How 5 ATS Systems Actually Parse Your Resume

How Do ATS Systems Read Your Resume?

ATS systems parse PDF and DOCX resumes using Apache Tika, extracting raw text coordinates horizontally and stripping all CSS, columns, and Canva graphics. Workday, Greenhouse, Lever, Ashby, and SmartRecruiters rebuild extracted text into relational candidate database profiles. Multi-column layouts cause critical data interleaving in Workday, producing auto-rejected, corrupted candidate profiles.

• Greenhouse extracts text sequentially because it processes PDF streams in document order. Two-column layouts interleave adjacent column text and corrupt candidate sections into unreadable output.
• Workday strips all CSS and works in raw ASCII because it prioritizes enterprise-scale parsing speed. Only single-column PDFs from Word or Google Docs parse reliably.
• Lever identifies sections using NLP pattern-matching on headers like "Work Experience" because it relies on keyword detection. Non-standard headings like "My Journey" cause job history to go unmapped.
• Ashby deploys ML boundary detection because it serves engineering-focused companies requiring higher tolerance, though icon-based contact sections still cause critical field-mapping failures.
• SmartRecruiters applies cloud OCR because it supports scanned documents, though mixed Unicode scripts without structural breaks cause field misassignment in the candidate database.
• Unlike resume builders such as Canva or Zety, Jobloo generates a uniquely tailored, ATS-parsed resume for each job description and submits it through authenticated Workday, Greenhouse, and Lever endpoints, combining CV writer, ATS compliance engine, and application agent in one automated pipeline.

Two hundred applications sent. Qualified for every one of them. And nothing back. Not even an auto-reply. Your file format was the reason. This is a problem every auto-apply tool faces. including tools like Sorce. Our Sorce vs Jobloo comparison covers how both platforms approach this formatting problem differently. And nothing back. Not even an auto-reply. Your file format is what's killing you.

I build CV parsers for a living. I run Jobloo, a platform that reads resumes and submits tailored applications to real company career pages. To build our parser, I had to reverse-engineer exactly how the five most popular ATS platforms extract data from your resume, and where they fall apart.

What ATS actually does to your resume

An ATS is a data extractor, full stop. It takes your PDF, strips all the formatting, and tries to shove the resulting text into database fields: name, email, phone, experience, education, skills. If your formatting is simple, this works. If it is not, things get ugly fast.

When the extraction breaks, your name merges with your address. Your most recent job disappears. Your skills section becomes a wall of noise. That is why picking the right tool matters, if you want to see how the leading AI job tools handle this differently, the Jobloo vs LazyApply vs Sonara vs Seekario comparison breaks it down across four platforms. Your most recent job disappears. Your skills section becomes a random string of characters mixed into your work history. The recruiter opens your profile, sees a wall of garbled text that looks like a corrupted database entry, and clicks past you in two seconds. They thought it was a bad application. It was a parsing failure.

That is the part nobody tells you. A human rejected you, but they were looking at a mangled version of your document that you never approved and never saw.

1. Greenhouse

Used by Airbnb, Spotify, HubSpot, Stripe, Cloudflare, Notion, and thousands of mid-size tech companies. Greenhouse reads your document top-to-bottom, left-to-right, extracting text in the order it appears in the file's internal structure. It handles single-column PDFs well.

Two-column layouts are where it breaks. Greenhouse reads straight across the page, so if you have a left column (skills, contact) and a right column (experience), it interleaves them: left line 1, right line 1, left line 2. Your experience becomes word soup. The recruiter sees "Python SQL 5 years at Deloitte" merged into a single line that makes no sense.

Canva templates are the other common failure point. When your name or section headers are part of a graphic rather than selectable text, Greenhouse cannot extract them. Your name field shows up blank. Your contact info vanishes. Information placed in PDF headers and footers gets dropped entirely, so if your phone number is up there, consider it gone.

Fix: single-column PDF, standard headings, zero graphics, text you can actually select with your cursor.

2. Lever

Netflix, Shopify, KPMG, Atlassian. Lever is more modern than Greenhouse and better at detecting section boundaries automatically. It looks for known heading patterns like "Work Experience", "Education", and "Skills", then maps content to the right profile fields.

The problem is it takes those section headings literally. Call your experience section "My Professional Journey" and Lever may not recognize it as work experience at all. Your job history ends up unmapped. Similarly, if you used invisible tables to get your layout to look neat, Lever reads table cells in unpredictable order, usually wrong. Date formatting matters too: "Jan '22 to Present" fails where "January 2022 to Present" works.

Comma-separated skills in plain text. Standard month-year dates. Nothing fancy. That is the formula.

3. Workday

This is the one that should worry you. Deloitte, Amazon, Walmart, Siemens, Unilever, L'Oreal, most Fortune 500 companies. Workday processes an insane volume of applications, and speed takes priority over accuracy. It strips your PDF down to raw text, throws away all formatting context, and then tries to reconstruct your profile from scratch using NLP. It looks for patterns like job title followed by company name.

When it works, it works. When it fails, it fails spectacularly. Multi-column resumes produce output like "J ohn D oe" when letters from two adjacent columns get interleaved. I have seen this happen with resumes from senior engineers with 15 years of experience. Design-tool PDFs (Canva, Figma, InDesign) are even worse. The text is layered on vector shapes, and Workday's extractor just sees nothing.

For Workday specifically, the only safe resume is genuinely boring. Single column. Times New Roman or Arial. No colors. No graphics. Generated from Google Docs or Word and exported as a standard PDF. It will look like it's from 2008. That's the point.

4. Ashby

Ramp, Linear, Vercel, Loom, Retool. Companies that tend to care about engineering quality. Ashby's parser reflects that. It combines text extraction with machine learning to identify sections, handles modern minimal designs better than the legacy systems, and is generally more forgiving of non-standard formatting.

That said, icon-based contact sections break it. If your email is represented by a mail icon rather than the word "Email:", Ashby might miss your contact info entirely. Progress bar skill ratings (the "4 out of 5 stars for Python" type) also register as nothing. And resumes with sidebars nested inside the main content area create reading-order ambiguity that trips up even Ashby's ML layer.

If you're applying to companies on Ashby, you have the most latitude. A clean, minimal single-column design with real selectable text will parse fine.

5. SmartRecruiters

Visa, Bosch, LinkedIn, Equinox. SmartRecruiters uses a cloud-based parsing service that supports PDF, DOCX, and plain text. It is solid, but it has specific failure modes worth knowing.

Scanned resumes are the worst case. If you photographed or scanned your resume from paper, the OCR frequently fails or produces garbage. Mixing bold, italic, and different font sizes within the same text line confuses section boundary detection. And if your resume mixes English with Arabic or Chinese script without clear structural breaks, the parser struggles to assign text to the right fields.

Standard single-column PDF or DOCX with consistent formatting throughout. That covers 95% of cases.

ATS System Comparison: Parsing Capabilities. for a broader look at which AI job tools handle ATS submission best, see The 10 Best AI Job Search Tools in 2026. & Failure Modes

The Plain Text Test (Do This Right Now). or use the Jobloo Resume Grader to see exactly how an ATS reads your file before you submit it anywhere.

Here is a 30-second test that tells you whether your resume will survive any ATS:

1. Open your resume PDF.
2. Press Ctrl+A (select all text).
3. Press Ctrl+C (copy).
4. Open Notepad (or any plain text editor).
5. Press Ctrl+V (paste).

Now look at the result:

• If the text is readable and in the correct order (name at the top, then contact info, then experience, then education). your resume will parse correctly in all five systems.
• If the text is jumbled, out of order, or missing sections. your ATS submission is broken. Every application you send with this file is a wasted application.

This test works because it simulates exactly what an ATS does: extract the raw text layer from your PDF, strip all formatting, and read what remains.

The 5 Universal Rules (Works on All 5 Systems)

1. Single column only. No sidebars, no two-column layouts, no grids. Top to bottom, full width.
2. Standard section headings. Use "Experience" or "Work Experience", "Education", "Skills", "Languages". Not "My Journey", "Toolkit", or "Superpowers".
3. No graphics, icons, or images. No headshots, no skill bars, no star ratings, no logos. Text only.
4. Text-based PDF. Export from Google Docs or Word. Never from Canva, Figma, or Photoshop. The text must be selectable, not baked into an image.
5. Mirror the job description. If the job posting says "project management", write "project management", not "managed projects" or "PM". ATS keyword matching is literal, not semantic.

The Keyword Stuffing Myth

A persistent myth says you should hide keywords in white text (invisible to humans, visible to ATS). This worked in 2015. In 2026, every major ATS can detect hidden text, and it will flag your application as spam. Some systems auto-reject these submissions.

Modern ATS platforms like Greenhouse and Ashby use contextual matching: they do not just count keyword occurrences. They check whether the keyword appears within a coherent sentence that describes actual work. "Managed a cross-functional project management initiative" scores higher than "project management project management project management" hidden in a footer.

The correct approach: read the job description carefully, identify the 5 to 10 most specific terms (tools, methodologies, certifications), and weave them naturally into your experience bullet points.

Related articles

• I Processed 500,000 Job Applications With AI. Here Is What the Data Shows.. Callback rates by ATS, timing, and keyword match score from Jobloo's Q1-Q2 2026 dataset. Numbers behind why ATS format matters.
• The 10 Best AI Job Search Tools in 2026. Our ranked comparison of every major AI job search tool in 2026, with honest notes on what each one actually does.
• Tinder for Jobs: How Jobloo's Swipe UI Works, How Jobloo's swipe-to-apply interface works: what the Two-Pass AI engine does when you swipe right on a job.
• Jobloo vs LazyApply vs Sonara vs Seekario. An honest side-by-side of Jobloo, LazyApply, Sonara, and Seekario, which one gets the most interview callbacks.
• Sorce vs Jobloo: The Better Alternative, Where Sorce wins, where Jobloo wins, and which platform is the right choice depending on what you're optimizing for.
• Is Jobloo Legit? Honest Review. A founder-written review of Jobloo covering security, RGPD compliance, real application numbers, and where the product falls short.

Frequently Asked Questions

Why does my resume keep getting auto-rejected by ATS parsers?

ATS parsers (like Greenhouse or Ashby) don't actually reject resumes; rather, recruiters do because of parsing corruption. Behind the scenes, libraries like Apache Tika extract text coordinates horizontally. If you use a Canva template or multi-column layout, the text blocks merge out of order. For example, "Experience: Software Developer" and "Skills: Python" get read as "Experience: Software Skills: Developer Python". When the recruiter searches for candidates, your profile appears as garbled text and gets skipped. You can check your resume readability instantly using the Free ATS Resume Grader sandbox tool, which simulates client-side text layer extraction.

Which applicant tracking system (ATS) is the hardest to pass, and why?

Workday is widely regarded as the most difficult and unforgiving ATS. It strips all custom styles, tables, and columns to build its own relational candidate profile. This process frequently scrambles two-column layouts and non-standard fonts. Modern platforms like Greenhouse, Lever, and Ashby use advanced parsers but still fail when custom typography maps characters to incorrect Unicode values, making your contact details or experience unsearchable.

Is it better to submit a PDF or a DOCX resume file to job applications?

A text-based PDF is the industry standard for Greenhouse, Ashby, and Lever. The critical requirement is that the text layer is selectable, not flattened as an image. Scanned PDFs or designs exported from Canva and Figma often have scrambled text layers or zero text characters (scanned images), which require OCR. When an ATS parser reads them, they show up completely blank to recruiters. If you choose PDF, always ensure it is exported from a word processor like Microsoft Word or Google Docs.

How can I tell if an ATS is parsing my resume layout correctly?

You can perform the "Plain Text Test" by copying all text from your PDF and pasting it into a raw editor like Notepad. If your sections (Experience, Skills, Education) are out of order, or if sentences are merged, the ATS will see the exact same scrambled layout. Alternatively, use Jobloo's Free ATS Resume Grader to instantly run client-side text layer checks and see a live preview of the extracted database structure.

Does repeating keywords in a resume help bypass automated filters?

No. Modern ATS platforms use semantic scoring models (NLP) and reject spam-like keyword stuffing (such as hiding lists of keywords in tiny white text). Natural language processing models check for contextual relevance. Instead, mirror the exact phrases from the job description naturally inside your professional experience bullet points to ensure high relevance scores.

Once your CV is in good shape, if you are looking at tools that handle automatic applications, we compared the four main AI job application tools in detail: LazyApply vs Sonara vs Seekario vs Jobloo, which one actually gets interviews?

Your next interview is one fix away.

Jobloo reads your CV the same way these ATS systems do, fixes what would break, tailors your application to each job description, and submits it on real company career pages. You swipe right. We handle the rest.

Try Jobloo Free