Skip to content
Back to Blog
AI ATS and Semantic Search

AI Candidate Matching: How The Match Score is Actually Built

A match score is a weighted rubric, not a keyword count. How HyreTech builds one, what moves it, and how to use it to defend a shortlist to a client.

Tenia NovaliaTenia Novalia
7 min read
Bright AI hiring dashboard showing a candidate match score of 76/100, with weighted skill criteria, individual scores, and supporting candidate profiles in the background.

AI candidate matching scores every applicant against the same written rubric, one criterion at a time, and combines those scores into a single match score. In HyreTech, each criterion gets a score from 1 to 100 with a written justification. The criteria carry weights that should add up to 100, and the match score is their weighted average. If you approve the role's deal-breakers, each one a candidate fails cuts the score to a fifth. The number is useful for sorting. The per-criterion reasoning is what lets a recruiter answer the client's question: why these five?

"Match score" gets used loosely. In some tools it is a keyword overlap percentage, and in others a similarity number between two blocks of text. Neither can tell you why one person ranks above another, and that is the question a client asks.

Below is how HyreTech builds the number, what moves it, and how to use it when you submit a shortlist.

How The Match Score is Built

Every role has a rubric: a handful of criteria and a weight for each. HyreTech drafts it from the job description as soon as the role is created. You edit it, and every applicant is scored against it.

Here is the arithmetic for one candidate on a senior backend role:

Criterion Weight Score Contribution
Ships end-to-end features 30% 80 24.0
Depth in our stack 25% 100 25.0
Works without a spec 20% 60 12.0
Production ownership 15% 60 9.0
Written communication 10% 60 6.0
Match score 76

Each of those five scores comes with a sentence or two explaining what in the resume earned it. How weights become a score covers the arithmetic in more detail.

A low score on a single criterion still shows when the total is high, and its reasoning says what the resume didn't show. It is a marker for a person to look at, not a filter.

Must-Haves, Deal-Breakers and Preferred Skills

A weighted average has one weakness: a strong score on four criteria can hide a hard requirement the candidate fails, such as location. So HyreTech lets you state the role's requirements explicitly.

When you regenerate a role's criteria, you review the requirements it extracted from the job description: required skills, preferred skills and deal-breakers. You approve them, which uses AI credits. After that, required and preferred skills shape the criteria HyreTech generates for the role, so a gap in them shows up as a lower criterion score. Only deal-breakers adjust the weighted average itself, and one more check applies to every role:

Signal Effect on the score (defaults)
Each failed deal-breaker (for example, location) × 0.2, with no floor
A resume flagged for possible prompt injection (any role) Capped one point below the Shortlist cut-off

Applied to the 76 above:

  • Missing one required skill (say, payments experience): no extra multiplier. The gap already counts through the score on that skill's criterion, so the result stays 76.
  • Matching three preferred skills: no bonus either. Preferred skills also count through their criterion, so the score stays 76.
  • Failing a location deal-breaker: 76 × 0.2 = 15.
  • Failing two deal-breakers: 76 × 0.2 × 0.2 = 3.

Until 7 October 2026, missing required skills and matched preferred skills moved the score too. We removed both after checking them against hiring outcomes on HyreTech roles: the criterion scores already reflect those skills, so the extra adjustments counted them twice and ranked candidates worse.

A deal-breaker has no floor because it is meant to be close to disqualifying. Each failed one is listed next to the score, for example "failed deal-breaker: based in Singapore", so you can see why the number moved. The injection cap covers resumes flagged for text that tries to give the AI instructions. Their score is capped one point below the Shortlist cut-off, so they can reach Consider but never Shortlist.

Skip the approval step and no deal-breaker applies: the match score is the plain weighted average, still subject to the injection cap.

Each role also has two cut-offs, set when its criteria are generated, that sort candidates into Shortlist, Consider and Reject. These are labels you filter on. HyreTech has no auto-reject: advancing or rejecting a candidate is always something a person does.

Match Score vs Keyword Match vs Reverse Matching

What it compares What it can explain
Keyword filter Words in the resume vs words in the job post Which terms appeared
Similarity search The meaning of the resume vs the meaning of the job post That two texts are alike, not why
Rubric match score Evidence in the resume vs each criterion you wrote A score and a reason per criterion
Reverse matching Talent-pool sign-ups vs a new role Who to look at first for the new role

HyreTech uses the last two together. The rubric score ranks applicants to one role. Reverse role matching runs the other way and re-matches your talent pool when a new role opens. Past applicants are not re-matched automatically; you find them with search, as covered in candidate rediscovery.

Using The Score to Defend a Shortlist

For a recruiter, the shortlist is the product, and the client is judging the reasoning behind it. With 141 applicants on a brief, the five you submit should be chosen after comparing all 141, not after running out of time around number 60. Scoring against one rubric makes that possible, because the applicant who arrived on the last day gets read the same way as the first.

When the client asks why candidate two ranked above candidate three, the answer is already written, criterion by criterion:

  • Candidate two: "Six years of Go, led the payments rewrite." Scored high on stack depth and on shipping.
  • Candidate three: "Strong distributed systems, lighter on payments." Scored high on stack depth, lower on the payments-domain criterion the client weighted highest.

Three habits make this hold up:

  1. Agree the rubric with the client before applicants arrive. HyreTech drafts the rubric the moment you create the role and starts scoring as soon as applicants come in. It does not wait for approval. Edit the criteria and weights on the kickoff call, then open the role. First scorecards walks through the order.
  2. Keep one rubric per search. A rubric belongs to a role and is never shared with another, so editing one client's criteria never rescores another client's applicants.
  3. Copy the reasoning into your submission. There is no client portal, so the per-criterion lines go into your shortlist email or deck. Editing the rubric by hand doesn't re-run the analysis for people who have already applied. Regenerating it with AI, or approving the role's requirements, recalculates their overall scores and labels from the per-criterion scores they already have, with no per-applicant charge and no new analysis. Re-analysing a candidate (1 credit) replaces their scores and reasoning, so send the shortlist from the version you agreed with the client.

One workspace holds every client's searches, and search works across all of them: a strong applicant from one client's search turns up when you search for another client's role. Talent-pool matching covers every Active role except those imported from a CSV, which aren't matched until you edit their description.

Roles you import from a CSV stay off the public HyreTech job board unless you list them. How this works for solo recruiters and for agencies is on HyreTech for freelance recruiters and HyreTech for recruiting firms.

What a Match Score Cannot Tell You

A match score measures the evidence in the resume against your criteria. If the evidence isn't written down, the score can't see it. A candidate who undersells a relevant project scores low on that criterion, and the reasoning will say the evidence is missing, which is your cue to ask.

For technical roles, a GitHub link in the resume adds evidence. HyreTech reads the profile's public repositories and its merged contributions to other projects, and weighs them as supporting evidence for the resume's claims. Scoring is set up not to count a missing profile against anyone.

A match score also doesn't check whether the resume is internally consistent. On Growth and above, a verification trust score flags timelines that don't add up, such as a technology listed years before it was released, and contradictions with the candidate's other resumes in your workspace. It is a prompt to check before a candidate goes to a client, not proof of anything.

One credit analyses one applicant. The Free plan includes a one-time 20 credits, and you can buy top-up credits on any plan. The Starter plan covers three active roles with 300 credits a month. Current rates are on the pricing page.

Start free and edit the drafted rubric before your first applicant arrives, or book a demo and we will run it on a brief you already have open.

FAQ

How does AI candidate matching work?

HyreTech scores each applicant from 1 to 100 on every criterion in the role's rubric, with a written justification, then combines those scores into one match score using weights that should add up to 100. If you approve the role's deal-breakers, each one a candidate fails multiplies the score by 0.2.

Is a match score the same as a keyword match?

No. A keyword match counts shared words between the resume and the job post. A rubric match score rates the evidence in the resume against each criterion you wrote, and explains each rating, so experience described in different words still counts.

Which platforms support exporting shortlists and writing decisions back to our ATS?

HyreTech does neither: there is no shortlist or candidate export, and nothing is written back to your ATS. Imports are one-way, by CSV. Every score carries per-criterion reasoning that you copy into the shortlist you send a client, and if your ATS stays the system of record, you enter the decision there yourself.

Does HyreTech reject candidates automatically?

No. Candidates are labelled Shortlist, Consider or Reject against the role's cut-offs, but nothing is rejected because of a score. A person always takes that action.

Does a missing required skill knock a candidate out?

No. A missing required skill lowers the score on the criterion that covers it, like any other gap in the evidence, and no extra multiplier applies. Only a failed deal-breaker, which multiplies the score by 0.2, comes close to disqualifying.

Ready to get started?

Hire smarter with AI-powered candidate screening — start free today.

Start free
AI Candidate Matching: How The Match Score is Built | HyreTech