When every resume is AI-written, the resume stops being a filter.
Keyword screening worked on one quiet assumption: that a resume was a costly, somewhat honest signal of fit. Generative AI removed the cost and the honesty in the same stroke. What's left is a document that matches everything and means nothing.
A resume was always a proxy. Nobody believed a bullet list captured how a person thinks — but it was a useful filter because it was expensive to fake well. Tailoring a resume convincingly to a specific role took effort, and that effort correlated, loosely, with genuine interest and relevant experience. The keyword match was never measuring competence. It was measuring the residue of effort.
That correlation has collapsed. When a candidate can paste a job description into a model and receive a perfectly optimized, keyword-dense resume in seconds, the effort signal vanishes. Every applicant now "matches." The applicant-tracking system dutifully ranks two hundred near-identical documents, and the ranking means nothing, because the thing it was implicitly measuring — the cost of a good match — has gone to zero.
The core shift
Keyword screening didn't get worse. The world underneath it changed. A filter calibrated for scarce, effortful documents is now processing abundant, effortless ones — and returning confident nonsense.
The arms race nobody wins
The instinctive response is to fight AI resumes with more AI screening — better parsers, smarter ranking, AI to detect the AI. This is an arms race, and it's one the screener structurally loses. The generator and the detector are trained on the same distribution; every improvement in detection is training data for the next generation of undetectable output. You can spend real money to end up exactly where you started, with a pile of documents you can't trust.
The deeper problem is that even a perfect resume tells you almost nothing you need. It's a record of where someone has been, not evidence of how they reason, communicate under pressure, or handle a problem that isn't in the script. Those are the things interviews exist to surface — and they were never in the document, optimized or not.
Where the signal went
If the resume no longer filters, something has to. The honest answer is that the signal moved to where it always lived: the conversation. Fifteen minutes of a candidate reasoning through a real, role-relevant problem — with follow-ups that probe the answer rather than accept it — produces more decision-grade information than an hour of resume review ever did. The interview is expensive to fake precisely because it's adaptive; you can't pre-generate your way through a follow-up you didn't see coming.
That has always been true. What's new is that the resume can no longer do the pre-filtering that let enterprises reserve interviews for a handful of finalists. When the document is noise, you either interview earlier and wider, or you make your most important decisions on the least reliable data you have.
The uncomfortable implication
Most hiring funnels are built upside-down for this world. They apply the weakest filter — the resume — to the largest, highest-stakes population (everyone who applied), and reserve the strongest instrument — the structured interview — for the smallest one (the final few). Generative AI didn't create that inversion. It just made it impossible to ignore.
The organizations that adapt won't be the ones with the best resume parser. They'll be the ones that stopped asking the resume to do a job it can no longer do, and moved the real evaluation to the first step instead of the last.