What does an AI candidate match score actually mean?

A candidate match score can look precise without having a clear interpretation. Before a hiring team uses a score to order applications, it should be able to explain what that number represents and what evidence changes it.

Ask the product provider to define the measure in plain language. "AI-powered" is not a definition.

Find out what the score measures

Does it represent matched criteria, a model's assessment of relevance, or a comparison with other applications? Is the scale fixed for the vacancy, or does the number change when more people apply?

A displayed percentage is not automatically a probability of performing well in the role. Treat such a claim as something requiring evidence for the intended use, not something implied by the percent sign.

Ask whether different job families use different scoring rules and whether the organization can inspect those rules.

Read the evidence behind the total

Look for the requirements considered, the resume passages supporting them, and any unknowns. A single total can hide a missing essential requirement or overvalue a preference that the hiring team did not intend to prioritize.

As a test, compare two fictional applications with similar totals. One might have clear evidence of an essential skill and several unconfirmed preferences. The other might match many optional terms while leaving the essential skill unconfirmed. The total alone does not resolve that difference.

Check how uncertainty is handled

Ask what happens when a resume omits a date, uses unfamiliar terminology, or fails parsing. "Not found" should remain distinguishable from a confirmed absence where the workflow allows it.

Test whether a retry changes the score and whether the product records the version and criteria used. If it does not expose those details, keep that limitation in the evaluation record.

Use the score for a defined purpose

Decide whether the output helps navigate applications, flags evidence for review, or supports another bounded task. Keep a person responsible for the application decision and provide a way to question unsupported conclusions.

Review a sample of lower-ranked applications to look for missed evidence. When reviewers disagree with the score, record why so the team can distinguish document errors from unsuitable assessment rules.

An explainable score is one the team can discuss in relation to the job and source material. If nobody can account for why a candidate received a number, avoid making that number the deciding factor.

Further reading

• Employment Discrimination and AI for Workers
• EEOC Launches Initiative on Artificial Intelligence and Algorithmic Fairness | U.S. Equal Employment Opportunity Commission

Related guides

• Define the screening criteria
• Review the shortlist with source evidence
• AI resume screening: how it works and where it can go wrong