Twenty CVs came in for one Marketing Executive role. Seventeen of them list "proficient in ChatGPT, Canva AI, Midjourney."

You tick the AI-skills box. Next CV.

But what did you actually just verify? That the candidate created some accounts. That's it.

"Knows how to use AI" isn't a skill. It's a list of logins.

And here's the part nobody says out loud: that line is in the JD because we put it there. Nobody — not you, not the hiring manager — can define what "proficient" means for this role.

So what are you screening for?

Why "AI skills" on a CV tells you nothing

A criterion is only worth something if it separates people. This one doesn't.

Everyone claims it, so nobody stands out

Seventeen out of twenty said the same thing.

When 85% of your pipeline clears a bar, that's not a bar. That's the floor.

The CV making the claim was written by AI

This is where the loop closes.

The only evidence that a candidate can use AI well is the document in front of you — and AI wrote it. You're reading a receipt printed by the store.

Cleaner prose doesn't mean a sharper thinker. It means a better prompt.

Vague JD in, vague screening out

Pull up the last JD you posted. Odds are there's a line like "familiarity with AI tools is a plus."

No level. No context. No expected output.

You can't filter against a sentence you can't measure.

You run out of follow-up questions

You ask: "How do you use AI in your work?"

They answer: "I use ChatGPT to write content."

And then? Nothing. The conversation ends exactly where it should have started.

The pattern:

Every JD says it.
Every CV says it.
Signal value = zero.

The 3T Framework: how to assess AI skills in candidates

No special tooling required. A spreadsheet and three better questions will do.

The idea is simple: each layer of evidence has a different price. The cheaper a claim is to make, the less it's worth.

Layer 1 — Tool: what they use

The cheapest layer:

• Tool names
• Tool count
• Online course badges

Anyone can produce this list in 30 seconds. Use it to rule out people with zero exposure. Never use it to rule anyone in.

Don't score this layer. It's a prerequisite, not a capability.

Layer 2 — Task: what they use it for, and what changed

This is where real signal starts. Ask for specifics:

• How long did this take before?
• How long now?
• What still gets done by hand?
• How is the output different?

Two answers for the same content role:

"I'm proficient with ChatGPT." → a sentence.

"I used to write 8 product descriptions a day. Now I generate 30 drafts and rewrite every spec line myself, because it invents numbers." → a workflow.

The second answer isn't more polished. It's just checkable.

Layer 3 — Trade-off: when they choose not to use AI

This layer separates people who use these tools from people who've tried them.

• When has AI given them a wrong answer they caught?
• What do they deliberately keep away from it?
• How do they verify the output?

A senior backend engineer who says "I let Copilot handle boilerplate, but I write transaction logic myself because its suggestions skip rollback cases" has hit the tool's edge. You only hit the edge by working there.

The strongest AI users aren't the ones who use it most. They're the ones who know where to stop.

Putting 3T into your process this afternoon

Three changes. None of them need approval:

1. Rewrite the JD line. Replace "familiar with AI" with one sentence describing an actual task in this role.
2. Add one application question. "Describe one thing you used AI for last week: how long it took, and what you fixed by hand."
3. Score by layer. Layer 1 gets zero points. Only Layers 2 and 3 count.

That application question filters harder than you'd expect. People who actually work this way answer in three lines with numbers in them. Everyone else writes a beautiful paragraph containing nothing.

The Knoot Angle

Screening speed was never the real problem. The problem is what you're comparing each CV against. Fuzzy criteria produce fuzzy shortlists whether a human or a machine does the filtering.

That's the thinking behind Knoot's AI Screening module. It doesn't grade CVs on vibes — it checks each profile against criteria pulled from your own JD, so nothing is ever screened blind. Every candidate on the shortlist arrives with a match percentage and a written reason for it. When a skill claim doesn't line up with the experience listed underneath it, the system raises a flag so you look closer. It never removes the candidate for you.

The judgment stays yours. The machine just points at the parts worth a second read.

AI does the cross-checking. You do the asking.