How AI ranking actually works

What it reads

When a candidate applies, the model compares their résumé and screening answers against the role you published — not against a generic template for that job title. That is why it works on a dental hygienist role and a Laravel role without any configuration from you.

The four verdicts

Verdicts are deliberately coarse. A 1–100 score invites false precision and endless debate; four buckets tell you what to do next.

1. 1Strong YesMeets every must-have with relevant, recent experience. Interview these first — usually the top three to five of a fifty-candidate pool.
2. 2YesMeets the must-haves with one soft gap. Worth a screening call.
3. 3MaybePartial match, or the résumé is too thin to tell. Read these when the Yes pile runs out.
4. 4NoMissing a deal breaker. Still searchable and still in your talent pool — never deleted.

Deal breakers

Marking a screening answer as a deal breaker is the strongest signal you can give. A candidate who fails one lands in No regardless of how strong the rest of their application is. Use it for genuine hard requirements — a licence, a work authorisation, a location — and nothing else.

When it gets it wrong

Override any verdict manually; your override always wins and is recorded in the activity log with your name against it. If you find yourself overriding the same way repeatedly, tighten the role requirements rather than fighting the ranking.