ChatGPT Resume Review: 7 Prompts That Get Honest Feedback

The short answer: "review my resume" gets you a compliment sandwich. To get a real critique out of ChatGPT (or Claude, or Gemini), give it three things: a role to play (a tired recruiter, not a helpful assistant), a rubric to grade against, and an instruction to quote your exact lines as evidence. Ask for the diagnosis first and the rewrite second, and check every number in anything it rewrites against what you actually did. The seven prompts below do each of those jobs, and you can paste them as they are.

Why the default answer is too nice

A general assistant is built to be helpful and agreeable. Ask it an open question about something you made and it tends to open with what's working, soften what isn't, and close with encouragement. That's pleasant to read and not much use the night before you send fifty applications.

It isn't that the model can't be critical. It can. It just needs to be told what "critical" means here: who is reading, what they're looking for, and how strict to be. That's what a prompt is for. Most of the difference between a useless AI review and a useful one is in the question.

Before you paste anything

Three things first.

Take out your phone number and street address. Nothing in a resume review needs them. Leave your name off too if you like. The feedback won't change.

Check your data settings. ChatGPT has data controls in its settings that decide whether your conversations can be used to improve its models. Other assistants have their own version. Look once, set it how you want it, and move on.

Have a real job posting ready. A resume is only good or bad relative to a job. Two of the prompts below need a posting, and the rest get sharper if you paste one.

Then paste your resume as plain text. If you upload a PDF with columns or text boxes, ask it to repeat back your most recent job title and dates before you go further. If it gets those wrong, it's reviewing a scrambled copy.

The seven prompts

Use them in one conversation, in order, or pick the two or three that match what you're worried about.

1. The blunt grade

This is the one to start with. It replaces "what do you think?" with a fixed scale.

Act as a recruiter who has read 200 resumes today and is looking for reasons to say no. Grade my resume from A to F on each of these: (1) clarity of what I do and at what level, (2) results with numbers versus duties, (3) relevance to the role I name below, (4) filler and buzzwords, (5) formatting a tired person can scan. Give one letter grade per category and one overall. For every grade below B, quote the exact line from my resume that cost me the grade. Do not rewrite anything yet. Do not open with praise. Target role: [role]. Resume: [paste]

The two instructions doing the work are "quote the exact line" and "do not rewrite anything yet." Quoting forces it to point at evidence. Holding the rewrite keeps it from skipping the diagnosis and handing you a prettier version of the same problems.

2. The six-second read

You have six seconds with this resume, the way a recruiter does on a first pass. Read only what a fast scan would catch: my name, current title, current company, dates, previous title, and education. Tell me what job you think I do, what level you think I am, and what you'd guess I'm applying for. Then tell me which of my three strongest accomplishments you would NOT have seen in those six seconds, and where on the page they are.

The order in that prompt comes from Ladders' eye-tracking research on how recruiters scan a resume, which we walk through in the 6-second resume test. A chat can only describe that scan in words. If you'd rather watch it, the 6-Second Replay plays a simulated recruiter's first pass across your own resume and marks the lines it never reaches.

3. The numbers audit

Go through every bullet point on my resume. Sort them into three lists: (A) bullets with a specific number or measurable result, (B) bullets that describe a result but give no number, (C) bullets that only describe a duty. Give me the count for each list. For list B, tell me what number is probably missing and ask me for it. Do not make numbers up.

"Do not make numbers up" matters more than it looks. Left alone, an assistant asked to improve a bullet will often add a tidy percentage you never gave it, because a bullet with a number reads better than one without. We wrote about that habit in what happens when ChatGPT writes the resume.

4. The buzzword sweep

List every phrase on my resume that could appear unchanged on anyone else's resume in my field: empty adjectives, clichés, inflated verbs, and claims with no evidence behind them. For each one, quote it, say in one line what it's standing in for, and ask me the question whose answer should replace it.

The last clause is the useful part. "Results-driven" is standing in for a result. The fix isn't a synonym. It's the result, and only you have it.

5. The match against a real posting

Here is a job posting and my resume. List the requirements and keywords in the posting, in order of importance. For each one, tell me whether my resume shows it clearly, shows it weakly, or doesn't show it at all, and quote the line you're relying on. Then list the five changes that would close the most gap, using only experience that is already on my resume. Do not invent experience I haven't listed. Posting: [paste]. Resume: [paste]

This is the single most useful thing to do with any resume before sending it. It's also what our free Job Match tool does in one step: a 0-100 fit score and the keywords the posting asks for that your resume doesn't show.

6. The level read

Ignore my job titles. Based only on the evidence in my bullet points (scope, ownership, size of what I was responsible for, who I influenced), what seniority level does this resume read as: entry, junior, mid, senior, or lead? Quote the three lines that most support your answer and the three that pull it down. Then tell me whether that matches the titles I've listed.

A senior title sitting over mid-level bullets is hard to see in your own resume, because you know what the work really involved. The reader only has the bullets. Level Check runs this same read if you want a second opinion on it.

7. The interview test

You are the hiring manager for [role]. Based on this resume, write the six questions you would most want to ask me, and for each one, name the line on my resume that made you want to ask it. Flag any claim you would expect me to struggle to back up in person.

This one is quietly the best lie detector for your own resume. If a question makes your stomach drop, the line that triggered it is either inflated or underexplained. Fix it now, not in the room.

Then, and only then, ask for a rewrite

Once you've read the diagnosis and argued with it, this is the prompt for the fix:

Rewrite only the bullets I list below. Keep every fact exactly as I gave it. Where a number would help and I haven't given one, put [NUMBER?] in brackets so I can fill it in. Plain verbs, no buzzwords, one line each. Bullets: [paste]

The brackets are the point. They keep the holes visible, so you fill them with the truth and nothing gets filled in for you.

How to tell if the answer is any good

Three checks, about two minutes.

Did it quote you? A review that never quotes your resume could have been written without reading it. If the feedback is all general advice ("consider adding metrics"), ask again and insist on quoted lines.

Does it hold up twice? Paste the same prompt into a fresh conversation. If the grades move by more than a notch, treat both as rough opinions, not measurements. This is the real limit of a chat: there's no fixed scale behind it, so before-and-after comparisons are shaky.

Is everything in the rewrite true? Read every number and every skill in the new version against what you actually did. Anything you couldn't explain in an interview comes out.

Where a chat stops and a tool starts

We'll be straightforward about our stake in this: we run a resume review tool, and it's built on a large language model too (Anthropic's Claude). The difference isn't a secret model. It's that the prompts above are already written, the rubric is the same for every resume, and the grade is worked out from that rubric the same way each time, so a B means the same thing on Tuesday as it did on Monday.

So here's the honest split. A chat is better when you want to go back and forth: argue with a suggestion, try six versions of a bullet, get the cover letter done in the same sitting. A purpose-built tool is better for the first step, the blunt verdict you can measure again later. The free roast gives you a letter grade, a Buzzword Density Score and the specific lines behind both in about 30 seconds, with no prompt to write and no account. We laid out the full comparison, including where ChatGPT wins, on the RoastMyResume vs ChatGPT page.

Most people would do well to use both, in that order: get the verdict, take the specific problems into a chat to work on the wording, then check the grade again.