Can an ATS Actually Tell You Used AI to Write Your Resume?
In June 2025, Greenhouse announced a product called Real Talent. Ask a recruiter what it does and you’ll hear some version of: it stops fake candidates. Ask a job seeker, and most guess wrong, assuming it scans resume prose for ChatGPT phrasing and rejects anything that sounds AI-written. It doesn’t. Neither does anything else currently sold to Greenhouse, Workday, or the other big applicant tracking platforms.
So: do ATS systems detect AI applications the way people assume? Mostly, no. But that answer hides three separate systems that keep getting treated as one, and untangling them changes what’s actually worth worrying about.
Three Different Systems Get Called “the ATS”
Every major applicant tracking system runs AI somewhere in its pipeline, and job seekers tend to lump all of it into one imagined function: a bot reading your resume for proof you didn’t write it yourself. That’s not what any of it does. Jobscan’s own review of six major platforms, Workday, Greenhouse, iCIMS, SAP SuccessFactors, Lever, and Oracle Taleo, states it plainly: the AI inside each one “is designed for candidate matching and resume screening,” and “none of it is built to detect who wrote your bullet points.”
Parsing is the oldest layer and the least controversial one. It extracts your job titles, dates, and skills into structured fields so a recruiter can search and filter. That’s the same engine behind keyword matching, which is a separate, older fight, one we’ve covered directly in our breakdown of how ATS keyword filters actually work. Matching and ranking sit a layer above that, scoring how closely your parsed resume lines up with a job requisition. Neither layer reads for tone. Neither one cares whether a sentence sounds like it came out of a language model.
The third layer is new, and it’s the one causing the confusion. Fraud and identity detection, the category Greenhouse’s Real Talent belongs to, showed up in 2025 to solve a different problem entirely: bots mass-applying to postings, and candidates using deepfakes to get through video interviews. It was never built to read your resume for AI phrasing. There’s no ATS AI flag sitting in the code waiting to catch a prompt-generated bullet point, because nobody built one.
So, Is AI Resume Detection Actually a Feature Anywhere?
No, not as a shipped feature on any platform tested so far. Jobscan’s reporting on Greenhouse notes the company is explicit that its matching AI never ranks candidates or assigns numerical ratings on its own, and that every hiring decision stays with a human reviewer. A separate 2026 audit of ten ATS platforms by StylingCV, including Workday, Greenhouse, Oracle Cloud HCM, SAP SuccessFactors, and Lever, found the identical pattern: all ten use AI to parse resumes, and, in the audit’s own words, “zero of them use AI to detect whether a human or machine wrote your bullet points.”
Parsing AI and detection AI are not the same capability. Every major ATS currently has the first one. None of them, publicly, have the second.
That distinction is worth sitting with, because it changes what you’re actually optimizing against. There’s no scoring penalty hiding inside a resume-parsing engine for AI-typical phrasing, because that engine was never asked to look for it in the first place.
What Greenhouse’s Real Talent Actually Checks
Real Talent, announced by Greenhouse in June 2025 and rolled out through the company’s Fall 2025 release cycle, is built around a partnership with CLEAR, the identity verification company most people recognize from airport security lines. Candidates verify themselves with a government ID and a selfie inside Greenhouse’s own candidate portal. From there, according to Greenhouse’s own description, “sophisticated algorithms detect bots, fake job applicants, mass applications and patterns that suggest deception,” and the system flags “inconsistencies and unusual behaviors” through the rest of the hiring process. Greenhouse names deepfake video and cloned-voice interview fraud specifically as the newer threat category driving this.
| Real Talent flags | Real Talent doesn’t touch |
|---|---|
| Government ID mismatched to a live selfie, via CLEAR | Formal or “AI-sounding” phrasing in your resume |
| Bot-driven mass-application patterns across postings | Repetitive bullet structure or a generic cover letter |
| Deepfake video or cloned-voice signals in live interviews | Grammar that reads a little too polished |
| Behavioral inconsistencies flagged through the hiring process | None of the above. That’s still a human call. |
None of that reaches your word choice. If your bullet points read a little generic, Real Talent has nothing to say about it, and neither does the matching layer it sits on top of.
The Bottleneck Nobody’s Automated Yet
Which is where the actual risk sits. In Insight Global’s 2025 AI in Hiring survey, 88% of hiring managers said they can tell when a candidate used AI on an application. A separate May 2025 survey of 600 hiring managers by TopResume found that 33.5% correctly identified AI-generated resumes in a blind test, and among the ones who guessed right, the average time to spot it was 20 seconds. Resume Genius’s own hiring-manager research lands in a similar place from a different angle: 74% say they’ve encountered AI-generated content in applications, and 47% say they’ve specifically seen AI-written resumes or cover letters.
I’d treat that 88% figure skeptically. Believing you can spot AI and actually spotting it correctly aren’t the same claim, and none of these surveys tested hiring managers against a labeled sample to check their accuracy. But even discounted, the pattern holds up: recruiters are pattern-matching against your writing on instinct, in under half a minute, and no ATS product update touches that.
The Same Bias AI Detectors Had With Non-Native Writers
There’s a 2023 study out of Stanford worth knowing about here, published in the journal Patterns. Researchers ran 91 real TOEFL essays, written by non-native English speakers under exam conditions, through seven commercial AI-text detectors. On average, 61.3% of those essays got flagged as machine-written. Native-English writing samples from the same test got flagged 5.1% of the time. The detectors weren’t actually reading for AI. They were reading for plainer, more formal sentence structure, the kind non-native speakers tend to produce under exam pressure, and mistaking that for a machine.
That study tested software, not recruiters, and I don’t have a comparable dataset proving humans make the identical mistake at the identical rate. But the underlying signal, formal phrasing, lower idiom use, fewer contractions, is close to what a recruiter skimming for “AI vibes” in 20 seconds would flag too. If your natural writing register already reads as more formal because English isn’t your first language, an unconscious human heuristic is arguably a bigger risk to you than any actual ATS feature.
What This Actually Means If You’re Applying With AI Help
Use AI to draft. Don’t submit its first draft. The applications that get flagged, by a person, not a bot, don’t get flagged for using AI. They get flagged for reading like everyone else’s AI output: the same three-bullet structure, the same “spearheaded” and “orchestrated,” the same shape a thousand other candidates pulled from the same prompt. Rewrite the output in your own sentence rhythm before you send it. Swap the generic verb for the specific one. Add the number only you would actually know.
If you want a check on the more boring half of this problem, whether a parser can even read your resume correctly in the first place, LastRound’s free ATS resume checker scores that directly: contact fields, structure, bullet quality, and a keyword-match pass against a real job description, all run in your browser. It has nothing to do with AI detection. It’s the parsing layer, the one that actually decides whether you clear the first filter.
And if you’re applying at a volume where hand-tailoring every version starts to break down, that’s the actual case for a tool like LastRound’s Auto-Apply: a review queue where you approve every tailored resume and cover letter before it goes out, not a bot firing blind. Ten applications a month on the free plan, up to 400 on the top paid tier, and every single one still passes through you first. That review step is what keeps the output from turning into the templated flood recruiters already say they’re tired of.
FAQ
Can Greenhouse detect if I used ChatGPT to write my resume?
No, not directly. Greenhouse’s matching AI is built for candidate-to-job comparison, not writing-style detection, and the company is explicit that it never ranks or scores candidates on its own. Real Talent, the newer feature launched in 2025, checks for identity fraud and bot application patterns through a CLEAR partnership. It doesn’t read your resume’s prose for AI phrasing.
Do ATS systems reject AI-generated resumes automatically?
No. Automatic rejection would require an ATS to score resumes on writing style, and none of the major platforms tested, including Greenhouse, Workday, and Lever, currently do that. When an AI-written resume gets rejected, it’s usually because a recruiter read it, decided it looked generic, and passed. That’s a human decision wearing an automated system’s reputation.
The software was never really the risk. A person spends about 20 seconds on your resume, decides it reads like everyone else’s in the stack, and moves to the next one before you get a fair read. Fix that problem, and the ATS question mostly answers itself.
