What an applicant tracking system actually does
What the software does, what it does not do, and which of the advice you have read is folklore.
6 minute read
It is a database before it is a filter
An applicant tracking system is, first and mostly, a place to keep applications. It receives your file, extracts text from it, tries to identify a few fields — name, contact details, employers, dates — and stores the result so a recruiter can search and sort it later. The extraction is the part that concerns you, because everything downstream reads what it produced rather than what you sent.
The popular image of a robot scoring your resume and rejecting it before a human sees it is mostly wrong, and where it is right it is configured by the employer rather than built into the software. Most systems rank; they do not reject. The ranking is usually crude, and a recruiter still opens the file.
What actually goes wrong
The failures worth caring about are extraction failures, and they are boring: text that is not text, reading order that is ambiguous, and fields that are not where a parser looks.
- A resume exported as an image, or scanned. There is no text layer, so there is nothing to extract at all.
- Contact details in the page header or footer. Several parsers drop those regions before reading, which loses your email.
- A two-column layout. A parser that reads column-wise recovers your resume correctly; one that reads line-wise interleaves the two columns into nonsense. Which one you get is not up to you.
- Text inside a table, a text box, or a graphic. Same problem, less obviously.
- Dates written only as “2019–2021” beside an employer, with no month. Recoverable, but it costs the parser a guess it may get wrong.
You can check every one of these on your own file, in your own browser, at /check. It shows the text a parser extracts and which fields it recovered — not a score, and not a prediction.
What we will not tell you
We will not tell you a resume is guaranteed to pass. Those systems are private, they are configured per employer, and anyone quoting you a pass rate is guessing. We will not quote an interview-rate improvement either, because we have not run that study and neither has whoever you last read it from.
What is defensible is narrower: a single column of real text with standard section headings is the arrangement that the widest range of parsers reads correctly. That is a statement about software behaviour, and it is checkable.