Why a resume scores low on ATS readability checks

What the score measures, and what it does not

The free check is deterministic: fixed rules over the text extracted from your file, with no AI model involved. The same document scored twice returns the same number. That is a deliberate design choice — a readability measurement should be reproducible rather than an opinion that varies between runs.

What it measures is whether your resume survives automated processing legibly: did the text extract, are there recognizable sections, is the formatting clean, do the bullets read like accomplishments, and — if you supplied a posting — does your text contain the terms that posting emphasises.

What it cannot measure is whether you are a good fit for a role. It has no model of the employer, the hiring manager, the other applicants, or the seniority bar. A high score means your document processes cleanly. It does not mean you will be shortlisted, and a low score does not mean you will not be.

The four components behind the number

The total is a weighted blend of four independent components, each scored 0–100 before weighting. Knowing which one is dragging is most of the diagnosis:

• Keyword match — how many of the terms extracted from the job posting appear in your resume.
• Sections — how many of the four recognized section categories were detected.
• Format — a 100-point budget reduced by specific, itemized deductions.
• Action verbs — how many bullet lines begin with a recognized accomplishment verb, plus a bonus for variety.

Pasting a job description changes the weighting

With a job description supplied, keyword match is weighted at 50%, sections at 20%, format at 15%, and action verbs at 15%. The comparison against the posting dominates, which is appropriate — that is the most role-specific evidence available.

With no job description, keyword matching is skipped entirely and its weight is redistributed: sections 40%, format 30%, action verbs 30%. The check becomes purely a document-quality assessment.

This has a consequence people find surprising. The same unchanged resume can score differently depending on whether you pasted a posting, and neither number is wrong — they answer different questions. If you are comparing two versions of your CV, compare them under the same conditions or the difference is not meaningful.

Exactly what reduces the format score

The format component starts at 100 and subtracts for each problem detected. These are the deductions, in full:

• Extracted text under 300 characters: −25. Almost always an extraction failure rather than a short resume.
• Extracted text over 8,000 characters: −10. Long enough that key content is likely buried.
• No email address and no phone number found anywhere in the text: −15.
• Fewer than two recognized section headings: −20.
• Lines longer than 120 characters: −5 each, capped at −15. Dense paragraph blocks rather than scannable bullets.
• Tab characters or runs of three or more spaces: −10. A common fingerprint of column or table layout surviving extraction.
• More than three pages: −15 — but only when the parser can report page count reliably. When it cannot, no page-count finding is produced rather than a guessed one.

The single most common cause of a very low score

If your score is far lower than the quality of your resume would suggest, the likeliest explanation is not that the scorer disagrees with your writing. It is that it never received your writing.

An image-only PDF — a scan, a photo, or a document flattened to an image on export — contains no extractable text. The file opens correctly and looks perfect, so nothing warns you. What reaches the scorer is a few characters of stray metadata, which triggers the 300-character penalty, finds no sections, finds no contact details, and finds no bullets to check for verbs. Every component collapses at once.

That signature is diagnostic. A score that is low across all four components, rather than weak in one, points at extraction rather than content. Before changing a single word, open your file and try to select a sentence with your cursor. If the text does not highlight, re-export from the original document and re-run the check.

Why a well-written resume can score mid-range

The action verb component is the one that most often surprises people who write well. It examines lines beginning with a bullet marker and checks whether each starts with a recognized accomplishment verb — built, led, designed, delivered, improved and similar. The score combines the proportion of bullets that qualify with a bonus for using a variety of verbs rather than repeating one.

Two things reduce it without reflecting weak content. Bullets written as noun phrases — "Responsible for the migration" rather than "Led the migration" — do not qualify. And a resume written as prose paragraphs with no bullet markers at all has no bullet lines to examine, so the check falls back to every line in the document, which almost never scores well.

The keyword component has its own quirk worth understanding. It extracts up to twenty of the most prominent terms from the posting and checks whether each appears in your text, matching on the exact term and on a punctuation-stripped form. That second pass is why "Node.js" matches "nodejs" — but a genuine synonym you never wrote will register as missing, because it is comparing text, not meaning.

Reading a missing-keyword list correctly

The missing list is the output people most often misuse. It reports terms from the posting that are absent from your resume — nothing more. It is not a list of things to paste in.

Split it into three groups. Terms describing work you have genuinely done but described differently: rewrite that line using the posting's wording, which is an accurate edit and the highest-value change available. Terms describing work you have not done: leave them out — an unsupportable keyword survives a text match and fails the first interview question about it. Terms that are noise, such as company boilerplate the extractor happened to rank highly: ignore them.

Chasing the list to 100% is the failure mode this design tries to avoid. A resume padded with unearned terms scores well and interviews badly, which is a worse outcome than a mid-range score on an honest document.

A diagnostic order that saves time

Work structural before textual. Fixing extraction is fast and moves every component at once; rewriting bullets is slow and moves one.

• Confirm the text extracted at all — select and copy a sentence from your file.
• Confirm your section headings sit alone on their own lines.
• Confirm an email address or phone number appears in the document body, not only in a header or footer region.
• Replace paragraph blocks with bulleted lines, and start each with a verb.
• Only then compare against a posting and revise wording where the change is truthful.
• Re-run under the same conditions as your first check, so the two numbers are comparable.