In this article
Before you start
Why interviewers ask this
Candidates ask whether software will flag them, and recruiters may ask candidates how they use AI, so the question appears from both sides of the table. The useful answer is not a yes or a no. It is a three-part picture: what the core system does, what an employer can add, and who decides. Candidates who give that picture show that they read documentation instead of repeating forum rumours, and that they understand hiring as a process with people in it.
A weak answer says "ATS tools catch AI" or "they never do", and neither is documented. A strong answer names what vendors publish, says what is not documented, and treats the text as the candidate's own responsibility to stand behind.
How an ATS handles AI, from first principles
In plain words
- Applicant tracking system
- An applicant tracking system is a type of software that an employer uses to collect, store and move job applications through hiring stages. Recruiters search and filter it. They do not read every application in a pile.
- Resume parsing
- Resume parsing is the step in which software reads an uploaded resume. It sorts the text into fields such as name, employer, title and skills. Those fields can then be searched.
- AI-text detector
- An AI-text detector is a program that estimates whether a passage came from a language model. It returns a likelihood, not a proof. It is a different tool from the ATS itself.
Think of an ATS as a filing cabinet with a clerk. The clerk sorts what arrives. Optional extra readers sit on a nearby shelf. The clerk files and routes. An extra reader scores or checks. A person decides. The analogy breaks here. A real clerk reads every page. An ATS mostly stores and routes, and leaves the reading to people unless a scoring add-on is switched on.
In one precise sentence, the two core systems read for this page (Greenhouse and Ashby) document recruiter help, workflow and fraud or spam signals, while an AI-text check is an optional separate tool that some employers connect to them.
Step 1: What the core ATS documents
The two core ATS vendors read for this page document recruiter assistance and fraud or spam screening, not an authorship check. Greenhouse's support page on AI features, updated in September 2026, lists seventeen features, among them scorecard summaries and talent matching, and none detects AI-written applications (Greenhouse support, read October 2026).
Greenhouse presents Real Talent as its response to "fraudulent applications, AI-generated resumes, and identity misrepresentation". Its listed features screen for fraud using "objective digital signals", blocklist known malicious internet protocol (IP) addresses, email addresses or domains, and offer identity verification through a named third-party provider. They check who applied, and none is described as judging who wrote the text (Greenhouse Real Talent, February 2026).
Ashby's candidate fraud detection is available on all plans (Ashby docs, March 2026). Neither vendor describes an authorship check.
An application sent by an AI auto-apply tool can be flagged as spam or a bot, though the flag would come from submission signals, not from judging who wrote the words. This describes two vendors' documentation, not every ATS.
Step 2: What scoring and screening add-ons check
A scoring or screening add-on ranks fit or checks who applied, and its documentation says nothing about AI authorship. The AI Screened integration, described on a Greenhouse support page updated in August 2026, writes a fit percentage from 0 to 100 and a "Strong Fit" or "Not a Fit" recommendation onto the candidate's profile, and the article says recruiters choose which screened candidates get an interview (AI Screened integration page, read October 2026).
The Brainner integration is described as a resume screening and fraud detection platform. Its identity check returns No Issues, Flagged or High Risk, using signals such as phone and email fraud scoring and employment-history consistency (Brainner integration page, read October 2026). Fraud detection here means fake or high-risk applicants, not AI-written text. Neither page describes an AI-text check. The input is the parsed application; the output is a score or a flag.
Step 3: Where a separate AI-text detector can be added
A dedicated detector, where an employer buys one, is a separate product that connects to the ATS. Pangram's help center says its integration with Ashby "automatically scans candidate application materials for AI as candidates apply", shows results inside the Ashby interface for recruiters, and is available on Pangram's Team and Enterprise plans, with setup arranged through Pangram (Pangram help center, reviewed July 2026).
So the honest answer to "does ATS detect AI" has an "it depends on what the employer added" clause. The documents show that an AI-text scan exists as an integration, not that every employer uses one, and they do not give a list of employers who do. The input to this step is application text, and the output is a result shown to a recruiter.
Step 4: Detectors make mistakes, and a person decides
An AI-text detector gives a probability that can be wrong, and the research on such tools warns against using them to judge people. Liang and colleagues (arXiv 2304.02819, revised July 2023) found that several widely used detectors consistently misclassified non-native English writing samples as AI-generated, while native samples were classified accurately. They caution against using these detectors in evaluative settings, where they could exclude non-native speakers.
That research is on general writing samples, not on resumes, and detector products have changed since. It fits how these vendors present results: the Greenhouse page for AI Screened describes recruiters making the interview call, and the Pangram page shows results to the recruiter. The input is a flag, and the output is a human decision: a separate detector can scan application text, and a recruiter decides.
Step-by-step description
- It parses, routes and flags fraud signals.
- An optional add-on may rank fit or screen.
- An optional tool may scan for AI text.
- A person reviews flags and decides.
- The ATS core parses, routes and flags fraud signals.
- A separate detector can scan application text, and a recruiter decides.
Text description of the diagram
- The Greenhouse AI features page lists seventeen features and none detects AI-written applications.
- The Pangram integration for Ashby scans application materials for AI as candidates apply.
Knowing it is half the job. Saying it is the other half.
Interview Clozo is an AI interview copilot for real interview rounds: it hears the question and puts a strong answer on your screen, and you say it in your own words. See the features or how it works.
Start Free Trial →The Staff-level answer
Does an applicant tracking system detect AI-written resumes?
The two core systems I have read about, Greenhouse and Ashby, document recruiter help and fraud or spam checks, not an authorship check. Greenhouse, for example, lists seventeen AI features and none detects AI-written applications. Some employers add a separate AI-text detector that connects to the system, and one vendor documents exactly that for Ashby. Detectors return likelihoods, and research found they can misclassify non-native English writing, so the tools read here present results for a person to review. My practical position is to write nothing I could not explain, because the real check is a recruiter or interviewer asking me about any line of my resume.
What makes it Staff-level is the structure: it separates the core system from add-ons, states the limit of the evidence, and ends on what the candidate controls.
The rubric
| Dimension | Weak answer | Strong answer | Staff answer |
|---|---|---|---|
| Correctness | "ATS tools catch AI" or "never" | Greenhouse and Ashby document no authorship check | Core versus add-on, with the vendor named |
| Depth | No examples | Names one vendor feature list | Names scoring, identity and detector tools |
| Trade-offs | Ignores false flags | Mentions detectors can be wrong | Cites the bias finding and human review |
| Communication | Confident guess | States what is documented | States what is not documented |
| Follow-up readiness | Stuck on "can you prove it" | Can say "it depends on the employer" | Ends on owning every line of the resume |
Three variants of this question
Can an ATS flag an application sent by an AI auto-apply tool?
Possibly, though not for its wording. Greenhouse Real Talent documents fraud signals, a spam blocklist and keeping bots out, and Ashby documents fraud detection, so a bulk or suspicious submission can be flagged as a bot. Neither page describes reading the text for AI authorship.
What if the employer has connected a detector such as Pangram?
Then application text can be scanned as candidates apply, and the result is shown to the recruiter inside Ashby, according to Pangram's help center (July 2026). Which employers connect one is not published, so assume a recruiter may see a flag and be ready to explain the line.
How would you explain a resume line a detector flagged?
Say what the line describes, show what you did, and offer detail. A flag is a likelihood, not a verdict, and research found some detectors misclassified non-native English writing as AI-generated. A calm, specific account of the work answers the concern better than a defence of the wording.
Common mistakes
- "The ATS will reject anything AI wrote" (wrong). No documented core feature does this, and detectors connect as optional tools whose results a person reviews (right: describe what is documented and what depends on the employer).
- "No tool can ever tell" (wrong). An add-on detector exists for at least one ATS, and recruiters read closely (right: assume your text may be scanned and reviewed).
- "A flag proves the candidate was dishonest" (wrong). Detectors return likelihoods and were found to misclassify some writers (right: a flag starts a conversation, not a verdict).
- "If the text is clean, the interview is safe" (wrong). Interviewers ask you to explain claims in your own words (right: write only what you can discuss in detail).
- "I should hide how I wrote it" (wrong). Some employers state rules on AI in applications, and the safe move is to read them and follow them (right: check the posting and ask the recruiter if unclear).
Follow-up questions
Can job recruiters tell if you use ChatGPT?
Not from the core ATS pages read here: Greenhouse and Ashby document no AI-text check. A recruiter can still notice generic wording, and a connected detector can show a flag, so the dependable test is the conversation. Prepare with the tell me about yourself answer and the guide to how to answer interview questions.
Do employers check if you use AI?
Some publish rules on AI in applications, and an employer can connect a detector, but no list of employers exists. Read the posting, and ask the recruiter if it is silent. Expect questions such as "tell me about this project", as in the behavioural interviews guide.
Is it okay to use ChatGPT for a job application?
Follow the employer's stated rule if there is one. What matters most is that every claim is true and that you can discuss it in your own words; the questions to ask the interviewer list helps you ask about the rule.
How accurate are AI-text detectors?
Detector results are likelihoods, not proofs, and independent research found that several widely used detectors misclassified non-native English writing as AI-generated. Treat any score as a prompt for a human to look, not as evidence.
- Greenhouse and Ashby document recruiter help and fraud or spam checks; Greenhouse lists seventeen AI features, none an authorship check, and an AI auto-apply tool can be flagged as spam or a bot.
- A separate detector can be connected, and Pangram documents it for Ashby.
- Detectors return likelihoods and were found to misclassify non-native English writing.
- A person reviews flags, so own every line you submit.
Sources
- Greenhouse AI features — Greenhouse support, 3 Sep 2026
- AI Screened integration — Greenhouse support, 6 Aug 2026
- Brainner integration — Greenhouse support, 25 Aug 2026
- Greenhouse Real Talent — Greenhouse, 1 Feb 2026
- Making hiring safer and more human: An update on Real Talent — Greenhouse, 31 Jul 2025
- Running candidate fraud detection checks and reviewing fraud signals — Ashby docs, 20 Mar 2026
- Can I use Pangram AI detector with Ashby? — Pangram help center, 24 Jul 2026
- GPT detectors are biased against non-native English writers (arXiv 2304.02819 v3) — arXiv (Liang, Yuksekgonul, Mao, Wu and Zou), 10 Jul 2023
