Every résumé in your inbox this morning is well-written. Tight bullets. Clean structure. Keywords that mirror your JD almost perfectly.

That's the problem.

A polished résumé used to tell you something. This year it tells you exactly one thing: the candidate knows how to use AI. Everyone does.

Candidates use AI to write. You use AI to screen. Two models, both trained on the same job description, meeting in the middle. And the signal you leaned on for a decade — the polish, the phrasing, the keyword fit — just evaporated.

Why a “good résumé” stopped being a signal

Keyword screening was built for a world where résumés were written by hand. Now a candidate pastes your JD into ChatGPT and says “rewrite mine to match.”

Eight hundred applications. Every one hits 90%. A filter everyone passes isn't a filter.

Polish stopped separating people

A carefully written résumé used to carry a quiet signal: this person put in effort, respects the role.

Now that effort costs one click and ten seconds. You can't score effort when effort is free.

The contradiction you're living in

Your company uses AI to read résumés. Then you toss the ones that “sound AI-written.”

You're penalizing candidates for doing exactly what you're doing. That's not a filter. That's a filter arguing with itself.

800 résumés for one backend role

A Senior Backend Java role used to pull 80 résumés. Last week it pulled 800.

You open the first 40. All 40 read the same — same structure, same power verbs, same keywords you wrote into the JD. Not because the candidates are identical. Because one model wrote for all of them.

Eight hundred copies. Eight hundred different signatures. Three hours later, your shortlist is basically random.

A résumé is no longer evidence. It's the output of a prompt. And you're trying to rank prompts.

Stop scoring the prose. Start testing for consistency.

When every surface is glossy, the only thing left that tells the truth is whatever AI doesn't control. There are three layers of it — call it the “three signals AI can't fake.”

Layer 1 — Internal consistency

Do the timelines add up? A “senior” who graduated two years ago? Skills claimed that no listed project would ever have used?

AI writes fluently — it doesn't audit itself. The contradictions stay right there on the page, if you actually read closely.

Layer 2 — Specificity

“Optimized the system” versus “cut p99 latency from 800ms to 120ms by collapsing queries and adding a read-through cache.”

Generative text is brilliant at generic. Real detail is hard to invent. The more specific it gets, the harder it is to fake.

Layer 3 — Evidence off the page

A repo, a shipped product, a measurable contribution, a name you can check. Things that exist outside the PDF.

A PDF can be generated. A track record can't.

You can do all three by eye. The catch: at 800 résumés, your eyes run out of battery around number 40.

The Knoot Angle

When every résumé is polished, a recruiter's edge isn't reading faster — it's seeing through the polish.

That's why Knoot's AI Screening doesn't grade “smoothness.” It scores each résumé against criteria pulled straight from your JD (via JD Analyzer), instead of matching keywords blind.

In practice, AI Screening helps you:

• Flag risk: overlapping timelines, skill claims the experience doesn't support, inflated scope.
• Rank a shortlist with a match % — and explain why each candidate made it.
• Read the reasoning, not 800 files.

And the part that matters most: AI flags, you decide. It never rejects anyone for you.

AI can write the words. It can't write the truth. That part is still yours.