Can employers really spot AI-written job applications, or are they judging AI use they can't directly verify, and what does that change?

This explores whether employers can spot AI-written applications, and what changes in hiring when they can't. The short version is that the corpus has no study that tests detection accuracy directly. What it does show is more interesting: employers react to AI use without needing to catch it, and in some cases the screening tools prefer the AI-written version.

Start with what recruiters say about themselves. In Greenhouse's survey, 91% of recruiters say they spot deception. Yet 41% of job seekers admit to hiding prompt injections in their applications to get past automated filters, and a third of recruiters spend half their week clearing out spam Are job applicants and employers locked in an escalating AI arms race?. Recruiters saying they catch deception doesn't mean they catch most of it. The same data describes an arms race, with applicants gaming the filters and employers tightening them, and nobody has measured who is winning.

The better evidence comes from a real marketplace. On Freelancer.com, an AI cover letter tool made polished, tailored letters cheap to write. Employers didn't respond by trying to detect them. They stopped trusting letters. Letter quality became much worse at predicting who got interviews and offers, and employers relied more on work history and reputation Does AI-generated cover letter access weaken hiring signals?. That may be the real answer to the question: once a good letter costs nothing, it no longer tells employers much, whoever wrote it. Detection matters less than the fact that the signal has stopped working.

The surprise is that when employers use AI to screen, the screener can reward AI writing. In a test with 2,245 resumes, eight of nine language models preferred resumes they had rewritten themselves over matched human versions. Preference rates ran as high as 98%, the effect got stronger in larger models, and it came from familiar writing style, not better content Do language models favor resumes they rewrote themselves?. So an applicant who runs a resume through the same kind of model the employer uses may gain an advantage instead of getting caught. Disclosure complicates this further. LLM raters showed demographic preferences when AI use was hidden, and those preferences disappeared once AI involvement was disclosed, while human raters penalized disclosed AI use the same way across groups Do LLM raters show hidden demographic preferences that disclosure erases?.

Applicants seem to know all this. Across four experiments with 4,439 people, AI users expected to be judged less competent and less diligent, and they were less willing to tell managers they had used AI Do people fear judgment when they use AI at work?. Meanwhile, recruiters in a conjoint experiment gave 8 to 15 percentage points more interview invitations to candidates who listed AI skills, mostly without checking whether those skills were real Do AI skills help candidates get more job interviews?. Together, these point to an odd incentive: say you're good with AI, but don't show you used it on this application. If you want to understand the hiring side of AI, the most useful question is not whether employers can detect it. It is which signals still mean something once anyone can produce a polished application.