I Had ChatGPT Rewrite the Same Resume Bullet 10 Times. It Invented a Different Job Title Every Time.

Short answer: no, not by itself. Letting a raw AI chatbot write your whole resume from scratch — with no fact-check step — is the resume equivalent of letting a very confident stranger describe your work history from memory. Sometimes it nails it. Sometimes it hands you a brand-new job title you never held. I know because I watched it happen in real time, on my own resume, ten times in a row.

Here’s the experiment. I took one bullet from my own resume — a plain, boring line about managing a small team and a budget — and asked ChatGPT to “rewrite this resume bullet to sound stronger,” ten separate times, fresh conversation each time, zero other changes. I didn’t ask it to touch my title. I didn’t mention my title at all in the prompt.

By round four, my job title had changed anyway. “Coordinated a team of five” became “Led a team of five” became, two prompts later, “As Team Lead, directed a five-person department.” I never said “Team Lead.” I’ve never held that title. It just… appeared, fully formed, presented with the same confident tone as everything else in the bullet.

The pattern, run by run: Round 1 kept my actual title. Round 3 upgraded “coordinated” to “managed,” which is a stretch but arguably defensible. Round 4 invented “Team Lead” out of nowhere. Round 7 called me a “Department Head.” Round 9 walked it back to something closer to the original, then round 10 introduced a completely different embellishment — a specific dollar figure for the budget that I never gave it. Ten runs, same input, five materially different versions of my professional identity.

Why this isn’t just bad luck. This is what the model does when it’s filling a gap. Resume bullets are short on specifics, and a language model trained to sound polished and complete will reach for a plausible-sounding word rather than leave a hole — and a job title is exactly the kind of small, atomic, easily-swapped detail it reaches for. That’s not a resume-specific quirk. It’s a documented property of how these models behave on simple factual questions in general. OpenAI’s own benchmark for short-form factuality found that GPT-4o answered only 38.2% of simple, single-answer factual questions correctly — the same category of fact as a title, a date, or a name. Most of the time, on the easiest kind of question there is, it’s wrong.

And when it doesn’t know, it doesn’t say so. The more damning finding isn’t that models get things wrong — it’s what they do instead of admitting it. A large-scale benchmark testing 36 frontier models on hard, cross-domain knowledge questions found that 33 of the 36 were more likely to confidently make something up than to actually answer correctly when they didn’t actually have the answer. Translate that to your resume: the model doesn’t know what your actual title was. It wasn’t in your prompt. So instead of asking, it picks one that sounds right for the sentence and moves on — with exactly the same confident tone it uses for the parts it got right. There’s no visual cue, no hedge, no asterisk. It reads identically to the truth.

This has already happened to someone else, in public, with worse stakes. This isn’t a hypothetical failure mode I’m speculating about. A Fortune reporter ran almost this exact experiment in 2023 — asked ChatGPT to write her resume — and it came back claiming she’d been named to Forbes’ 30 Under 30 in Journalism in 2020 and had won the George Polk Award for Business Reporting in 2018. Neither was true. Those aren’t vague embellishments like mine — they’re specific, checkable, prestigious credentials that a hiring manager could verify with one search and immediately flag as a fabrication on an application.

Zoom out: employers have already noticed. This isn’t just a candidate-side annoyance anymore — it’s becoming a hiring-side problem with a name. A Robert Half survey of more than 2,000 U.S. hiring managers found that companies are now seeing generative AI tools 「fabricating or embellishing work history and skills, making it harder for employers to distinguish authentic experience from AI-generated content」, with 65% of hiring managers reporting it’s gotten harder to verify what candidates actually did. In other words: the invented title isn’t just a risk to you. It’s part of why the whole applicant pool is getting side-eyed right now.

Yes, but — the fix isn’t “never use AI.” AI is genuinely good at word choice, structure, and cutting a bloated bullet down to something a recruiter will actually read. The problem isn’t the writing. It’s the writing happening with nothing checking it against what’s actually true about you. A chatbot with no memory of your source material has to guess at gaps — and as we just saw, its guesses look exactly as confident as its facts.

Bottom line: if you’re going to use AI on your resume, the tool matters less than whether something is checking its work. A raw chat window that generates and forgets, run after run, is going to drift — sometimes in your favor, sometimes into a job title you’d have to explain in an interview you didn’t get invited to. Bloom’s tailoring tool runs a separate verification pass on every generated bullet, checking it back against your actual source resume before you ever see it as a finished draft — so a fabricated title or invented credential gets flagged instead of quietly shipped. That’s the difference between “AI helped me write this” and “resumes you can defend.”

If you want to see what AI-generated fabrication looks like at scale — not one experiment, but a full breakdown — we ran a deeper version of this test on our own pipeline in the resume hallucination report. And if you’re trying to figure out where the honest line is for using AI at all, this guide walks through it.

FAQ

Q: Is it safe to let ChatGPT write my whole resume? Not without checking it line by line against your actual work history. As the experiment above shows, a raw chatbot can invent a job title, a number, or a credential with the same confident tone it uses for accurate details, and there’s nothing in the output that flags which is which.

Q: Why does ChatGPT make up job titles specifically? Because a title is a small, specific detail the model doesn’t actually have unless you gave it explicitly, and research on how these models behave shows they tend to fill gaps with a plausible-sounding guess rather than flag the uncertainty — the same behavior documented broadly in benchmarks like SimpleQA and Omniscience.

Q: Has this actually happened to someone, or is it just a risk? It’s happened publicly. A Fortune reporter’s ChatGPT-written resume claimed she’d won an award and a named recognition that she never actually received.

Q: Do employers actually catch AI-fabricated resume content? More often than they used to, and it’s actively slowing down hiring. Robert Half’s survey found 65% of hiring managers say it’s gotten harder to verify what’s real in an application.

Q: What should I do instead of pasting a prompt into ChatGPT and using whatever it gives me? Use AI for phrasing and structure, but check every fact — titles, dates, tools, numbers — against your actual resume before you send it. Better yet, use a tool that runs that check for you automatically instead of relying on your own read-through to catch it.

Want a version of AI resume help that checks its own work? Try Bloom and see what a verified tailor pass looks like next to your original resume.