How ATS Resume Parsers Actually Read Your CV in 2026

Most resume advice online still treats Applicant Tracking Systems as a single mysterious filter that "robots" use to discard 75% of applications. That mental model was wrong in 2018 and it's even further from the truth in 2026. The real picture is more boring — and far more actionable.

This post walks through what major ATS platforms (Workday, Greenhouse, Lever, iCIMS, Taleo, SmartRecruiters) actually do with your resume the moment you press Apply.

Step 1: file ingest

The ATS doesn't open Word or Acrobat. It runs your file through a server-side parser. For DOCX files it reads the underlying Office Open XML directly — the same structured tags Microsoft Word emits. For PDFs it has to do something harder: re-derive text from the visual glyph positions inside the PDF. That's where most "ATS-friendly" advice breaks down.

If your PDF was generated by Word or Google Docs, parsing works fine. If it was exported from Canva, Cake Resume, Notion, or Figma, the parser is much more likely to:

• Misread multi-column layouts as one continuous run of jumbled text.
• Skip text rendered as an image or SVG path.
• Glue adjacent words together because the glyphs lack reliable space-character metadata.
• Lose every character in a custom display font that doesn't ship a ToUnicode map.

Plain DOCX bypasses all of that. This is why kairesume now ships both PDF and DOCX in the download bundle — DOCX for ATS submission, PDF for human-facing emails and LinkedIn.

Step 2: section detection

Once text is extracted, the parser looks for canonical section headings: Experience, Education, Skills, Projects, Certifications. Anything inside those sections is then bucketed for downstream analysis.

Two common failure modes:

1. Creative headings. "My Journey" instead of "Experience" — parser misses the section entirely and your jobs end up unlabeled.
2. Section bleed. No clear visual break between sections, so the Education list gets appended to your last job's bullets.

Keep headings boring. The parser is grateful.

Step 3: entity extraction

This is where ATS platforms have actually gotten dramatically better. Modern parsers identify:

• Job titles (matched against an internal taxonomy with hundreds of variants).
• Companies (cross-checked against a database — known employers boost the match score, unknown ones are accepted but flagged).
• Dates and durations ("Jan 2022 – Present" → 3.4 years tenure).
• Skills and tools (a list of canonical skills, each with synonyms — "K8s" matches "Kubernetes", "TF" matches "Terraform").
• Education credentials with degree, institution, and graduation year.

Whatever the recruiter typed into the requisition gets matched against these extracted entities. That match — not raw keyword density — is what drives most modern ATS scoring.

Step 4: ranking signals

The score the recruiter sees blends several inputs:

• Skill overlap with the requisition (weighted by how recent the experience is).
• Years of experience in the matching skill area.
• Title progression — has this candidate moved up?
• Tenure stability — average years per role.
• Location and right-to-work signals.
• Education match (for roles that require a specific degree).

Notice what's not on the list: how pretty your resume looks. Fancy templates don't help. ATS-clean templates don't hurt your score; they just stop the parser from corrupting your data.

Common myths, briefly

"Keyword stuffing works." It used to, in 2010-era systems. Modern parsers detect repetition without context and discount it. Use each target keyword once or twice, in a sentence where it makes sense.

"White text on white background fools ATS." It also gets flagged by every QA check the recruiter runs and is grounds for instant rejection at most large employers. Don't.

"ATS rejects PDFs." Not true. Well-formed PDFs from Word, Google Docs, or LaTeX parse fine. Visually-designed PDFs from Canva and similar tools are what cause problems.

"The robot decides — humans never see it." Almost always wrong. Most ATS configurations rank candidates and present the top 20-50 to a recruiter, who reads them. The filter doesn't reject — it reorders.

What this means for your resume

A few practical implications:

• Use a single-column, plain layout. Headings: Experience, Education, Skills.
• Include the exact skill names from the job ad — not paraphrases.
• Date every job (Jan 2022 – Present), not just years.
• Submit DOCX when the form accepts it; otherwise a Word-exported PDF.
• Write bullets that pair a tool with an outcome ("Migrated 30 services to Kubernetes, cutting infra cost by 22%").

That's it. Not flashy, not clever — just legible to the system on the other side. If you want this done in under a minute against a specific job description, paste it into kairesume and we'll generate an ATS-clean version with the right keywords highlighted.