What your resume actually looks like to the machine
Most ATS posts on LinkedIn say "use keywords." Useful, but not concrete. I spent a weekend testing five ATS platforms (Greenhouse, Lever, Workday, iCIMS, Ashby) with the same resume in different formats. Here is what your resume actually looks like once it goes in.
Test 1 — The two-column resume disaster
I uploaded a resume with a left sidebar (skills, contact) and a right column (experience). Visually clean. To the parser:
John Smith
Frontend Engineer
React
TypeScript
Next.js
2024 - present
Senior Engineer at Acme
Led migration to App Router
Reduced bundle size 40 percent
The sidebar got stuck above the experience block. Skills floated above context, dates merged into the next line. A recruiter searching for "React + Next.js + 3 years" would not find me — the parser had no idea those skills connected to a 2024-present role.
Fix: Single column. Always. Pretty design loses to a parser every time.
Test 2 — Headers the machine cannot read
Half the templates from popular sites use icons next to "Experience" or replace "Skills" with "Toolbox" / "What I do well."
Greenhouse and Workday parsed "Experience," "Education," "Skills" cleanly. They both choked on "What I do well" and dumped that section into a generic "Other" bucket that recruiters never look at.
Fix: Use the boring section names. The robot is not impressed by your creativity.
Test 3 — Dates in the wrong format
The parser handles these well:
Jan 2024 - present01/2024 - present2024 - present
The parser fails on:
January '24 - nowSpring 2024 - current- Date ranges in side comments
When the parser cannot read your dates, your role becomes "experience without timeline" and gets ranked below candidates with parseable dates.
Test 4 — PDF versus DOCX
Three of the five platforms preferred DOCX. Specifically: Workday and iCIMS produced cleaner parses from DOCX than PDF. Greenhouse, Lever, and Ashby were tied.
If a job application asks for either, send DOCX. If it asks specifically for PDF, send PDF and check that text is selectable (not an exported image).
Test 5 — Skills in a table
I put skills in a 3-column table. Parsing result: every row got concatenated into a single line. So my "Languages: Python, JavaScript, SQL" became "Python JavaScript SQL Frameworks Django React Node Tools Git Docker Kubernetes" — readable to a human, useless to a keyword search that looks for "Python developer."
Fix: Use comma-separated lines under each header. Plain text.
Test 6 — Bullet points starting with "Responsible for"
Every parser kept the bullet as text. But every recruiter search ranks "Responsible for managing X" lower than "Managed X — reduced cost by 30 percent." The verb plus number combination is what the human reading the parsed output gravitates to.
Fix: Verb + outcome + number. If you have no number, write the verb anyway. "Designed" beats "Was responsible for designing."
Test 7 — File name matters more than I thought
Two of the five platforms include the file name in the search index. Resumes named resume_v3_FINAL.docx ranked lower than Smith-John-Frontend-Engineer.docx for searches on "Frontend Engineer."
Fix: Name the file with your name and target role.
Test 8 — Hidden white text
Old trick: white text crammed with keywords so the parser sees them but the human does not.
Three platforms now flag this. Workday explicitly warns recruiters when text color matches background. Greenhouse strips it. The trick used to work in 2022. It is dead.
Quick rebuild checklist
- Single column layout
- Boring section names: Experience, Education, Skills
- Dates in
Mon YYYY - Mon YYYYformat - DOCX file (unless PDF specifically requested)
- Skills in comma-separated lines, not tables
- Bullets starting with action verbs, ending with numbers
- File name
Firstname-Lastname-Role.docx
What to do today
Open your resume. Run it through a parser preview (free here). If the output looks like a wall of merged text, the recruiter sees the same thing.
The parser does not care about your design taste. It cares about structure. Win the parser, then the human gets to see your work.
Related deep dives:
- Your Job Application Is Being Scored by AI — original ATS explainer
- I Built a Job Description Keyword Extractor — pair this with the parser test
Top comments (0)