Stop asking ChatGPT to write your resume. Do this instead.

I read maybe four hundred resumes a year. Since GPT-4 came out, they've started to blend together in a way that's almost funny.

The verbs are the same. Spearheaded. Orchestrated. Drove. Elevated. Leveraged. The structure is the same: a STAR-method bullet that mentions the framework you used, the metric you moved, and the team size you led. The opening paragraph is the same: "Experienced engineer with 7+ years of expertise in scalable systems and a passion for clean, maintainable code."

You can spot an AI-polished resume in about eight seconds. We've all started spotting them. Recruiters faster than anyone.

This isn't a moral problem. The problem is that you're spending a polished resume against thousands of other polished resumes, and there's no signal in any of them. None of you stand out. The recruiter is reading her seventy-third "drove a 32% increase in throughput by leveraging modern serverless patterns" of the week. Her eyes are not on your sentence.

AI-generated phrase	What it actually means to a recruiter	What to write instead
"Leveraged AI-powered tools"	You used ChatGPT. Everyone does.	Name the tool + what you shipped with it + how long it took.
"Passionate about cutting-edge tech"	Cliché. Skipped.	Don't write this. Skip the trait, describe the work.
"Drove cross-functional alignment"	You attended meetings.	Who you talked to, what you agreed on, what shipped because of it.
"Spearheaded a strategic initiative"	You did some of the work.	What was the problem, what was your part, what was the outcome (with numbers).
"Results-oriented professional"	Filler.	Show one result with a number near the top of the resume.

What recruiters actually grep for

I asked a few recruiter friends what they actually scan a resume for. The answers were depressing in their unanimity.

Names of specific projects, products, or systems they recognise.

Specific tools and frameworks, used as recently as possible. The version matters.

Numbers that feel real. Not "increased efficiency by 40%". Numbers that come with the unit attached and the cost in tension with the benefit. "Cut p99 latency on the checkout endpoint from 1.2s to 280ms across 4 services, while preserving the existing retry semantics." That sentence does not come out of any LLM I've ever used, because it has actual content.

Anything that suggests you've shipped something to real users. Side projects with a URL. A talk you gave. A bug report someone else closed because you filed it well.

Recruiters do not grep for the third synonym for "drove". They are exhausted by it.

The thing nobody is doing

Write your resume yourself. Badly. Just type it out, in plain language, the way you'd describe the work to your friend who works in a totally different field. Don't try to make it sound impressive. Try to make it clear.

Once you have the bad draft, do one of three things:

You can ask the model to clean up the language while keeping the content. Be aggressive about rejecting changes that flatten your voice. The model's instinct is to make every sentence look like every other sentence. Your instinct should be the opposite.

The actual rule

A resume's job is not to summarize you. Its job is to earn an interview. Specifics win interviews. Adjectives don't.

You can ask the model to point out the bullets that are too vague. Then go back and add the specifics. The actual project name. The actual technology. The actual person who was upset before, who is no longer upset.

You can ignore the model entirely and just send the badly-written, clearly-written, specific resume. This will work better than you think. The bar for "interesting human" has dropped because the median resume is now sand-blasted into perfect anonymity.

A practical four-step routine

If you want a process: this is what I'd do for a hiring search this month.

Step one. Write down the five projects you actually shipped or rescued in the last two years. Just the facts. Two sentences each. No marketing language.

Step two. For each project, write down: what would have happened if you hadn't been there? Be honest. Some of them, the answer is "nothing would have changed; the team would have done it without me." Cut those. Keep the ones where the answer is something specific you uniquely did.

Step three. Now write the bullets. One project, one to three lines each. Specific verbs. Specific numbers. Specific names where you can use them without breaking NDA. Read each bullet out loud. If it could appear on someone else's resume word-for-word, rewrite it.

Step four. Send it to a friend who knows you and ask: "does this sound like me?" If they say no, rewrite again. If they say yes, you're done.

Closing

The bar to look impressive used to be high and is now low. The bar to look like a specific person used to be low and is now high. That's the swap nobody told you was coming.

The good news: a resume that sounds like a real human, with specific stories and a slightly imperfect voice, is now genuinely rare. It will stand out without you trying. You don't need to be more polished. You need to be more yourself.