ATS-Friendly Resume Templates: What Actually Gets Parsed
Twelve resume templates, printed to A4 and read back with a text extractor. Every word survived in all twelve; what moved was the order. In two layouts the entire side column arrives before the candidate's name. Plus what the ATS vendors actually document, and the one statistic nobody can source.
Seventeen specific probe strings buried in a single-page PDF, printed from a layout engine and pushed through a text extractor. Every word came back out. The layout does not stop your text getting through an applicant tracking system. What gets parsed is text. The layout simply decides the sequence that text arrives in.
This is not about where uploaded resume data goes. This is strictly the mechanical step of pulling words out of a file. The text survives. Everything after that is evidence.
What the vendors actually publish
Product documentation fractures immediately. Greenhouse published a support page on 2 March 2026 listing reasons parsing fails. They cite columned layouts, graphics, and word art. They explicitly warn against putting contact information in a header, a footer, or a text box. Their system stops at a 2.5MB file limit. A candidate FAQ from 1 June 2026 suggests uploading a PDF for best results, noting parsing makes mistakes.
Workday admin documentation from 23 June 2023 says parsing results vary by format and word order. They advise against image-based styles.
Oracle Taleo documentation carries no date and paints a different picture. They claim parsing ignores bold, italic, and bullet formatting, and state their tool can detect text in the header and footer. Their 100-kilobyte limit on file size is 25 times smaller than the Greenhouse ceiling.
Greenhouse says a header breaks contact extraction. Oracle says its parser reads headers fine. The applicant tracking system as a single, uniform machine is a fiction.
Reading order and the real measurement
No vendor documents that its software reads a page top to bottom or left to right. The W3C published a technique on 9 September 2026 stating that if a PDF is not properly tagged, a screen reader may interpret two columns as one. The tag order dictates the reading order. To see what extraction does to a layout, all twelve ResumeRoast templates, from the Mocha category, were tested on 13 September 2026. They were printed from the app's 2.2.5 stylesheet using headless Chrome 152, then read back with pdftotext from poppler 26.09.0. A text extractor is not an applicant tracking system. It stands in for the extraction step only.
Nothing was lost anywhere. All seventeen probe strings and all four date ranges survived. In Executive, the name returned as MORGAN A. WHITFIELD because the template uppercases it in CSS. The sequence is what moved. Six templates kept a human's reading order. Five moved something, and Bold shifted only its job title. The two that scrambled the sequence hardest are the two with a full-height side column that comes first in the markup.
In Modern and Sidebar, the entire left rail arrives before the candidate's name. Email, phone, location, website, LinkedIn, GitHub, all eight skills, both languages, and both certifications pour out first. Only then does the name appear. It sits 302 characters into 1,786 in Sidebar. In Modern, it lands 325 characters into 1,845. That is about a sixth of the way down the page's text.
Compact and Timeline are not side-column designs, yet both reordered the text. Compact opens the stream with three date ranges and a GitHub address before reaching the name. Timeline pushes the skills, languages, projects, and certifications ahead of the work history. Creative drops the contact block after the work history and ends the stream on the skills list. The layout dictates the sequence.
Modes disagree and spaces disappear
Two modes of the exact same tool disagreed on nine of the twelve files. Content-stream order, which is how the page paints, and position-sorted order, which re-sorts blocks by coordinates, matched only for Elegant, Minimal, and Executive. For Corporate, sorted mode put the email, phone, and location before the name, while the raw stream put the name first.
Ten of the twelve layouts emitted almost no spaces. Out of roughly 1,800 characters, Compact produced exactly one space. Sidebar and Corporate produced six. Elegant and Minimal offered seven. Modern, Classic, Creative, Timeline, and Bold gave up eleven or twelve. Only Executive and Technical kept real word spacing, holding 260 and 238 spaces. Those are serif and monospace designs, making it a font difference, not a layout trick. In the other ten files, a job title comes out of the raw stream as DirectorofEngineering. The position-sorted mode reconstructs spaces by measuring gaps between glyphs. The raw mode cannot. Spacing on your screen is not necessarily spacing in the file.
The bullet dots are missing entirely. No bullet character sits in front of any list item in any of the twelve templates. The marker is just a circle drawn by the stylesheet. A geometric shape. Nothing is lost by that, but nothing marks a list as a list either.
Where the filter statistic comes from
The claim that an applicant tracking system rejects 75 percent of resumes is repeated everywhere. It traces back to a Computerworld article published on 4 March 2012. The figure is attributed to a vendor selling resume-optimization services. No study, dataset, or sample size is cited.
Academics at Mount Royal University and the University of Calgary wrote in The Conversation on 5 July 2026 that the statistic originated in a 2012 sales pitch by a company that folded the following year. Practitioners find nothing either. Christine Assaf noted in October 2020 there is no concrete source data to back up the statement. There is no trustworthy figure to put in its place.
Which layout to send
As the measurement shows, ripping the photo out doesn't fix your text flow. Compact is photo-free, but still opened its stream with three date ranges before your name. A clean reading order tracks with a single column of real text. Modern and Sidebar scrambled things hardest. Both lean on a full-height side column that comes first in the markup.
This gives you a plain rule. When a posting goes through an automated system, send Classic. It is the only option in ResumeRoast literally labelled "Classic (ATS)". The listing states that ATS-friendly templates stay text-only for applicant-tracking systems.
Keep the rest for applications a person will actually open. Switching a template reflows the same content into the new layout, so trying another look costs nothing. Duplicate your resume for each job. Send Classic to the portal, and keep the side-column designs for a portfolio submission where the look is doing work.
Where this breaks
The export from this app is A4 only. The paper size is fixed in the code. There is no US Letter option. US Letter is 8.5 by 11 inches, the standard in the United States and Canada. Anyone applying there sends a page that isn't their market's paper size.
Only Classic carries the ATS label. The other five photo-free templates are untested against any system and not certified against anything.
What was run here was a text extractor. It shows what the extraction step yields. It doesn't show what any real system does with the result, how it scores the text, or if anyone reads it. Only the Windows print path was tested. The macOS export goes through a different browser engine and may order output differently. This was one page, one resume, on one machine on 13 September 2026. Nobody can guarantee a resume parses correctly everywhere, and this post doesn't.
Extraction is a messy mechanical process, and the rules change depending on whose server answers the request. If you want to run your text through the Classic template and see what the PDF looks like, grab a membership and fire up the app.