What an AI match score actually measures, and what it doesn't

People treat a match score as a verdict. It isn't. It's a structured comparison between what a role asked for and what a profile can evidence, expressed as a number so you can sort by it. Understanding the difference stops both the overconfidence and the suspicion.

What goes into the number

• Relevant experience: depth and how close the domain is to the role, not just total years.
• Required skills and tools, weighted by whether the posting marked them essential.
• Seniority and scope: team size, budget, whether the level roughly lines up.
• Hard practicalities: salary expectation against band, availability and notice, location and remote preference.
• Sector and regional context, where the role genuinely depends on it.

Two profiles with the same years of experience can score 30 points apart because one evidences the specific responsibilities the spec asked for and the other lists a job title and leaves the rest to imagination.

What it cannot see

• Whether someone will stay two years.
• How they behave under pressure or with a difficult supplier.
• Whether they'll get on with the hiring manager.
• Anything genuinely absent from the profile — an unlisted skill is an invisible skill.

So a 92 isn't "this person will succeed". It's "this person's evidence lines up closely with what you asked for". A 58 isn't a bad candidate; often it's a good candidate against a spec that asked for the wrong things, which is why reading the reason matters more than reading the number.

Why the reason is the product

Every score comes with one or two plain sentences explaining it. That sentence is what makes the number arguable — and a number you can argue with is the only kind worth acting on. If the reason repeatedly penalises something you don't actually need, the correct response is to fix the spec, not to distrust the score.

Scores narrow the pool consistently. Judgement still belongs to you, and it should.