What an ATS can actually read from your resume

Why an ATS score is not an employer decision

The category collapses four different questions into one percentage, and the averaging is where the information dies.

Readable is not the same as relevant

Passing a parser test only tells you the document can be read. It does not tell you whether the experience inside it makes you competitive for this job.

This is where we think the category has it backwards, and where we will argue plainly for our own method. Keyword overlap can be raised by anyone in an afternoon, which is exactly why it predicts so little — a resume can contain every phrase in the posting and still be a weak application, and a resume with lower overlap can carry far stronger evidence. RoleSprint separates machine readability from job evidence instead of hiding both inside one opaque score, and then goes further than any parser can: it reads the real posting against the real evidence in your resume and tells you whether the role is worth your effort before you spend it. That last decision is the economically significant one, and it is the one an ATS percentage is standing in for, badly.

What nobody can tell you from outside

We would rather be useful about the limits than confident about things we cannot see.

• Whether a specific employer auto-rejects anyone. Screening rules are configured per employer and per job. The software makes it possible; the employer decides whether to use it.
• What knockout questions a posting uses. Those are application-form questions the employer writes — work authorisation, location, licences — and they filter far more decisively than resume wording. They are also invisible until you reach the form.
• Which parser an employer runs, or how it is tuned. Different systems and settings extract the same file differently, which is why we describe our result as one honest sample rather than a verdict.
• Whether a human will read it. That depends on volume, recruiter workload, and process, none of which is a property of your document.
• Whether a particular employer uses AI matching at all. Greenhouse, for instance, documents its Talent Matching as a paid add-on the employer enables, calibrated with criteria and weights the hiring team sets themselves — and states plainly that it does not automatically advance or reject candidates. None of that configuration is visible from outside, which is exactly why an external match percentage cannot mean anything.
• What a status label means at a specific company. Statuses are employer-configured, so the widely-published tables of what each one supposedly means are describing something that is not fixed. We write up what the vendors themselves document instead, for Workday and for Greenhouse.

How this test works

Everything on this page is mechanically derived from your file, with no model involved, so you can check any of it yourself.

• Your file is uploaded to our server, its text is extracted in memory with the same libraries the product uses elsewhere (unpdf for PDF, mammoth for DOCX), and the file is discarded. It is never written to storage and never logged.
• Field detection is pattern matching over the extracted text: an email, a phone-shaped string, a link, and date ranges in the common formats. The name line is explicitly a guess from position and shape, and is labelled as one.
• Format risks are fingerprints, not layout inspection. We do not have your document's internal structure, so we report what a layout does to extracted text — heavy line fragmentation, column-like spacing, characters that failed to decode — and say what each pattern usually means.
• The two readings we show, machine readability and structure risk, are OUR coarse bands over those findings. They are not an employer's ATS score and are never combined into one number.
• With a job description supplied, the requirement language check is literal word presence and nothing more. Finding a phrase proves the phrase is present, not that you meet the requirement. It is shown unrolled on purpose.
• This is RoleSprint's own diagnostic. It is not Workday's parser, Greenhouse's parser, Lever's parser, or whatever your employer has configured, and we do not have access to any of those. That is not a weakness to apologise for — it is the reason we only claim what this diagnostic can actually observe, while a percentage implies knowledge of a configuration nobody outside the company has.
• PDF extraction often returns text with no line breaks at all. Where that happens we say so and skip the checks that read line shape, rather than reporting them as passed.

For what the vendors themselves publish, we have written up Workday and Greenhouse from their own documentation, with sources cited.

Questions about ATS and resumes

What makes a resume ATS-friendly?

Text that can actually be extracted, and a structure that survives extraction. In practice: real selectable text rather than an image or a scan, a layout that does not scramble when flattened into a single stream, standard section headings, and date ranges written so they stay attached to the right role. Those are testable properties of your file. Everything beyond that is the employer's configuration, not your document.

Does an ATS automatically reject resumes?

It can, but whether it does is the employer's decision, not the software's default. The filters that reject most decisively are usually knockout questions on the application form — work authorisation, location, licences — rather than anything about resume wording. Nobody outside the company can tell you how a given posting is configured, and any tool that claims to is guessing.

Is an ATS score worth anything?

Not much on its own, and it is not an employer's decision. A score generally blends how readable your document is with how many words it shares with the posting, and those are different things with different levels of certainty. Keyword overlap in particular is easy to raise and predicts little: a resume can echo every phrase in the posting and still be a weak application. We show the readability and the literal word overlap separately, labelled, rather than averaging them into a number that looks precise and is not.

Should I add keywords from the job description?

Adding words you cannot back up is a bad trade: it may raise a keyword count, and it makes the interview harder because you now have to substantiate claims your experience does not support. If a requirement is genuinely part of your history and simply is not written down, adding it is worth doing. If it is not part of your history, the useful question is whether this role is the right target at all.

Does this tool store my resume?

No. The file is read in memory to extract its text and then discarded. It is not written to storage, not logged, and nothing from it is sent to analytics — the events we record contain only coarse bands such as whether readability came out strong, mixed or weak.

Why does this not give me a single ATS score?

Because the number would be false precision. Different employers run different systems with different settings, and the strongest filters are configured per job and invisible from outside. A single percentage implies a universal answer to a question that has no universal answer. Showing the separate, checkable findings is less satisfying and considerably more useful.