Can Employers Tell If You Used AI on Your Resume?
If you drafted or polished your resume with ChatGPT, Claude or Gemini, you have probably wondered whether the employer can tell. There are two different questions here: can software detect it, and can a person notice it? The evidence gives different answers to each.
What AI detectors can and cannot do
Detection tools estimate how "predictable" a text is. Short, formulaic writing like a resume is a hard case for them.
The strongest evidence comes from OpenAI itself. As Search Engine Journal reported, OpenAI updated its classifier announcement to say: "As of July 20, 2023, the AI classifier is no longer available due to its low rate of accuracy." The company that built the model could not build a dependable detector for its output.
Independent research points the same way. In GPT detectors are biased against non-native English writers, Liang, Yuksekgonul, Mao, Wu and Zou tested seven widely used detectors on 91 TOEFL essays written by people. The detectors' average false-positive rate was 61.22%. Eighteen essays (19.78%) were flagged as AI by all seven detectors, and 89 (97.80%) were flagged by at least one. The same paper found that simple prompting could push detection of AI-written essays down from 100% to 13%.
Put those together: a detector would wrongly flag many human-written resumes, especially from non-native speakers, while missing lightly edited AI text. Employers who know this have little reason to rely on one.
What ATS software actually checks
Applicant tracking systems are built to parse and match, not to judge authorship. The vendor documentation reviewed for this series describes matching and grading features, and none of it mentions AI-text detection:
- Workday HiredScore's grading compares your resume's qualifications with the job's. Its August 2026 release note describes an LLM that "evaluates each screenable qualification individually against the candidate's parsed resume."
- Eightfold's vendor disclosure lists titles, experience, education, skills and resume text as inputs to a match score.
In other words, the screening software asks "does this resume match the job?", not "who wrote it?" A clean, well-matched AI-assisted resume passes the same way a human-written one does.
| Signal | Who notices | Reliability | What to do |
|---|---|---|---|
| AI-text detector score | Software | Low: OpenAI retired its own; high false positives on non-native writing (Liang et al.) | Nothing to optimise; do not paste your resume into 'humaniser' tools |
| Overused LLM vocabulary | Recruiter | Moderate: measurable word-frequency shifts in research corpora | Replace with plain verbs and specifics |
| Generic bullets with no numbers or names | Recruiter | High: the most common tell | Add the tool, the scope and the result |
| Claims you cannot explain | Interviewer | Very high | Only keep what you can talk about for two minutes |
| Resume and LinkedIn disagree | Recruiter | High | Keep titles and dates consistent |
What recruiters actually notice
The tell is rarely a single word. It is a texture: bullets that could belong to anyone, adjectives where evidence should be, and the same rhythm in every line.
Researchers have measured the vocabulary part of this. Kobak and colleagues analysed more than 15 million PubMed abstracts in Delving into LLM-assisted writing in biomedical publications. They found 379 "excess" style words in 2024. "Delves" appeared 28 times more often than expected, "underscores" 13.8 times and "showcasing" 10.7 times. Liang and colleagues, studying AI-conference peer reviews in Monitoring AI-Modified Content at Scale found "commendable," "meticulous" and "intricate" with 9.8-, 34.7- and 11.2-fold increases. These are academic corpora, not resumes, but the same models write both.
So a recruiter reading "Meticulous, results-driven professional who delves into complex challenges" is not running a detector. They are recognising a pattern they now see every day.
Using AI without sounding like it
The fix is not to avoid AI. It is to give it your facts and keep only what you can defend.
- Start from your real material. Paste your current resume and the job posting, and ask the model to rewrite only using facts in your resume.
- Ban the stock vocabulary in the prompt. Tell the model to avoid the words listed above and to use plain past-tense verbs (our resume action verbs list has alternatives).
- Demand specifics. For each bullet, require a named tool, a scope (team, budget, volume) and a result. Where you lack a number, the bullet should say so rather than invent one.
- Read every bullet aloud. If you cannot explain it for two minutes in an interview, cut it.
- Vary the rhythm. Not every bullet needs a percentage and three clauses.
- Check consistency with LinkedIn: titles, dates and employers should match.
A prompt you can copy:
Rewrite my resume bullets for the job below. Use only facts in my resume. Do not invent numbers; where a result is missing, write [add metric]. Avoid these words: delve, meticulous, commendable, intricate, showcase, underscore, leverage, spearhead, dynamic, results-driven. Use plain past-tense verbs and name the tools.
Some tools build this rule in. The ATS Resume AI optimizer marks posting terms it cannot trace to your resume as [confirm: …] and missing numbers as [add: …] rather than filling them in. For a full workflow, see how to use AI to tailor a resume and our ChatGPT prompts for tailoring.
A before/after: the same bullet, generic and specific
Here is a bullet of the kind a model produces when it is given a job title and little else:
Spearheaded dynamic cross-functional initiatives, meticulously leveraging data-driven insights to drive transformative operational excellence.
It contains "meticulously," a form of a word both studies above flagged, plus a stack of stock buzzwords, and no tool, no scope and no result. A recruiter cannot tell what the person did, which is a bigger problem than how it was written.
Here is the same bullet after the person supplies the facts:
Ran the weekly inventory review with purchasing and two warehouse leads; rebuilt the reorder report in Power BI, which cut stock-outs on top-50 items from 14 to 5 a month.
Same person, same job, perhaps the same AI tool. The second version names a team, a tool and a before-and-after count, and every part of it can be explained in an interview. (The figures here are illustrative; use your own.) If a model wrote it from your notes, nothing about it signals "AI," because it is specific to one person.
Is it against the rules?
The real risk is misrepresentation, and that applies whether the misstatement came from you or from a model. An invented metric is a problem in an interview or a background check regardless of who typed it. If an employer's application asks you to certify that the information is accurate, that certification covers AI-drafted text too.
Sources
- Search Engine Journal — OpenAI shuts down flawed AI detector
- Liang et al., arXiv — GPT detectors are biased against non-native English writers
- Kobak et al., arXiv — Delving into LLM-assisted writing in biomedical publications through excess vocabulary
- Liang et al., arXiv — Monitoring AI-Modified Content at Scale
- Workday — HiredScore Spotlight Fit and Gap enhancement release note
- Eightfold — OFCCP AI vendor disclosure (PDF)
Frequently asked questions
- Do applicant tracking systems scan resumes for AI writing?
- The ATS vendor documentation reviewed here (Workday HiredScore, Eightfold) describes parsing your resume and matching it to the job's qualifications. None of it describes an AI-authorship detector.
- Should I run my resume through an AI detector before applying?
- It tells you little. OpenAI withdrew its own classifier for low accuracy, and research by Liang et al. found detectors flag human writing by non-native English speakers at high rates. A better check is to read each bullet and ask whether it names a real tool, scope and result.
- Will a recruiter reject me for using ChatGPT?
- Recruiters reject resumes that read as generic or that do not match the job. If AI helped you produce specific, accurate, well-matched bullets, there is nothing to reject. If it produced vague adjectives, the vagueness is the problem.
- Is it dishonest to use AI on a resume?
- Using a tool to write is not dishonest; stating things that are not true is. Keep every claim traceable to your actual experience and be ready to explain it.