AI candidate matching
Also called: candidate-job matching, AI matching
What “matching” means in practice
The model produces a similarity score between a candidate and a role. The score is shown to the reviewer alongside an explanation: “matches on Python, 5 years backend, EU work authorization ; weak on team-leadership signals.”
Good matching tools expose what drove the score. Black-box matching tools that show only a number are less useful — the reviewer can’t argue with a number; they can argue with an explanation.
The two layers
- Skills match : surface-level keyword and skill overlap. Fast, explainable, prone to false negatives (a candidate whose CV uses “TypeScript” instead of “JS” looks like a worse match than they are).
- Semantic match : embedding-based comparison of CV and role description as text. Catches relevant experience the skills layer misses, at the cost of being less explainable.
Most modern matching combines both, with a hybrid score.
Where it sits in the funnel
Candidate matching feeds two upstream uses:
- Inbound ( AI screening ) : rank applications against the role.
- Outbound ( AI sourcing ) : surface passive candidates whose profiles match the role.
Same mechanism, different funnel stage.
What candidate matching is not
It is not the hiring decision. A high match score means “worth a real conversation,” not “make an offer .” Treating it as the latter is the most common failure mode of AI in hiring.
Where TalentGleam fits
TalentGleam exposes match scores with per-dimension explanations, so the reviewer always sees what the AI is keying on. The reviewer’s judgment stays primary. See the features page .
