ATS-Friendly Resume: What the Software Actually Reads

A resume is ATS-friendly when an applicant tracking system can read the file and put each piece of it into the right field. That is the whole job. Not a score and not a pass mark. A copy operation from your document into a database. Most advice about the ATS-friendly resume gets the rules roughly right and the reasons wrong, which is why nobody can tell you which rules are safe to bend. So this post sources every rule to the systems themselves: five applicant tracking vendors' own documentation, one resume-parsing company's published measurements, and a peer-reviewed benchmark that scores 50 resume templates one by one. Short version: if you send a PDF or a Word file your file type is not the problem, your file size and your contact block sometimes are, and the layout rule everyone repeats turns out to be weaker than it sounds.

What "ATS-friendly" actually means

An applicant tracking system does two separate things with your application, and only one of them touches your resume file.

The first is parsing. The software opens your document, pulls the text out, and tries to work out which piece of text is your name, which is your phone number, which is an employer, which is a job title. Lever's help centre describes its parser as extracting "their name, organization, and contact details" and writing them onto the candidate profile. That is the entire ambition. Nothing in this step reads your resume for quality.

The second is screening, which mostly happens on the application form rather than in the file, and which is what an applicant tracking system actually does with your application once you have submitted it.

Here is the part almost every guide gets wrong. When parsing fails, Greenhouse's own documentation says the formatting problems behind it "may result in a partial resume parse that will need to be manually corrected and verified." A half-filled profile that a recruiter finishes by hand. Your resume is still attached and still readable. A bad parse costs you the automatic version of your profile and a bit of the recruiter's patience. Greenhouse describes a correction step, not a discard rule.

That changes what you should optimize for. You are not trying to pass a test. You are trying to make a copy operation easy.

Which file types every ATS accepts (and the size limit nobody mentions)

Five vendors publish exactly what they take, in their own documentation:

Applicant tracking system	Accepted for a resume	Documented size number
Greenhouse	.doc, .docx, .pdf, .rtf, .txt	100 MB upload, 2.5 MB parse
Lever	Word (.docx), PDF, RTF, WordPerfect, HTML, MS Office HTML, ODF	not published
Workable	.pdf, .doc, .docx, .rtf, .html, .odt	5 MB
Breezy HR	doc, docx, pages, pdf, rtf, txt	50 MB to a candidate profile
SmartRecruiters (partner application API)	PDF, DOC(X), RTF, JPG, PNG	2 MB per file

Line those five lists up and only three formats clear all of them: PDF, DOC or DOCX, and RTF. Plain text passes two of the five. OpenDocument passes two. HTML passes two. Apple Pages passes one. Nobody sends an RTF, so the practical answer is PDF or Word, and the argument about which one is better is settled by the vendors, because every one of the five takes both. Spend the energy somewhere else.

Now the number that matters and that nothing on page one of Google mentions. Greenhouse accepts candidate uploads "up to 100 MB." Greenhouse also states, in a different article in the same help centre, that it "can't parse resumes larger than 2.5MB," and names the cause: "if a candidate has an especially large file (for example, if their resume contains a lot of high-quality images), it may be over this file limit."

Read those two sentences together. The upload limit is forty times the parse limit. A resume with a headshot and a few high-resolution graphics can clear the upload and still land on the recruiter's side as an attachment with none of the fields filled in, and nothing in the documentation says the candidate hears about it. SmartRecruiters' partner API, the route job boards use to push applications in, documents an even tighter 2 MB per file. Keep your resume under 2 MB and you are inside every limit these five vendors publish. Right-click the file and look at its size. That is the whole check.

One more distinction worth holding onto, because it is where the worst failures live. Accepted and parseable are two different lists. SmartRecruiters accepts JPG and PNG attachments. Lever states flatly that it "cannot parse information from image files (like JPG or PNG)." A screenshot of your resume, or a PDF exported as a picture rather than as text, is a perfectly valid upload and an empty document to the parser.

Are two-column resumes ATS-friendly?

This is the most-asked version of the format question, and two published tests of it reach opposite conclusions.

Enhancv ran one resume through 17 of its own templates and 7 different extraction tools, 357 tests in all, and concluded that "two-column resumes lose almost no text during extraction in every test we ran, so the core myth doesn't hold up." ATSVerification ran one resume through six layouts and found that "the two-column layout was the only one of six to draw a critical flag, and the only one to lose points, dropping from a clean 100 to 85." Both are published by companies that sell into the resume market. Both are worth reading.

They do not actually disagree about what happens. ATSVerification: "The text is there; the order is jumbled." Enhancv: "The real risk is scrambled reading order." They disagree about whether that matters, and neither cites the company whose day job is answering exactly this.

Textkernel parses CVs for a living. It estimates that "at least 15% of CV documents use a column layout," and describes the mechanism plainly: a system reading left to right and top down across a two-column page produces a rendering where "the information from different sections of the CV is mixed together." Then the line that settles the argument: "An imperfect text rendering can still be useful for certain tasks: searching for keywords is still possible, and humans can still easily read the document. But when automated systems try to extract structured information from an imperfect rendering, problems compound very quickly."

Keyword search survives a scrambled render. Field extraction does not. That is why one test can say "no text was lost" and another can say "critical flag," and both be right.

Textkernel then published what fixing it was worth. After replacing its rule-based column detector with a trained model, correct column-separator decisions went from 60% to 82%, well-rendered CVs across about 700 documents went from 62% to 90%, and across more than 12,000 CVs, "the contact information fill rates are increasing by 4% to 10% absolute." Those are one vendor's before-and-after figures for its own parser, not an industry average. But they name the casualty precisely: in column resumes the old system was failing on, the parser "would sometimes extract less information, in particular contact information like name, phone, and address."

Greenhouse belongs here too, on the other side of the ledger. Its parse-failure list names "resumes that have a columned layout" outright, so this is not only folklore. The consequence it attaches is the same partial parse as everything else on that list.

So the risk is real, it is not rejection, and it lands in one place more than anywhere else. Keep the sidebar if you love it. Keep your name, phone and email out of it.

The worst template in the benchmark was single-column

ResumeBench, published at EMNLP 2025, scored 50 resume templates one at a time: 2,500 synthetic resumes across those templates, 30 career fields and 5 languages, read by 24 different language models. It measures models reading resumes, not one named commercial ATS, so treat it as a ranking of layout difficulty rather than a rejection rate.

Most of the 50 templates clustered near the average. Two did not, and the identity of the loser is the interesting part.

The worst performer was Template 7, at 12.23% below the benchmark average. The paper describes it as "a single-column layout with numerous sections and a tendency to include deeply nested experience entries." Second worst was Template 48, at 11.15% below average, "a two-column layout with an approximate 3:7 ratio between the left and right columns," with "dense personal information and language details" packed into the narrow left column. The paper's own summary of the pattern: "deeply nested single-column and asymmetric two-column formats degrade performance."

Neither failure is about having a column. One is about nesting your roles four levels deep and running a dozen sections. The other is about cramming a narrow sidebar so full that it stops looking like a column and starts looking like noise.

Single column is still the safer default, because it removes the reading-order question before it can be asked. It is not a guarantee. A single-column resume with sub-bullets under sub-bullets under every role is the exact shape the benchmark punished hardest.

And the way it fails is familiar by now. The paper's error taxonomy describes valid output with "structural misalignment (e.g., incorrect nesting, cross-role conflation) or semantic omissions." Smaller models collapsed the layout, merging bullet points across roles and dropping one role's description entirely. Words survive. Structure does not.

The three fields that actually break

Sources that do not cite each other converge on the same short list.

Your contact block. Textkernel's fill-rate measurement named contact information as the field that moved. Greenhouse lists "resumes with the name and contact information in the header, footer, or text box" as a cause of parse failure. Lever's parser writes name and contact details straight onto the candidate profile, which is what makes these the fields with the most to lose. Put your name, phone, email and city as plain text in the body of the document, at the top, outside any header, footer, sidebar or text box. If you change one thing after reading this, change that.

Your employer names. Greenhouse lists "company names that don't include identifying words such as Inc., Co., LTD, or LLC" as one cause of a failed parse. Enhancv's test found employer names were the single field that AI-based parsing did not repair. Neither says every company name needs a suffix. Both say this field is the fragile one, and writing "Acme Analytics, Inc." once instead of "Acme" costs you nothing.

Your job titles. Greenhouse's example is verbatim: "Sr. Account Exec" instead of "Senior Account Executive." Spell the title out. Save the abbreviation for conversation.

Notice what is missing from that list. Fonts and column count are where most of the advice goes, and neither of them is on that list.

An ATS-friendly resume format, rule by rule

Here is every common rule with the evidence behind it, including the ones that do not have any. The last column is the honest part.

Rule	What it protects	Documented in	How firm
Contact details in the body, not the header or a text box	Name, phone, address fields	Greenhouse; Textkernel measured it	Documented and measured
Real selectable text, never an image or scan	Everything; an image parses to nothing	Lever; Greenhouse	Documented
Keep the file under 2 MB	The parse running at all	Greenhouse (2.5 MB), SmartRecruiters (2 MB)	Documented
Send PDF or .docx	Acceptance at upload	All five vendors above	Documented
Spell out employer names and job titles	Employer and title fields	Greenhouse; Enhancv	Documented
Clear, conventional section headings	Section boundaries	Greenhouse lists unclear sections as a cause	Documented
Avoid tables and text boxes	Reading order and field assignment	Greenhouse; MIT CAPD	Documented
Avoid headers and footers	Contact block, mainly	Greenhouse	Documented
No letter-spacing tricks to fill a line	Word boundaries	Greenhouse lists "a resume with spaces between the letters"	Documented
One column, or a sidebar with nothing critical in it	Reading order	Textkernel, ATSVerification, Enhancv, ResumeBench	Measured, and weaker than you have been told
Shallow structure: few sections, no deep nesting	Role hierarchy	ResumeBench	Measured
Common fonts, 10pt or larger	Character fidelity	MIT CAPD	Plausible, but no ATS vendor documents it

PDF or Word: which file format for an ATS resume?

Either. Every vendor in the table above accepts both, which is as close to a settled question as this subject gets. Send a PDF if you want the layout to survive being opened on someone else's machine, send a .docx if the posting asks for one, and check that the text is selectable either way. The formats that actually carry risk are the ones only some vendors take: plain text, OpenDocument, HTML and Apple Pages.

ATS-friendly fonts

MIT's careers office gives a reason where most pages give a list: "If the font is not commonly used, the ATS software may try to convert the font and inadvertently delete or ignore critical information." That is a real mechanism, and it is more than most pages offer, but no ATS vendor documents it, which is why the table marks it plausible rather than documented. Use Arial, Calibri, Georgia or Times anyway, at 10pt or larger. It costs nothing and it reads better.

What a template cannot fix

A template does not choose your export settings, your file size, how you write your employer names, or whether you dragged your phone number into the header because it looked tidier there. Those are the parts you own no matter which template you start from.

How to check your resume yourself, in two minutes

1. The highlight test. Open your resume and try to select the text with your cursor. Lever's help centre gives this one directly: "If you cannot highlight text, the document is likely not parseable." An image-based PDF fails here instantly.
2. The plain-text test. Select all, copy, paste into any plain text editor, and read it top to bottom. That is close to what a naive extractor hands the parser. Your name and contact details should come first, and no sidebar item should have landed in the middle of a job description.
3. The size check. Look at the file size. Under 2 MB and you have cleared every documented limit above. If it is over, look at the images first: high-resolution graphics are the cause Greenhouse names.
4. The employer-name check. In that pasted text, find each company you have worked for and ask whether it reads as a company name on its own.

Four checks, no tool, no upload. Then keep an eye on what changes. If you rewrite the top of your resume, it helps to track which versions of your resume get replies rather than guessing.

What is not a formatting problem

Screening questions are not a formatting problem. Work authorization, location, notice period and the rest are answered on the application form, and that is where the documented reject rules live.

Keywords are not a formatting problem either. Getting the right skills and phrasing into the document is a content job, and a different one. Tailor the content to the job description posting by posting.

If you would rather not rebuild the file yourself, JobMason has 12 resume templates and the application tracker, both free with no card and no expiry. The AI generation and PDF export are the paid part, so you can put a resume together and track where you sent it without paying for anything.

Frequently asked questions

What does it mean for a resume to be ATS friendly? It means an applicant tracking system can open your file, extract the text, and assign each piece to the right field: name, contact details, employer, job title, dates. An ATS friendly resume is one that survives that copy step intact. It is not a score, and there is no pass mark.

Can an ATS read a PDF? Yes. Greenhouse, Lever, Workable, Breezy and SmartRecruiters all list PDF among the file types they accept for a resume, in their own documentation. The two real constraints are that the PDF must contain selectable text rather than a picture of your resume, and that the file should stay small: Greenhouse does not parse anything over 2.5 MB.

Are two-column resumes ATS-friendly? Mostly, with one caveat that is worth taking seriously. Two published tests this year reached opposite verdicts, but both found the same mechanism: the words come through and the reading order can scramble. Textkernel, which parses CVs commercially, measured contact information as the field that suffers when column rendering goes wrong. Use two columns if you like them, and keep your name, phone and email out of the sidebar.

How can I check if my resume is ATS-friendly? Try to highlight the text with your cursor. Lever's guidance is that a document you cannot highlight is likely not parseable. Then copy everything, paste it into a plain text editor, and read the result in order. That is roughly what the software sees, and anything out of sequence there is out of sequence for the parser too.

The one change worth making today

If your name and phone number are sitting in the document header, move them into the body. That single edit takes about thirty seconds and it protects the fields that three independent sources say break first. Everything else on the ATS-friendly resume checklist is worth doing, and none of it is worth as much as that.