How ATS actually works

Does the Recruiter See Your Match Score?

Almost never, and where a score does exist it belongs to the recruiter who calibrated it. What that means for how much time to spend chasing a percentage.

Somewhere in your browser history is a tool that told you your resume was a 62% match. Perhaps it was 41% and you spent an evening fixing it. The question nobody asks about that number is the only one that matters: who else can see it?

Nobody. It was calculated by the tool you pasted the posting into, it lives in that tool, and there is no field in the employer's system that could receive it even if somebody wanted to send it. This is true of our score too, and it is worth saying plainly before anything else.

What actually reaches the employer

When you apply, the employer's system receives your form answers and your file. That is the whole payload. The file gets converted into fields: Oracle publishes Taleo's list, and it runs to name, contact details, education level, institution, programme, dates, and per job the start date, end date, employer, job function and responsibility. Greenhouse lists what its matching feature extracts: skills, job titles, years of experience, employment start and end dates, company names, and a derived industry classification.

No percentage crosses that boundary. What the employer's side does next is up to the employer's side.

The systems that show no score at all

For most of the pipeline, the honest answer is that there is no score to see.

Lever's documented candidate search looks over "candidate names and any parseable content attached to candidate profiles, including resumes, notes, and feedback forms" and sorts results by last interaction date, most recent first. Ashby's candidate search is a boolean full-text search of the resume, with modes for exact match, exact phrase, exact capitalisation, and word variants. Ashby's application review lets the reviewer choose to start with the newest application or the oldest. Greenhouse's keyword search returns matching candidates with a snippet showing how the terms are used in the profile.

Date order. Text retrieval. A snippet a person reads. Across four vendors' own documentation there is no ranking of every applicant by fit, because that is not what these screens are for.

The exception, and it is a real one

A lot of career advice, including things written a year ago, states flatly that recruiters never see a score. That is now too strong, and pretending otherwise would be exactly the folklore this cluster exists to correct.

Greenhouse documents a feature called Talent Matching, part of its Real Talent add-on. A recruiter builds a calibration, described as "a structured list of job-related qualifications and weights", applies it to a pipeline, and "each candidate receives a match score based on how closely their resume aligns with the recruiter's calibration". Recruiters see resume highlights, a short justification, and a list of matched and missing skills.

Read the constraints Greenhouse publishes alongside it, because they are the interesting part.

Workday, on its own marketing pages, says that "many platforms rank candidates by match percentage based on the job description, though they don't just count matching words". That is a vendor describing a category rather than product documentation of a specific screen, and we would not build advice on it. We include it because it is the strongest published claim in the other direction and you should see it.

So the honest summary is three sentences

The score you were shown belongs to the tool that showed it. Most employer screens have no score in them at all and work on search and date order. Where a score does exist, a specific recruiter defined its criteria and weights, it cannot reject you on its own, and it stops mattering the moment a human opens the file.

What follows for how you spend an evening

The reason this is not academic: a percentage is addictive to optimise because it moves when you type. Three hours of typing can lift a number that nobody will read. Meanwhile the four things that decide the outcome sit untouched.

1. The form answers. The only stage that rejects automatically. Greenhouse's own Auto-Reject examples are licence and location requirements.
2. Whether your file parses. If the fields do not extract, you are not in the search result, and in a scored pipeline you are shunted to a manual-review pile instead of an ordered one.
3. Whether your page uses the words a recruiter would type. Exact terms are what a boolean search finds. Ashby's "Similar" mode is available, not compulsory.
4. Whether page one, first third, says who you are. Every scored or unscored path ends with a person reading fast.

The calibration problem, in one example

Two recruiters hiring the same job title will build different calibrations. One weights the industry heavily because the last three hires failed on domain knowledge. The other weights a specific tool because the team has a migration running. Same posting, same words in the advert, two different orderings of the same pile of applicants.

There is no document you can write that is optimal for both, and there is no percentage a third-party tool can compute that predicts either. What you can do is make sure the facts a calibration is built from are present and readable: the job titles, the years, the employers, the dates, the named skills. Greenhouse's own list of what its matching step extracts is that list. Everything a weighting can act on has to survive parsing first, which puts the parse back at the centre of the problem.

This is also why "my score went up when I added the word" is a weak signal. It went up in the tool that scores. Whether it moves anything on the employer's side depends on a calibration you cannot see, in a feature the employer may not have bought.

Optimising past the point where a search finds you produces a document that reads worse to the human who then opens it, which is a bad trade at any score.

Why we still show you a number, and why it is two

We do calculate scores, so it would be convenient to skip this part. Involve Resume shows Fit, which asks whether this job matches your background, and CV readiness, which asks whether the document as written survives the screen. They are never averaged, and the reason is the behaviour a single number produces.

A blended percentage can be raised by typing. Split in two, the diagnosis is unavoidable. Low readiness with decent fit is an evening's work on a document. Low fit is a signal to apply somewhere else, and no rewriting changes it. The full reasoning is here.

Neither number is sent anywhere. They exist so you can decide where to spend an hour, which is the only thing a score is honestly good for.

A short test for any tool that scores you

Ask it these four questions. Ours answers them the same way.

1. Who computed this number? If the answer is not "the employer", it is advisory.
2. What is it computed against? A posting you pasted, or a calibration a recruiter built. Those are different things.
3. Can it reject anyone? If the tool is on your side of the wall, no.
4. What would change if the number were ten points higher? If the honest answer is "the wording", the number is measuring your wording.

Run one real posting through the two-score view and look at what it says about the document rather than at the number on top of it.

Sources: Greenhouse, Operational readiness guide: Talent Matching policy, Greenhouse, Talent Matching Data Processing FAQ, Greenhouse, Search resumes for keywords, Ashby, Candidate Search, Ashby, Application Review, Lever, Searching the database for candidates, Oracle Taleo, Candidate Management, Workday, What is an applicant tracking system?