Does ATS Detect AI Resumes? We Researched the Top 10 Systems

Right away, the answer is no. None of the major ATS detect AI in resumes. Not Workday. Not Greenhouse. Not Oracle (which is literally generating AI content for candidates). It’s zero out of ten. And this is by design, not because of a lack of capability.

Here’s a real statement from social media, racking up thousands of likes and hundreds of comments:

Users are quick to believe when marketing speak is delivered with confidence. No wonder they're willing to pay for AI humanizer tools. Some become so paranoid that they’re running their resumes through three rounds of paraphrasing.

If you’re also losing sleep over whether the algorithm can smell ChatGPT on your resume, then we have some good news for you. We investigated ten of the most-used applicant tracking system (ATS) platforms.

The fear being spread is fake. Let’s take a look at each popular ATS and see how they actually work.

What ATS systems actually do?

Here’s the first thing you need to know. There’s more than one type of AI that could be integrated in ATS.

Parsing AI is what every ATS has. It reads your resume, pulls out titles, dates, skills, and education, and maps them to job requirements. Its goal is to figure out if you’re a suitable candidate for the role.

AI detection analyzes writing patterns, sentence variation, and statistical predictability to guess if a human or AI typed the words. But it’s not present in any ATS platform.

Here’s a side-by-side comparison.

Most applicants hear that ATS uses AI and assume that ATS detects AI. But this is simply not the case. Here’s why.

10 ATS platforms, zero AI detectors

Let’s go over what’s known about each system and look for evidence that they don’t use AI detection.

Workday

It uses agentic AI for candidate ranking, predictive hiring analytics, and HiredScore acquisition that strengthens its scoring. None of these features examines your resume for AI-generated content.

iCIMS

iCIMS uses Coalesce AI, sourcing agents, and a feature called Generative Engine Optimization. No AI detector included. Coalesce AI's responsible-AI framing mentions "explainable" governance, which is its own architectural reason not to ship detection.

Greenhouse

Greenhouse uses LLM-powered semantic matching, integrates with OpenAI, and runs WardenAI bias audits.

Their own product documentation mentions that “recruiters are drowning in AI-generated applications.” Its solution wasn't building a detector. Instead, they built MyGreenhouse, a candidate intent portal that asks applicants to declare what they are actually after.

Oracle Cloud HCM

Here’s the finding that’s probably the strongest example in our research. This ATS itself is producing AI writing on the applicant’s side of the process.

Oracle's ATS reads resumes and then uses generative AI to create work summaries or cover letters, which it attaches to the application on the candidate's behalf. In other words, the system itself acknowledges the efficiency of AI-generated content.

SAP SuccessFactors

This ATS literally deletes the uploaded resume file after 30 minutes once it has extracted the structured data. Its marketplace partner Sapia.ai has third-party reviews specifically asking for "accurate AI detection," which means SAP doesn’t deliver it.

Lever

Lever's AI capabilities run under IBM’s watsonx.governance framework, which requires explainability for any automated decision.

Meanwhile, AI-detection models are inherently unexplainable. This means the technology is architecturally incompatible with Lever's compliance stack, so it doesn't exist there and won't anytime soon.

Workable

Workable uses a 400-million-candidate database with AI features built for volume sourcing. Surprise, surprise, no AI detection included.

The product is designed to push more applications through faster, not to flag which ones look written by an LLM.

SmartRecruiters

SmartRecruiters uses Winston AI for its screening purposes. No AI detector is present here as well.

Here’s one famous anecdote related to this system: the Wellmark “hippopotamus” honeypot. An employer hid a prompt-injection trap word inside a SmartRecruiters job description, betting that AI applicants would accidentally include it in their resumes.

Ashby

No AI detection in Ashby’s stack. Here are two pieces of evidence from our investigation for this.

First, GPTZero, the leading AI-detection company on the market, uses Ashby to hire its own engineers.

Second, another Ashby customer (Close) warns applicants in its portal that “obviously AI-generated applications will be disregarded” because the platform can't do it for them. The warning is doing the work the software won't.

BambooHR

And finally, BambooHR has no AI resume detection capabilities either. It also has perhaps the most damning data point in the industry: BambooHR's own research found that 79% of people think they can spot AI writing, but only 30% actually can.

BambooHR published the proof that detection doesn’t work. It still didn't build a detector. That's a vendor acting according to their own math.

For easy reference, here’s a side-by-side comparison.

Ten vendors, zero AI detectors. So getting past AI while using AI isn’t the issue. Clearly, detection isn’t being added to these systems on purpose.

Why don’t ATS platforms use AI resume detection?

Detecting such applications is possible. Even customers keep asking for it. How come nobody delivers a product then? Well, here are the three main reasons.

The false-positive problem

GPTZero, currently the best-known consumer AI content detector, runs at a roughly 1-2% false-positive rate on careful tests. Let’s transfer that at an enterprise scale. With hundreds of thousands of applications a month, that's tens of thousands of real candidates getting flagged as AI.

Here's the kicker. The false positives aren’t random occurrences.

They concentrate on:

• ESL speakers
• Neurodivergent writers
• People writing in formal English

In other words, anyone whose prose has low burstiness.

And under NYC Local Law 144 and the EU AI Act, an automated hiring tool that disproportionately filters out protected classes is begging for a lawsuit.

ATS vendors spend serious resources on bias-audit partnerships (Greenhouse with WardenAI, Ashby with FairNow, iCIMS with TrustArc) to defend their existing AI features. Naturally, they're not going to bolt a volatile detection layer on top and blow that protection up.

The arms race is practically unwinnable

So, the way detection works is by spotting statistical patterns in machine-generated text. But advanced prompting flattens those patterns.

For example, you can ask an LLM to:

• Vary sentence length.
• Mimic a slightly impatient first-person voice.
• Add small grammatical errors here and there.

This way, detection accuracy drops to near random. BambooHR's own data shows humans land at 30% accuracy. But even an advanced algorithm at 60-70% is operationally useless for automated decisions at the enterprise scale. In the end, you can't reject candidates on a coin flip.

Detection is meaningless once the ATS strips your prose

Here's what every panic thread on Reddit isn’t mentioning. ATS vendors don't see AI-written resumes as their problem because, as we established, they also use AI.

• Workday automatically re-ranks pipelines.
• Oracle generates summaries for candidates.
• SAP discards your prose after 30 minutes and keeps only data points.

By the time your resume reaches the recruiter's queue, in some cases, the ATS has often reduced it to something like “Python, 5 years, Remote, $120K target.” It doesn't matter who typed the sentences around those tokens. The narrative is gone. Detecting authorship of a narrative that the system has already deleted is operationally meaningless.

If the system keeps tokens and throws prose away, your strategy should factor this knowledge. This means getting the right resume keywords into the document naturally, so the parser sees them where it expects to find them.

How are employers reacting to AI resumes?

Since ATS vendors won't build detection, companies are improvising.

Here are some examples:

The trend is that the industry is somewhat moving away from trusting the resume document, not toward better scanning it. Detection is the wrong investment because the resume itself is becoming the weakest signal in the automated hiring stack. It only becomes valuable once again when a human recruiter is finally reviewing it.

Now that we’ve debunked one of the common ATS myths, let’s move on to some practical advice for making an ATS-friendly resume. Because clearly the takeaway here isn’t “just let AI write everything.”

So what should you actually do?

For the time being, you need to accept that AI-written content will sound like AI. This is because it writes from probability distributions.

• If bullet points are the highest-likelihood structure, every output gets bullet points.
• If “spearheaded” ranks highest among resume action verbs, every resume gets "spearheaded."

Take a look at ChatGPT resume writing, for example. Same triadic lists. Same semantic triples. Same pre-formatted answers to predictable prompts.

You’re not crazy for having a feeling about whether something is AI-generated. There’s a sameness to it. A flatness. Humans pick it up even when they can’t articulate why, which is what the BambooHR data was really measuring. People can’t reliably name what's wrong, but they can sense when something feels off.

This is why the idea of humanizing AI output with more AI is paradoxical. And you shouldn’t waste your time with it. The whole concept is broken. You’re running a resume through ChatGPT, then through a humanizer, then through a paraphraser. It’s three layers of a machine trying not to sound like a machine. The wording changes. The rhythm doesn’t. And the absence of anything specific about you is obvious.

Quantifying achievements like that is what separates a real bullet from a generated one. The detail is found in your memory and nowhere else. It’s messy and awkward but it’s real, and that’s exactly what will set you apart from the AI-generated slop.

The AI resume builder with real-time feedback

That’s the principle Enhancv’s AI resume builder is built around.

• The AI gives you a draft.
• The feedback engine flags vague bullets and missing metrics.
• The content checker compares your resume against the job description.

Yes, Enhancv uses AI to write your resume, but it keeps you in the loop. It doesn't replace your judgment. It makes the organizing of information and writing faster, so the time you save goes into making the resume sound like a person who actually did the work.

Meanwhile, you can use our free ATS resume checker to make sure your resume is up to standard. And make sure you stand out to recruiters from the rest of the candidates with one of our beautiful ATS-tested resume templates.

Conclusion

Back to the original Facebook post.

We’re pushing back on it, as clearly as possible:

No ATS on the market detects AI in resumes. We checked the ten most-used ones. The capability doesn’t exist, and the vendors with the budget and incentive to build it have actively chosen not to.

The real risk isn’t a machine rejecting your resume. It's a human skimming past it because nothing on the page suggests that behind the content is a real person.

So stop worrying about what the machine thinks of your writing. Instead, focus on telling a compelling career story.