Why do ChatGPT prompts to tailor a resume go wrong?
They fail when the model fills a gap with something plausible. The US National Institute of Standards and Technology calls this confabulation in its generative AI risk profile, NIST AI 600-1: "The production of confidently stated but erroneous or false content." On a resume, that false content is usually a number, a tool or a bigger job title.
A resume is a set of claims an employer may check. A reference call, a background check or one interview question can test any line on it. So a tailored resume is only useful if every fact in it is yours.
The prompts on this page all work the same way. They give the model your real facts, tell it to use nothing else, and ask it to flag gaps instead of filling them. None of them makes errors impossible, so the last step is always a check by you.
What should you give ChatGPT before you prompt it?
Give it three things: your full work history, the job posting, and a rule against new facts. A master resume with every role, project and number you can prove is the best source. Paste the posting in full. Then state the rule plainly at the top of every prompt, rather than trusting the model to assume it.
Anthropic's guide to reducing hallucinations lists techniques that apply to any chat model. It says to explicitly give the model "permission to admit uncertainty." It also suggests telling the model to "only use information from provided documents and not its general knowledge." Both ideas are built into the prompts below.
Remove anything you would not want stored, such as your home address, phone number and references. The model does not need them to rewrite a bullet. Add them back in your own copy at the end.
Which ChatGPT prompts help you tailor a resume?
Five prompts cover most of the work: map the requirements, rewrite single bullets, draft a summary, list missing skills, and check the new draft against the old one. Run them in that order, one at a time. Each one below is an example you can copy, followed by the way it most often fails.
Prompt 1: map the posting to your evidence.
Here is a job posting and my resume. List each requirement in the posting. For each one, quote the exact line from my resume that supports it, or write NO EVIDENCE. Use only my resume. Do not rewrite anything yet.
Where it fails: the model may count a loosely related line as evidence, or paraphrase a line and call it a quote. Check that every quote appears word for word in your resume. Anthropic's guide recommends this quote-first step because it grounds the answer "in the actual text."
Prompt 2: rewrite one bullet at a time.
Rewrite this bullet so it uses the job posting's wording where my facts support it. Use only the facts in the bullet. Keep it under 25 words. If a number would help and I haven't given one, write [NUMBER?] instead of guessing.
Where it fails: even with the instruction, a model can add a percentage, a team size or a result you never mentioned. It may also swap your verb for a bigger one, so "helped with" becomes "led." Compare the new bullet with the old one, fact by fact.
Prompt 3: draft a summary from your bullets.
Write a 3-sentence resume summary for this job using only facts from the bullets below. Name the job title from the posting, two skills the posting asks for that my bullets prove, and one result from my bullets. No adjectives about my personality.
Where it fails: the output drifts toward generic wording, such as "results-driven professional with a proven track record." That phrase fits anyone, so it tells a recruiter nothing. The resume summary examples show what a specific summary looks like.
Prompt 4: find missing skills, without adding them.
List the skills and tools in the job posting that my resume does not mention. Do not add them to my resume. For each one, ask me whether I have used it and where.
Where it fails: the model may list a skill you do mention under a different name, or push you to add every term. Add a skill only if you can describe where you used it. The guide to resume keywords covers how to pick terms, and resume keyword stuffing covers when it goes too far.
Prompt 5: check the new draft against the original.
Compare my new resume with my original resume. List every number, job title, date, employer, tool and certification in the new version that does not appear in the original. Change nothing.
Where it fails: a long resume can make the model miss items, so run it one section at a time. Treat its list as a starting point, and still read every line yourself.
Where do ChatGPT resume prompts fail most often?
They fail in five predictable places: invented numbers, inflated scope, generic wording, stuffed keywords and changed dates or titles. Each one is easy to miss on a quick read because the sentence still sounds good. The table lists each failure with a made-up example and a way to catch it before you apply.
| Failure | Made-up example of what it looks like | How to catch it |
|---|---|---|
| Invented number | "Cut onboarding time by 40%" when you never measured it | Circle every number and find where it came from |
| Inflated scope | "Led a team of 8" when you coordinated with 8 people | Ask whether your old manager would agree with the line |
| Generic wording | "Dynamic self-starter who drives results" | Delete any phrase that could describe anyone |
| Stuffed keywords | Five tools from the posting listed in one bullet | Keep only tools you could answer questions about |
| Changed title or date | "Senior analyst" when your title was "Analyst" | Check titles and dates against your offer letters |
If your resume has no numbers to begin with, don't let the model supply them. The guide to a resume with no numbers shows other ways to show results. The page on why AI resume tools invent numbers explains where those figures come from.
Can recruiters tell ChatGPT wrote your resume?
There is no reliable way for a reader to prove it, and the guide to whether recruiters can tell AI wrote your resume covers the evidence. What recruiters can see is a generic line or a claim that falls apart in an interview. Some employers also publish rules on how candidates may use AI.
Anthropic's guidance for candidates asks applicants to write a first draft themselves and use its AI to refine it. It warns that AI-written answers produce "Generic content with experiences you haven't actually had." Amazon's guide to writing a resume with generative AI says these tools should not be "your sole source of truth."
Check the employer's own careers page for an AI policy before you apply. If one exists, follow it. When there is none, the safe rule is the same: every claim should be something you can explain in your own words.
Does using AI on a resume help at all?
Better writing can help, according to one large field experiment. Wiles, Munyikwa and Horton gave nearly half a million jobseekers on an online labour market algorithmic writing assistance, and "Treated jobseekers were hired 8% more often." They found "no evidence that employers were less satisfied" with the people they hired.
That tool was an error checker. The paper says it improved writing "by identifying and providing suggestions to resolve common errors," and treated resumes had fewer grammar errors and redundancies. It did not write new content, so the study supports using AI to clean up your own words. It says nothing about letting a model add claims.
The safest use of ChatGPT for a resume follows that pattern. Let it fix grammar, tighten long bullets and match the posting's wording for skills you have. Keep it away from your facts.
How do you check a tailored resume before you send it?
Read every line and check two things: that it is true, and that you could explain it in an interview. Then compare the numbers, titles and dates with your original resume. Finally, read the posting once more to confirm the strongest evidence for each requirement sits near the top of each role.
A short checklist helps.
- Every number traces back to something you measured or can estimate and explain.
- Every job title and date matches your records.
- Every tool and skill is one you have used.
- No sentence could describe a different person.
- The file is a single-column .docx or text-based PDF, as the guide to how applicant tracking systems work explains.
The guide to how to tailor a resume without lying covers the method in more depth, and common ATS myths covers claims you can ignore. The free resume optimizer rewrites the lines that matter for a job and keeps your facts. Prept also writes tailored resumes only from facts you have confirmed.