How to use ChatGPT for your resume (without sounding like ChatGPT)

Recruiters and modern ATS rankers can both spot a raw ChatGPT resume in under five seconds. The tell isn't any one phrase — it's the texture. Sentences are uniformly polished. Verbs are slightly too formal. Every bullet starts with one of the same eight verbs. The summary uses the word "passionate" or "dynamic" or "results-driven." Every accomplishment has a number that ends in 0 or 5.

The model didn't do anything wrong. It produced exactly what it was asked for: a resume. The problem is that "a resume" is the average of every resume on the internet, and the average is terrible. To get a useful output you have to ask for something more specific than "write me a resume."

• Never paste your old resume and ask ChatGPT to "improve" it.The output will be the average improvement, which means generic verb swaps and the addition of phrases the model thinks sound professional. Your resume is now worse and you've lost the version control.
• Never let ChatGPT invent metrics. If you give it a vague bullet, it will fabricate a number. Pleasant in the moment, disqualifying in a reference check. Always supply your own numbers.
• Never ship the first draft. The first draft is the average. Your job is to make it specific. That edit is the entire workflow.
• Never ask it to write the summary cold. The summary is the highest-signal slot on the page. The model writes it generically by default. Write it yourself, then ask the model to tighten it.

Forget "write me a resume." The prompt that works is a three-part instruction with a concrete role, a concrete seniority, and a clear taste constraint:

Three options. Specific seniority. Banned-word list. Placeholder for the number you supply. Every output that prompt produces will be 5–10x stronger than "rewrite this bullet."

The right way to use ChatGPT on a resume is one bullet at a time, not one document at a time. The loop:

1. Open your existing resume. Pick the weakest bullet.
2. Run the prompt above on just that bullet.
3. Pick the option that's closest to your voice. Edit one or two words to make it more obviously yours.
4. Fill in the real number.
5. Move to the next bullet.

Twenty bullets at fifteen seconds of generation plus ninety seconds of editing each is roughly thirty-five minutes for the whole document. That's the same time it would take to do a single bad full rewrite, and the result is a resume nobody can flag as AI-generated because each line was edited by hand on top of an AI suggestion.

• Replace any sentence that contains "passionate," "dynamic," or "results-driven" with one that contains a specific noun.
• Cut every "leveraged" and "utilized." They are the AI tells the model loves and humans never use.
• Vary verb starters across bullets. If three in a row start with "Built," one of them gets rewritten.
• Use one number that doesn't end in 0 or 5 per role. Real metrics are messy; round numbers are AI metrics.
• Add one sentence with a slightly informal word ("rewrote," "ripped out," "handed off") per role. Models avoid these. Humans use them constantly.

None of these are stylistic preferences. Each one is a signal modern LLM rankers and human recruiters use to distinguish a real document from an AI-generated one. The combined effect of all five is a resume that reads as written by a real person who used AI as a tool — which is exactly what it is.