What happens to hiring when AI writes every application, and why the filters meant to save you time are quietly working against you.
There's a specific kind of Monday that talent leaders have started to dread.
You open the ATS. The role went live on Friday. Over the weekend, 300 applications landed. And as you scroll, something feels off, not because the CVs are bad, but because they're all good. Every one is polished. Every one is keyword-perfect. Every one mirrors the job description back to you with uncanny precision.
And somewhere in that scroll, a quiet thought sets in: I don't believe any of this.
That feeling isn't fatigue. It's the sound of the hiring funnel breaking.
AI didn't just speed up applications. It broke trust.
For twenty years, the deal was simple: candidates wrote CVs, and the CV was a rough proxy for the person. Imperfect, sure but a signal. The applicant tracking system sat on top of that signal and filtered by keywords, and everyone more or less accepted the trade.
Generative AI dissolved that deal in about eighteen months.
Today, tailoring a CV to any job description, injecting the exact terms a filter is hunting for, rewriting bullet points to match, generating a cover letter in your voice takes a candidate ninety seconds and costs nothing. By some estimates, a large share of applications now contain AI-generated content, and more than half of job seekers openly use tools like ChatGPT to write or optimise their CVs.
The result isn't better candidates. It's indistinguishable ones.
And recruiters know it. In one 2025 survey, 59% of hiring managers said they suspect candidates use AI to misrepresent themselves but only 19% were confident their process would actually catch it. Read those two numbers together and you have the whole problem in a sentence: almost everyone senses the funnel is being gamed, and almost no one trusts themselves to spot it.
The filter now rewards the best prompter, not the best person
Here's the part that stings. The ATS keyword filter (the thing meant to save you from the flood) has become the easiest part of the process to defeat.
If a system rewards whoever uses the right words, then it doesn't select for the best candidate. It selects for the best optimiser. The person who knew to mirror the posting. The one who ran their CV through a scanner until the match score hit 90%. Meanwhile the genuinely strong candidate (the one who wrote honestly and didn't play the game) can sink to the bottom of the pile.
No major ATS on the market today has native AI-authorship detection. So the filter can't tell the difference, and increasingly, neither can you until the interview. Or worse, until the person is three months into the job.
That's the cruel maths of it: the tool that was supposed to protect your time now guarantees you'll spend more of it. You re-check everything. You interview people you don't trust, to confirm things the CV should have told you. One executive at a large staffing group put it bluntly to us recently, they now waste far more time in interviews than they used to, because they can't trust what's on paper anymore.
The volume makes it impossible to fight by hand
You might think: fine, I'll just read more carefully. But the numbers make that a fantasy.
LinkedIn alone now processes roughly 9,500 applications a minute, up 45% year on year. Popular roles collect hundreds of applications in the first day and over a thousand across a weekend. By many accounts, 70–80% of applicants don't even meet the basic listed requirements.
So the honest recruiter is trapped between two bad options. Read everything, and drown. This is now the number-one driver of recruiter burnout, with 61% reporting significant stress. Or triage by gut and by who applied first, and accept that the best person in the pool probably slips through unseen.
Neither option is a filter. Both are a coin flip wearing a lanyard.
Why "detect the AI" is a losing game
The instinct is to fight fire with fire: buy a detector, add a step, try to catch the fakes. But this is a race you can't win. Detection tools lag behind generation tools by design, and every false positive means rejecting a real, qualified human for the crime of writing well. You end up policing the funnel instead of hiring from it.
The deeper issue is that we're still evaluating the wrong artefact. A CV optimised for keywords tells you how good someone is at optimising for keywords. It was never evidence. It was always a claim.
The way out isn't a better lie detector. It's changing what you evaluate and how.
Two shifts matter. First, evidence over assertion: a hiring decision should rest on cited, verifiable signals of what a person can actually do — not on how confidently a document asserts it. If there's no evidence, there's no score. That simple rule is un-gameable in a way a keyword match never will be, because you can't fake your way past a requirement that has to be shown, not claimed.
Second, consistency over vibes: when every manager screens by their own gut, bias and noise creep in, 48% of HR leaders admit bias affects who they hire, and structured, standardised scoring has been shown to cut that bias by more than half. One standard, applied the same way to every candidate, is the only thing that makes two people's judgements comparable at all.
The real question for 2026
Quality of Hire is the metric talent teams say they value most, and the one only about a quarter of them feel confident they can actually measure. That gap is not a coincidence. You cannot measure the quality of who you hire if you can't trust the funnel they came through.
So the question worth sitting with isn't "how do we catch the AI CVs?"
It's "what would our hiring look like if we stopped trusting the document and started demanding the evidence?"
Because every CV looks perfect now. The teams that win the next few years won't be the ones who get better at spotting the fakes. They'll be the ones who build a process where faking it stops working at all.
Where this thinking led us
This is the problem we started BlindStairs to solve.
We stopped trying to detect gamed CVs and rebuilt evaluation around the two shifts above. Every process is unique, so there's no fixed template for a candidate to optimise against, you can't prepare for a process you've never seen. And every score is built from cited evidence: no evidence, no score. That makes the evaluation un-gameable at the source, consistent across every team and hiring manager, and (because the trail exists by construction) defensible if a decision is ever challenged. The same engine assesses 100% of the pool, so nobody strong gets missed for lack of time.
It's evaluation you can actually trust again which, when the funnel is full of perfect-looking CVs, turns out to be the whole game.
If the Monday-morning dread sounded familiar, we'd be glad to show you what evidence-based evaluation looks like in practice.