The Best ATS Resume Format in 2026 — Written as a Spec, With the Numbers

Search "ATS resume format" and you get the same advice everywhere: keep it simple, use standard headings, avoid graphics. True, and useless, because it never says how simple, which headings, or what happens when you don't.

We wrote the rules an ATS checker scores against, so we can be specific. What follows is the format as a spec — what the file is, how the page is laid out, which sections appear in which order, how long each part is, and the three mistakes that cap your score regardless of everything else. Every threshold below is the one our engine uses; they were set from how parsers behave and how recruiters read, and they are the numbers your resume is measured against when you run the checker.

1. The file

PDF with a real text layer and embedded fonts. The text layer is the non-negotiable part: a resume that is a scan, a photo or an exported image has nothing for a parser to read. Our engine caps any such file at a score of 20 out of 100 and stops evaluating — there is no keyword match to compute. Check by opening your PDF and trying to select a word; if you cannot, it is an image.

DOCX when the posting asks for it. Some employer systems and most government portals still specify Word. A DOCX from a modern word processor parses well; the risk is that whoever opens it can edit it, and that fonts you used may not be on their machine.

File name: Firstname-Lastname-Resume.pdf. Not resume_final_v3 (2).pdf. Some systems display the file name to the recruiter.

2. The layout

One column. The reading order of a single column is unambiguous for every parser ever built. Two-column layouts work only when the document's underlying order is main column, then sidebar — the way a person reads — rather than row by row across both. Table-built and text-box-built columns usually read across, producing lines like "Experience Skills Senior Engineer Python" that no keyword matcher can recover. Our engine's single-column reading order and no table-based layout checks sit in the parseability dimension, which carries 25% of the total score in role mode.

Contact details in the body. Name, phone, email, city, LinkedIn — as the first lines of ordinary text, not in a page header or footer and not labelled with icons. Parsers frequently skip headers and footers; icons carry no text. A resume whose contact block is only in the header is capped at 55 in our engine, because the record it produces has no owner.

Margins 0.6–1 inch, body 10.5–11.5 pt. Wider margins waste the page; narrower ones crowd the printer's safe area.

No text boxes, no images that carry information, no skill bars. A chart showing "Python ████░░ 80%" extracts as nothing. Write the skill.

Standard, embedded fonts. Inter, Roboto, Calibri, Arial, Georgia, Merriweather. No script, handwriting or display faces in the body. The standard embedded fonts check fails on decorative families because they are the ones that extract as wrong characters.

3. The sections, in order

Use these headings, spelled this way. Parsers map headings to fields by matching against a list of canonical names; "Career Chronicle" maps to nothing.

1. Name and contact block — as above.
2. Headline — your target title, e.g. "Senior Data Analyst". The headline aligns with the target role check compares this line to the role you are scanning for; a headline that says something different from the job you want is a match you are giving away.
3. Summary — 40–120 words, ideally 60–90. Title, years, specialism, one quantified result. Not an objective, not a personality statement.
4. Skills — 10–14 items, grouped (Tools · Methods · Domain). Fewer than five and the keyword matcher has too little to work with; more than twenty reads as padding and dilutes the no keyword stuffing check. Every skill listed should appear somewhere in your experience text, because the skills backed by experience text check measures exactly that ratio.
5. Experience — reverse-chronological. Each role: title, employer, location, dates (MMM YYYY – MMM YYYY), then 3–5 bullets of 12–28 words each. Open every bullet with an action verb; put a number in as many as you honestly can. The engine scores the share of bullets that start with a verb and the share that contain a figure, and it penalises filler ("responsible for", "duties included") and first-person pronouns.
6. Education — degree, institution, year. Dated, because the education carries dates check expects it.
7. Certifications, Projects, Publications, Volunteering — only the ones that exist. One or two supporting sections lift the supporting sections check; empty headings hurt.

Standard names: Summary, Skills, Experience (or Work Experience), Education, Certifications, Projects. Not "Professional Journey", not "Toolbox", not "Where I've Been".

4. The lengths

These bands are graded, not binary: a 900-word resume loses a little, a 1,700-word one fails the check.

5. The three failures that cap everything

Most checks trade points against each other. Three do not — they set a ceiling on the whole score because they make the document unusable to the system:

1. No text layer (scan or image PDF) — capped at 20.
2. Under 60% of fields survive extraction — capped at 45. This is what a table-built two-column layout, heavy text-box use or a corrupted export produces: the parser gets text, but not enough of it lands in the right fields.
3. Contact details only in the header or footer — capped at 55.

Fix these before anything else. A resume with perfect bullets and a header-only contact block scores 55 at best.

6. Keywords: the part format cannot fix

Format gets you parsed. Keywords get you ranked. Once the record is clean, the engine compares your text against the expected vocabulary for the role — or, if you paste the job description, against the posting's own required and preferred terms, which carries 30% of the score in that mode. Use the posting's exact wording where it is true of you ("stakeholder management" if that is what they wrote, not "managing stakeholders"), put required terms in Skills and again in a bullet where you used them, and leave out anything you would be uncomfortable being interviewed on.

7. A one-page checklist

• PDF, text selectable, fonts embedded (or DOCX if asked)
• One column, or a two-column template whose document order is verified
• Contact block as body text, no icons, no header
• Standard headings in the standard order
• Summary 60–90 words with a result
• 10–14 grouped skills, each evidenced in a bullet
• Every role dated, reverse-chronological, 3–5 verb-led bullets of 12–28 words
• Education dated
• 350–850 words, 1–2 pages, no shrinking below 10.5 pt
• No unexplained gap over a year
• Scan it, fix the ranked list, scan again

Our templates are built to the layout half of this spec; the content half is yours. The checker tells you which lines above you have missed, in order of how much each one costs.