale.
Employers can sometimes tell, but they cannot be sure. No tool identifies AI text with certainty, and a letter written by hand can be flagged by mistake. What readers notice is sameness: the same opening and the same adjectives in one application after another.
If you edited the draft and every fact in it is true, there is little to worry about. A letter that claims experience you do not have is a different matter, because the claim fails in the interview whoever wrote the sentence. Our guide to ChatGPT resume writing lists the sentences that give an unedited draft away.
Prezumi's free cover letter generator takes two inputs: your resume as a PDF and the job description. It writes only from what is on the resume. Where the posting asks for something the resume does not show, the tool leaves it out of the letter and lists it separately, so you can decide whether it is true for you.
That fits what the research found. A draft built from real work history gives you checkable facts to start from, and the list of gaps tells you where your 20 minutes of editing should go. You get one letter a day without an account.
If the list of gaps is long, the problem may sit in the resume. The free ATS resume checker compares your resume with the same job description and shows which keywords are missing, and the guide to what to write when you don't meet every requirement covers how to handle the gaps that are real. If you are applying in another country, cover letters by country shows the layout each one expects.
Yes, as a draft. In a study of an AI writing tool on Freelancer.com, access to the tool raised callback rates, and more time spent editing the draft was linked to more hiring success. Check every fact against your resume and rewrite the opening before you send it.
Not reliably. Detection tools make mistakes in both directions. Readers do notice letters that open the same way and use the same phrases as many others, and they notice claims that fall apart in an interview. An edited draft with true, specific facts is not a problem.
Yes, though what makes a letter work has changed. Two studies found that a closely tailored letter became a weaker signal once an AI tool could produce one for anyone, and that employers turned to work history instead. A letter that carries facts the reader can check still helps. In a Resume Genius survey, 83% of 625 hiring managers said they read cover letters always or frequently.
Remove any claim your resume does not support, rewrite the first two sentences around your most relevant result, add one sentence about this company that could not be sent to another, and cut any paragraph that only repeats your job titles.
It can, if the tool only sees the job description. It then writes whatever fits the requirements. A tool that reads your resume and writes only from it avoids this. Prezumi's generator works that way and lists the requirements it could not find on your resume.
About 20 minutes is enough for a one-page letter: five to check the facts, five on the opening, five on a sentence about the company, and the rest on cutting repetition and reading it aloud.
Zubair Baqai is an AI engineer and the founder of Prezumi. He started it after researching what gets candidates interviews at companies like Google, Meta, Microsoft and Amazon. He writes Prezumi's guides on resumes, ATS screening and CV formats. About Prezumi