What ATS keyword matching actually does

Three different things get called "keywords". Only one of them is worth acting on, and it is the least discussed.

By Vitor Aguiar·Published 13 September 2026

“Optimise your resume for the ATS” is an industry. It sells scores, percentages, rewriting services and subscriptions, and it rests on a claim that is partly true — which is what makes it hard to argue with.

The useful move is to separate three things that all get called keyword matching, because they have very different evidence behind them.

One: literal string matching. Real.

Recruiters using an applicant tracking system can search and filter the candidate pool by terms. If a role requires Sage, someone types “Sage” and gets the subset of resumes containing it. This is a database query, it works the way you would expect, and it is the mechanism people are actually describing when they say a resume was “filtered out”.

The consequence is narrow and concrete: a term you never wrote cannot be found. If you spent two years using a system and left it off the page because it felt obvious, you are absent from that search.

The corollary is the part most advice skips. Matching is on strings, not meaning. These are different:

• BLS and Basic Life Support
• QA and quality assurance
• Postgres and PostgreSQL
• M365 and Microsoft 365

You do not know which form the employer searched. Writing both costs a few words and removes the guess — the registered nurse example shows what that looks like on a page where every certification has two names.

Two: a score the system gives your resume. Mostly invented.

This is where the industry lives, and it does not survive contact with how these products work.

There is no single ATS. There are dozens of unrelated products with different features, configured differently by every employer that buys them. Some offer a match percentage against a requisition. Many do not. Where it exists, the threshold is set by the employer, its formula is not published, and it is frequently ignored by the person doing the screening.

So when a tool tells you your resume scores 72% and will be rejected below 80, ask what it is measuring. It cannot know which system the employer uses, how it was configured, or whether anyone looks at the number. It is comparing your text to a job description using its own formula and presenting that as a verdict from a system it has never seen.

Where those tools are genuinely useful is duller: they show you which terms from the posting are missing from your resume. That is a checklist, and a checklist is worth something. It is just not a score.

Three: semantic and AI matching. Emerging, and not yours to optimise.

Newer products do match on meaning rather than exact strings, and some rank candidates with a model. This is real and growing.

It is also, from your side of the screen, unactionable. You cannot inspect the model, you do not know which employers use one, and the advice that would follow — “write clearly about what you did” — is what you should be doing anyway. Any guidance more specific than that is someone guessing at a system they cannot see.

What to actually do

Read the posting and use its vocabulary where it is honestly yours. If they say “stakeholder management” and you do that, write that. This is not gaming anything; it is describing your work in the language of the person reading it.

Write both forms of anything abbreviated. Once each, naturally placed.

Name systems, tools and standards explicitly. These are the terms that get searched, and they are the easiest to leave off because they feel like background.

Put the terms in your actual experience, not in a list. A skills block of twenty words carries far less weight than the same term inside a bullet that says what you did with it — and it reads as padding to the human who arrives second.

What not to do

Do not paste the job description in white text, or hide keywords behind an image, or shrink them to one point. Parsers extract text regardless of colour or size, so the hidden block arrives in full, visibly out of place. It is detected easily and it is read as exactly what it is.

Do not keyword-stuff a skills section. Forty terms make every one of them less credible, and interviewers ask about the ones you listed.

Do not rewrite a true sentence into a false one to match a phrase. The word gets you found; the interview is where the claim is tested, and the gap shows up in under a minute.

The honest summary

Keyword matching deserves about ten minutes of your attention per application: read the posting, check that terms genuinely true of you appear somewhere in your experience, expand your abbreviations. That is the whole of the actionable advice.

The rest of the time is better spent on whether the file parses at all, which is a mechanical question with a checkable answer — see how to check if your resume is ATS-friendly.