Key takeaways
ChatGPT is a capable cover letter generator, and that is the problem with relying on it. In April 2023, Freelancer.com added a one-click tool that writes a cover letter tailored to each job post. Yale economists Jingyi Cui, Gabriel Dias and Justin Ye studied what happened next in their paper "Signaling in the Age of AI". The letters got more tailored, and callbacks rose for the freelancers who used it, though that effect was statistically weak and tapered off after two months. Employers adjusted. After the launch, the correlation between how well a letter matched the post and whether it earned a callback fell by 51%, and the correlation with getting an offer fell by 79%. When anyone can tailor in one click, tailoring tells the reader less. Whatever tool writes the draft, what you still owe the reader is the evidence a model cannot invent for you.
Can ChatGPT write a good cover letter?
ChatGPT writes a good first draft, fast, and it deserves credit for that. Give it a resume and a job description and it will produce a letter that is grammatical, on topic, and shaped around the posting's priorities. For someone staring at a blank page, that is most of the dread gone in under a minute.
It is also good at the jobs around the draft. It can tighten a rambling paragraph, cut a 500-word letter to 250, or rewrite an opening that sounds like a form. It handles tone shifts well, so a letter for a startup and a letter for a bank can come from the same facts.
The free tier is enough for this. ChatGPT Projects, which SlashGear reported in September 2025 had come to free accounts, let you keep your resume and a set of standing instructions in one place (check OpenAI's current plan page, since limits change), so you are not re-uploading the same file for every application.
What ChatGPT cannot do is know things that are not in what you paste. It does not know why you want this job, which project you would lead with in an interview, or what the hiring manager said when you met at a meetup. A letter built only from a resume and a posting is a restatement of the resume in the posting's vocabulary, and that is the letter everyone else can generate too.
What does the research say about AI-written cover letters?
The best evidence comes from the Yale study of Freelancer.com's AI Bid Writer, because it measured real hiring decisions rather than opinions. The tool generated a tailored letter with one click, and freelancers could send it as written or edit it first.
Three findings matter for anyone using ChatGPT the same way. First, the tool worked for its users at the start. Letters matched posts more closely and callbacks rose, though the authors report the callback estimate as statistically weaker than the tailoring one and say the effect tapered off after two months. Second, employers trusted the signal less. The tool wrote a minority of the letters on the platform, yet the tailoring-to-callback link fell by 51%, and employers shifted weight toward harder-to-fake evidence such as a freelancer's past work on the platform. Third, editing went with better results. Most AI letters went out with little to no modification, yet among AI-assisted applications, about three and a half more minutes of editing (one standard deviation) was associated with a gain in the chance of winning the job equal to about 52% of the average offer rate. It is a correlation controlling for the worker, so treat it as a strong hint about editing rather than a proof.
The setting has limits. These are freelancers bidding on projects, not salaried candidates applying through an ATS, so read it as evidence about how readers respond to effortless tailoring, not as a forecast for every job.
Hiring managers report a similar instinct. In TopResume's survey of 600 US hiring managers, run in May 2025, 19.6% said they would reject a candidate with an AI-generated resume or cover letter, and 25% said cover letters should remain AI-free. TopResume sells resume writing services, so treat it as a vendor survey with a stated sample.
How do you get ChatGPT to write a cover letter that is not generic?
Give it the material only you have, and ask it to work in passes.
Start a Project for your search and upload your resume once. In the project instructions, set the rules that apply to every letter: under 300 words, no claims the resume does not support, no adjectives about the company that the posting does not use, and no invented figures about the company's size or results.
For each application, paste the full job description, not a summary. Then add two or three facts the resume does not contain. Why this company. One result you would want to talk about in an interview, with its real number. Anything specific you learned from the team, the product, or the posting itself.
Then ask for an outline before a letter. Which two requirements in the posting does your history answer best, and which story proves each one? Approve or change that outline, then ask for the draft. Last, ask ChatGPT to list every claim in the letter and point to where your resume or your notes support it. Anything it cannot point to comes out.
Our step-by-step guide to writing a cover letter with AI walks through a full worked example of this process.
What does ChatGPT still leave you doing?
Run that workflow honestly and the drafting is done. Here is what remains, every time.
You are the claims check. ChatGPT will list its claims if you ask, but it is grading its own work, and the letter's most common overreach is subtle. A resume that says you "contributed to" an audit becomes a letter where you "led" it. Catching that means reading the letter against the resume line by line.
You are the fact check on the company. Models fill gaps with plausible detail, and a confident sentence about a company initiative that does not exist is worse than no sentence. The failure modes are covered in our piece on when AI cover letters hurt your application.
You are the record keeper. Which version went to which company, with which resume? A chat history holds the answer somewhere, but nothing ties the letter to the application it belongs to.
And you are the second application. The next posting means pasting a new description, re-running the outline, re-checking the claims, and re-reading for invented facts. This is the step the Yale data says people drop, because most AI letters in that study went out with little to no editing, and editing time was the part associated with winning the job. A workflow that depends on you doing it by hand, every time, is a workflow that quietly stops getting done by the tenth application.
ChatGPT or a dedicated cover letter generator?
The drafting is close to a wash. Both run a frontier model over your resume and the posting, and the prose comes out comparable. The difference is what happens around the draft.
| What you need | ChatGPT | Four-Leaf |
|---|---|---|
| Draft quality | Strong | Strong, same class of model |
| Knows your resume | Yes, if you keep it in a Project | Saved once on your account |
| Knows the posting | Paste it into each chat | Paste it once, saved with that application |
| Claims your resume does not back | Only if you ask it to grade itself | Overstated verbs, managing managers and years flagged beside the draft |
| Invented company figures | Forbid them in the prompt, then check | Draft regenerated once if it cites scale the posting and resume lack, then you check |
| Revising | Ask for edits in the chat | Edit in place, or apply a chat edit in one click |
| Letters per application | One chat per company, if you keep it that way | One letter per application, written from the resume for it |
| Your reasons for this company | You add them | You add them |
| Cost | Free tier and paid plans | Free 3-day trial, then $5 for 5 days or $20 a month |
Two rows carry the argument. Claims your resume does not back is the check ChatGPT can only do on itself, and a checker that wrote the sentence is a weak reviewer of it. When Four-Leaf generates the letter, it checks three kinds of claim against your saved resume (a verb stronger than the resume uses, a claim to lead managers or other leaders, and years of experience) and lists each flagged sentence, with the matching resume line where there is one. It never edits the sentence for you, and it does not re-check text you edit afterwards. Letters per application is the record keeping a chat window leaves to your discipline.
One row is deliberately identical. Neither tool knows why you want this job. That sentence is yours in both workflows, and it is the kind of specific a one-click tool cannot hand every other applicant too.
What is overrated
Detection panic. Candidates spend effort trying to make AI letters "undetectable," as if the risk were a scanner. The risk the Yale study measured was a reader who gives a polished, tailored letter less weight, because polish and tailoring got cheap. Rewording a draft to dodge a detector leaves that untouched. Content the model could not have produced on its own is what changes it.
The opposite instinct is overrated too. Skipping the cover letter because AI made it meaningless ignores that the Yale study found editing time still associated with better outcomes. The letter lost value as a signal of effort. It kept value as a place to put evidence.
The playbook
- Keep your resume and standing rules in one ChatGPT Project or one saved account, so each letter starts from the same facts.
- Paste the complete job description for every application. A summary loses the vocabulary the reader wrote.
- Add two or three facts per company that are not on your resume, including at least one real number.
- Ask for an outline before the draft, and pick the two stories yourself.
- Check every claim about you against your resume, and cut any company detail you cannot source. Four-Leaf's cover letter generator flags overstated verbs, claims to lead managers, and years of experience beside the draft for you, which leaves the rest of the read to you.
- Read it out loud once. Anything you would not say to the hiring manager comes out.
Where this is heading
The Freelancer.com data shows a pattern that will repeat wherever a tool makes one part of an application effortless. That part carries less information, and on Freelancer.com employers moved their attention to what was still hard to fake, such as past work. Inside the letter, our read is that the equivalent is specifics. A result with a real number, a reason for this company that could not apply to its competitor, a detail from the posting that shows you read past the first paragraph. Our comparison of AI cover letter generators covers the other tools, and none of them can write that part either. ChatGPT can hand you a tailored letter in seconds. You still owe the reader the two sentences only you could have written.
