AI wrote the CV. Now what are you screening?

AI writing tools have changed the economics of applying for jobs: a candidate can generate a polished, tailored, keyword-matched CV and cover letter for any posting in seconds, and apply to hundreds of roles in an afternoon. The result lands on the hiring side as two problems at once: application volume has surged, and the documents no longer separate candidates, because every CV in the pile is now well-written. When the paperwork all looks the same, early screening has to sample the person instead. That is the case this post makes.

What changed

For decades, CV screening worked as a rough proxy: writing a decent CV took effort, tailoring it took more, and that effort correlated loosely with seriousness and communication skill. AI tools removed the effort. Mass application is now nearly free for candidates, so postings that once drew dozens of applicants draw hundreds, and recruiters and hiring teams widely report the surge. This is not candidates behaving badly; it is candidates responding rationally to a market where employers automated rejection first. But it breaks the proxy either way.

Why document screening now fails in both directions

False positives. A keyword-matched, confidently formatted CV used to suggest a strong candidate. Now it suggests a candidate with access to a chatbot, which is every candidate. The signal you were paying for is gone.

False negatives. Keyword filters and quick skims still reject people whose experience is real but whose paperwork is plain, and they do it faster than ever because the pile is taller. Your strongest applicant may be indistinguishable, on paper, from a hundred generated documents, and the skim has no way to know. The arithmetic of what that skim actually costs you is laid out in how to screen 200+ applicants without phone screens.

Doubling down on document-side automation does not fix this: an AI filter reading AI-written CVs is an arms race with no signal at the bottom of it.

Sampling the person instead

A structured recorded answer changes what is being tested. When every candidate answers the same specific questions on camera, in their own words, within a time limit, you observe things a generated document cannot carry: how the person communicates, how they organize a thought, and whether their claimed experience produces real detail when probed. The format is described in full in what is a one-way video interview.

Honesty requires saying the quiet part: candidates can use AI to prepare recorded answers too, and preparation is legitimate. The difference is that delivery cannot be outsourced. A rehearsed answer to a specific behavioral question still shows you the candidate speaking, and question design does most of the protective work: specific, story-based prompts of the kind collected in 50 one-way video interview questions are far harder to fake than definitional ones, because generated answers go generic exactly where real experience goes concrete.

The same principle applies on your side of the table. AI belongs in your screening as an assistant that transcribes, summarizes, and finds things, never as a judge that scores or rejects, a line drawn carefully in how AI should (and shouldn’t) be used in video interview screening.

Frequently asked questions

Why did applications to my job posting suddenly spike?

AI writing tools have made applying nearly effortless, so candidates apply to far more roles than before. Higher volume per posting is now the market default, not a sign your posting went unusually viral.

Can AI-written CVs be detected and filtered out?

Not reliably, and it is the wrong goal. Using AI to write a CV is not cheating, and detection tools produce false accusations against real candidates. The durable fix is weighting early screening toward formats where the candidate must perform rather than submit.

Can candidates use AI to cheat in recorded video interviews?

They can use AI to prepare, which is legitimate, and preparation shows. Specific, experience-based questions with time limits leave little room for generated filler, because a candidate reciting borrowed material sounds different from one recalling their own work.

Should I use AI to automatically reject the extra applicants?

No. Automated rejection is where AI screening has repeatedly failed and where regulation is most active. Use AI to reduce review effort, and keep every decision about a candidate with a person. This is general information, not legal advice.