What an ATS Selects For When Resumes Are Written by AI

Suppose two people apply for the same operations role. One has run a warehouse shift for six years and writes her own resume the night before, a bit awkwardly. The other has less experience, but he pastes the job description into an AI tool and asks it to improve his resume. Which of them does your applicant tracking system put near the top?

Almost certainly the second. The ATS is working as designed. The trouble is that it was designed for a world where people wrote their own applications.

Most of these systems run on simple logic. A resume that mentions the right skills, often enough, in the expected format, gets surfaced. That was crude, but it was roughly useful as long as a person wrote what they submitted. If someone listed "stakeholder management, cross-functional delivery, Python," you could assume they had at least run into those things at work.

AI-written resumes break that assumption. Give a model a job description and ask it to improve a resume, and it will work in every relevant keyword at the right density, arrange them in the structure a parser expects, and add achievements with specific-looking numbers. "Increased pipeline conversion by 34%" sounds credible. It might be true, roughly true, or made up, and the ATS has no way of knowing which. So the document ranks well because it was made to rank well. The shortlist looks qualified, but what it really samples is people who are good at using AI in a job search. That's a skill. It's rarely the one you're hiring for.

For a senior hire, where a recruiter reads every resume anyway, this is manageable. Experienced recruiters catch some of it. The damage shows up in high-volume hiring, where nobody reads everything. Suppose a role gets 400 applications, which isn't unusual now that aggregator sites have made applying almost free. The ATS decides who gets seen. If the top 40 got there because their applications were well optimized, you've filtered for skill with AI tools rather than fit for the job, and the people who applied without help, often the more operational ones who don't think much about application strategy, sink down the list.

It also gets worse over time. Applying to 50 jobs with an AI-polished resume now takes about as much effort as applying to five used to. So the number of applications goes up, the share that are polished goes up, and the ATS keeps ranking all of it on keywords. It's scoring documents that were written to score well against it, and at some point the score stops telling you much about the person.

Why detection won't save you

The natural response is to fight back in the text: build better parsing, spot AI-written resumes and discount them. I don't think that goes anywhere good. It's an arms race between models writing applications and models trying to catch them, and job seekers have every reason to keep winning it. It also punishes the wrong thing, since tidying up a resume with AI doesn't make someone worse at the job.

What would work is to stop treating the written document as the main thing you screen on, and ask people to talk instead. Someone who wrote "led enterprise software implementations" either knows what that meant day to day or doesn't, and a few minutes of conversation shows which. A question like "you mentioned managing a team, can you walk me through what that looked like?" is hard to answer well from a script someone else wrote.

We built Asendia to be that conversation, in place of the keyword ranking. Instead of scoring the document, it phones each person soon after they apply and has them explain their work in their own words, asking a follow-up when an answer is thin. Since it calls everyone rather than the 40 at the top of a ranked list, the warehouse lead with the awkward resume gets heard, and the polished resume nobody can back up doesn't advance. What each person said goes into the ATS as a written summary. We've also written about where a spoken screen fits in a pipeline that mostly runs itself.

I don't expect any of this to reverse. As long as polished applications do better in keyword filters, more candidates will use them, and the filters will tell you less. The resume is still good for some things, like knowing where someone has worked. But if it's still the main thing deciding who you talk to, it's worth asking whose skill your shortlist is really measuring.

Ready to transform your hiring strategy? Schedule a Demo with our founders today!