What Resume Keywords Actually Match Against

When a resume enters an applicant tracking system, the first substantive action the system takes is not a ranking — it is a parsing pass that converts the document into discrete, searchable fields. Only after that conversion does any form of keyword comparison begin. The matching step is therefore downstream of, and dependent on, how cleanly the system could read the file in the first place.

What the matching actually compares against is a specific set of terms derived from the job requisition — not a general sense of the candidate's background, not the resume as a whole document, and not inferred meaning. Understanding the source of those terms, and the rules that govern the comparison, explains why two resumes with similar experience can receive very different scores from the same automated filter.

How Resume Matching Actually Operates, Step by Step

The process begins at the requisition level. When a recruiter or hiring manager creates a job posting inside an applicant tracking system, the system indexes the language in that posting — job title, required skills, preferred qualifications, and sometimes specific credentials or years of experience. That indexed language becomes the reference corpus against which incoming resumes are compared.

Once a resume is submitted, the system's parser attempts to extract structured data: contact information, employment history, job titles held, dates, education, and a pool of skill or keyword tokens. The quality of that extraction depends heavily on the file format and the document's layout. A resume with unconventional formatting, tables, or text embedded in graphics may yield an incomplete token pool, which directly reduces the number of potential matches before any comparison has occurred. The mechanics of how a system extracts text from a submitted file are a prerequisite to understanding why keyword matching produces the results it does.

After parsing, the system performs the match itself. The most common approach is a form of term frequency comparison: the system counts how many of the requisition's indexed terms appear in the resume's extracted text, then expresses that count as a proportion of the total indexed terms — producing what is often called a match score. Some systems apply weighting, giving more credit to terms that appear in the required-qualifications section of the posting than to those in the preferred-qualifications section. Others treat all indexed terms equally.

Synonym handling varies significantly across systems. One platform may recognize that "machine learning" and "ML" refer to the same concept; another may count them as two distinct tokens, crediting a resume only if both exact strings appear. Similarly, stemming — the process of treating "manage," "managed," and "management" as variants of the same root — is present in some systems and absent in others. A resume that uses one grammatical form of a term while the posting uses another may receive no credit for that term in a system without stemming.

The resulting score is what surfaces in the recruiter's queue. What that score actually measures is the degree of lexical overlap between two documents — not demonstrated competence, not the quality of the candidate's experience, and not cultural fit. The score is an artifact of text comparison, and it functions as a triage signal rather than an evaluative judgment.

Roles and Systems Involved in the Matching Process

The applicant tracking system. This is the software layer that receives submissions, executes parsing, runs the comparison algorithm, and surfaces ranked or scored results to human reviewers. The system does not make hiring decisions; it orders information for the humans who do.

The recruiter or sourcing specialist. This role typically writes or approves the job posting whose language becomes the matching corpus. The recruiter also sets any threshold scores used to filter the applicant pool — for example, deciding that only resumes above a certain match percentage are surfaced for manual review. The recruiter's word choices in the posting therefore directly determine what the system rewards.

The hiring manager. In many organizations, the hiring manager contributes the technical requirements section of the posting, specifying tools, certifications, or domain knowledge. Because those terms are indexed alongside the recruiter's language, the hiring manager's vocabulary choices influence matching outcomes even though the hiring manager rarely interacts with the ATS directly during the screening phase.

The job posting itself. The posting is not a passive document — it functions as the matching template. Vague postings with few specific terms produce low-discrimination match scores; postings with dense, specific technical language produce higher discrimination but may also exclude resumes that describe equivalent skills in different terminology.

The resume document. The submitted file is the input to the parser. Its format, structure, and vocabulary determine the token pool the system has available to compare. A resume that uses the same terminology as the posting — without any other change to the underlying qualifications described — will score higher than one that describes the same experience in different words. The structural choices in a resume file affect how much of its content the system can actually read and index.

Where Keyword Matching Breaks Down or Produces Unexpected Results

Parsing failures upstream of matching. If the system cannot extract text from a section of the resume — because the content is inside a graphic, a header or footer region the parser ignores, or a multi-column layout that scrambles reading order — those terms never enter the token pool. A candidate whose most relevant experience appears in a visually formatted sidebar may receive a lower match score than one whose equivalent experience appears in plain body text, even if the underlying qualifications are identical.

Terminology divergence between industries or regions. The same role carries different titles in different sectors. A posting written by a recruiter from one industry vertical may use terminology that candidates trained in an adjacent vertical do not use, even when the work is functionally identical. The matching system has no mechanism to bridge that gap unless synonym libraries are explicitly configured — and most organizations do not audit those libraries regularly.

Over-indexed postings. Some postings are written by aggregating language from multiple prior job descriptions, producing a requirements list that is broader than the actual role. When the system indexes that language, it creates a large matching corpus. Resumes that happen to contain many of the aggregated terms — even if those terms represent peripheral skills — may score higher than resumes focused precisely on the core competencies the hiring manager actually needs.

Threshold effects at the filter boundary. When a recruiter sets a minimum match score to advance to manual review, the boundary is a hard cutoff. A resume scoring one point below the threshold receives the same outcome as one scoring far below it — neither is surfaced. The score difference between advancing and not advancing can be a single term, and the candidate has no visibility into where the threshold was set or which term was absent.

Score inflation from keyword density without context. A system counting term frequency does not evaluate the context in which a term appears. A resume that lists a skill in a summary section, a job description, and an education section may receive more credit for that term than one that describes a deeper but less repeatedly mentioned application of the same skill. Density and depth are not the same thing, and the matching algorithm cannot distinguish them.

What the ATS Record Shows — and What It Does Not

The record an applicant tracking system retains after the matching step typically includes the submitted resume file, the parsed data fields extracted from it, the match score assigned at the time of submission, and a timestamp. In systems that log recruiter activity, there may also be a record of whether the resume was opened for manual review and when.

What the record does not show is the reasoning behind the score. The log captures the output of the algorithm — the number — but not a breakdown of which terms matched, which were absent, or how weighting was applied. A candidate reviewing their own application status through a portal, where that option exists, sees a disposition code (received, under review, not selected) but not the score or its components.

The record also does not capture what happened to the posting's indexed terms after submission. If a recruiter edits the job posting after applications have been received — adding or removing required skills, changing the job title — the match scores already calculated may not be recalculated against the revised posting. The record reflects the state of the comparison at a specific moment, not an ongoing or updated assessment.

From a compliance standpoint, applicant tracking systems are required to retain records of applicants and hiring decisions under federal recordkeeping rules. The Equal Employment Opportunity Commission and the Department of Labor both have regulations governing how long those records must be kept and what they must contain. However, those requirements address the existence of records, not their interpretive completeness. A retained match score is a number in a log; it is not a documented rationale for a hiring decision.

Resume keyword matching is a text-comparison operation built on top of a parsing operation, and its outputs reflect the precision — or imprecision — of both steps. The score that determines whether a resume reaches a human reviewer is a measure of lexical overlap between two documents written by different parties, in different contexts, with no shared vocabulary standard. That is the mechanism the first filter runs on.

Note: This explains how hiring works as a system. It is not career coaching or legal advice, and it is not a substitute for a career professional or employment attorney. Check the cited sources for current labor-market and employment-law data.