ATS bots reject 75% of resumes. Here's exactly how to bypass them.
If you've sent 50 applications and heard back from one, the issue probably isn't your skills. It's that 38 of those CVs never reached a human. Applicant Tracking Systems — Workday, Greenhouse, Lever, Naukri RMS, the works — filter on a small set of signals before any recruiter sees the file.
This is the technical breakdown. We tear down the four major ATS engines used in India and walk through the six specific changes that get you through. None of this is “optimisation” trickery. It's just understanding what the bot reads.
What the bot actually does
Modern ATS engines run three steps:
- Parse the file. Convert your PDF / .docx into structured
data: name, email, education, work history, skills. This step alone fails on ~15% of CVs because of formatting.
- Match against the JD. The bot computes a similarity score
between your CV and the job description — TF-IDF in older systems, embedding similarity in newer ones (Workday 2024+, Greenhouse Mayan AI). Below ~70% match → auto-reject.
- Flag risk signals. Long employment gaps, inconsistent dates,
non-standard email domains, embedded images, “objective” sections — anything that looks like padding gets a penalty.
You can't argue with steps 1–3. You can only feed them the right input.
The 6 specific changes that work
1. Single column. No tables.
Tables get scrambled. The bot reads left-to-right top-to-bottom and concatenates whatever it finds. A two-column “skills | dates” layout gets parsed as “skills dates skills dates skills dates,” which is nonsense.
Fix: one column. Use whitespace and section headers for hierarchy. Looks blander to humans. Reads correctly to bots.
2. Standard fonts, embedded.
The parser uses font hints to detect headings. Custom fonts that aren't embedded in the PDF can fail to render — your headings get read as plain text. Stick to Helvetica, Arial, Calibri, Garamond, Times. Save with “Embed fonts” checked.
3. Filename matters more than you think.
A file named "untitled (3).pdf" looks like spam. "kundan-khatri-ml-engineer.pdf" gets through. Some ATS engines use the filename as a backup parser hint when the PDF parsing fails.
4. The keyword exercise.
This is the highest-leverage move. Open the JD. Bold every noun, verb, and tool name. Anything mentioned 2+ times is a must-have keyword. Your CV needs each must-have keyword to appear at least twice — once in skills, once in an experience bullet.
Example. JD says “PyTorch (5 mentions), distributed training (2), LangChain (3), MLflow (1).” Your CV needs PyTorch + distributed training + LangChain in both skills and a bullet. MLflow optional.
Don't keyword-stuff. Modern ATS (post-2024) penalise unnatural density. Use the keyword in a real sentence describing real work. If you can't, you don't have that skill — and the interview will catch it anyway.
5. Drop the “Objective” section.
Recruiters in 2026 don't read it. Bots specifically flag it as “low-signal padding.” Same goes for hobbies, marital status, declarations, photo, full address. These were standard in
- They burn space the bot won't credit.
6. PDF format with text-based content (not scanned).
Scanned PDFs are images. The OCR engines on most ATS are mediocre. Always submit a PDF generated FROM a document — Word, Pages, LaTeX, Google Docs — never a scan. Test: open your PDF, try to copy-paste the text. If you can, the bot can read it. If you can't, it's an image and the bot is guessing.
Test before you submit
Free tools that test ATS-readability:
- Jobscan.co — pastes your CV against a JD and gives a match
score. Get above 70%.
- Resume Worded — flags the obvious format / content issues.
- The plain-text test. Open your PDF, select all, copy, paste
into a plain-text editor (TextEdit, Notepad, VS Code with .txt). If the order of content is preserved and nothing's missing, your CV is ATS-safe.
The real fix is structural, not tactical
Everything above gets you through the bot. None of it gets you hired. The market is moving past CV-driven hiring entirely — the companies hiring AI/ML talent fastest in India in 2026 (Razorpay, Flipkart, the YC W26 batch) are searching by verifiable scores instead of pre-filtering on resume.
That's the bigger move. Ship a real project. Get a calibrated AI Score. Make it public. Recruiters search the score, not the CV. ZT exists for exactly this — first profile is free at zerotheory.live, score in 30 seconds.
If you've done the six fixes above and you're still hitting silence, the system is telling you something — not that you're under-skilled, but that the channel is broken. Switch channels.
— Kundan, founder of ZT