Candidate matching software: AI candidate matching and job candidate matching software with the evidence behind every match score

AI candidate matching turns a job description into a set of weighted requirements, reads each candidate's resume or profile for evidence of those requirements, and scores how closely the two line up. Good matching reads meaning rather than exact words, so "built payment APIs in Go" counts toward "backend services experience" even though the phrases share nothing.

Under the hood there are four steps, and every vendor does them with different care. First the software extracts the requirements from the job: skills, years in a function, industry, level, location and any hard constraints like work authorization or a license. Second it parses each candidate into the same structure, which is where messy resumes, gaps and unusual job titles either get handled or get lost. Third it compares the two, usually with language models that understand that a "Staff Engineer" at a 40-person startup and a "Senior Software Engineer II" at a bank can be the same level. Fourth it weights and totals the result into a score.

The fourth step is where most of the difference between tools sits. A matcher that treats every requirement equally will rank a candidate who ticks eight nice-to-haves above one who has the two things the hiring manager actually cares about. That is why HireAgent asks you to mark each criterion as a must-have or a nice-to-have when you write the brief, and why a missing must-have caps the score instead of shaving a few points off it.

Here is what that looks like on a real brief. Say you are hiring a senior backend engineer with five criteria: production Go or Java, payments or fintech systems, on-call ownership of a service, mentoring, and US work authorization. A candidate who spent four years building a card-processing service in Go, led the on-call rotation and never mentored anyone lands around 82. The card shows three green criteria with the resume lines behind them, one amber ("no mentoring evidence found") and the authorization question flagged for the recruiter to confirm, because a resume rarely states it.

Keyword matching, the older approach inside many applicant tracking systems, skips the understanding step and counts overlapping terms. It is cheap and fast and it misses anyone who describes their work differently from the job ad. If a vendor cannot tell you whether its matching is keyword-based or semantic, ask them to run a resume that uses none of the job's vocabulary and watch where it lands.

A match score with no reasoning behind it is a number you cannot defend to a hiring manager, and under New York City's rules on automated hiring tools it is also a number you may have to explain to a candidate. Before you buy, put these questions to every vendor on the demo call and ask to see the answer on screen rather than hear it described.

The single most revealing test takes ten minutes. Take a role you closed in the last six months, give the vendor the original job description and the original applicant pool, and see where the person you hired ranks. If your hire sits outside the top 15, the matcher is reading something different from what your team values, and no amount of tuning on a sales call will tell you that.

AI candidate matching is about as accurate as the brief it is given and the evidence it can see. On a clear brief with structured must-haves, semantic matching puts the people a recruiter would shortlist into the top of the list most of the time; on a vague brief it confidently ranks the wrong people. Vendor accuracy percentages are rarely comparable, because each vendor measures a different thing.

Measure it on outcomes you already track. The useful numbers are the share of the top ten that your hiring manager agrees to interview, the reply rate when you contact the top-ranked sourced candidates, and where your eventual hire ranked. If the manager rejects half the top ten, the problem is usually the brief rather than the model, and a good tool makes that visible because you can read which criterion drove each score.

Matching also gets worse in three predictable places. Very senior roles, where the decisive evidence is judgment and reputation that no resume captures. Roles that change shape during the search, where the brief stops describing what the team now wants. And thin profiles, where a strong candidate simply has not written down what they did. HireAgent flags low-evidence candidates instead of scoring them low, so a thin profile gets a human look rather than a silent rejection.

Yes. US employers can use AI candidate matching, but a tool that substantially drives who advances brings notice, audit and record-keeping duties in several places. The rules follow where the candidate is located, not where your company is headquartered.

New York City Local Law 144 requires an independent bias audit within the past year, a published summary of the results and candidate notice at least 10 business days before an automated employment decision tool is used for a New York City role. Illinois requires employers to notify applicants when AI is used in employment decisions and bars using zip code as a proxy for a protected class. California's automated-decision-system regulations under FEHA, in effect since 1 October 2025, prohibit discriminatory use and require four years of record retention. Title VII's adverse-impact test applies to any selection procedure, automated or not.

The practical consequence is that the score has to be explainable and the decision has to be human. HireAgent scores every candidate on the same structured criteria, stores the evidence and the criteria version for each score, and leaves every advance and reject to a person. That gives you the record an auditor or a candidate's lawyer would ask for. Our candidate screening software page carries the full state-by-state table.

Staffing and search firms buy matching to answer one question fast: who in the database fits the job order that just came in. An agency with 40,000 candidates in its ATS and a two-day window to submit three people needs the match to run against past candidates as well as new ones, and it needs the evidence written down, because the client will ask why each person was submitted. Our page on the AI recruiter for staffing agencies covers the submittal workflow in detail.

In-house talent teams usually buy matching to cope with volume. A single posted role at a US company with a recognizable brand can draw 300 to 1,000 applications, and nobody reads them all with the same attention. Matching puts the pool in order so the recruiter's reading time goes to the top 40. For roles that draw thousands, the high volume hiring software page covers the extra controls you need.

Startups and small teams have the opposite problem: too few applicants, not too many. For them matching is most useful pointed outward, at sourced candidates who have not applied, with the outreach drafted and the interview booked by the same tool. That is the job recruiting software for startups is built around.

Candidate matching software costs from about $15 per user a month for ranking the applicants already in an ATS, to $99 to $199 a seat for matching against outside profile databases, to $299 a month and up for agents that match, screen, contact and schedule. Several staffing platforms publish a price only for the plans that do not include matching.

HireAgent publishes every plan. Solo is $299 a month for one seat and up to three open roles, with outreach sent after your approval. Growth is $799 a month, or $666 a month billed annually, for three seats, up to ten open roles, autosend outreach and ATS integration with Greenhouse, Lever and Ashby. Scale is $1,999 a month for ten seats, higher volume and SSO. Every plan includes sourcing, matching, screening, ranking, outreach and scheduling.

The comparison that matters to most buyers is not another tool, it is the alternative way of filling the role. A 20% contingency fee on a $120,000 engineering hire is $24,000 for one placement. A year of Growth billed annually is $7,992 across every role you open. The AI recruiting software pricing page sets both against the rest of the market.