How ATS Really Works: An Insider's Technical Guide

I spent 12 years at ZipRecruiter, where I led design for products that connected 110 million job seekers with employers. During that time, I watched our engineering teams build and refine resume parsing systems that processed over 7 million resumes per month. I sat in on countless product reviews where we debugged why qualified candidates weren't surfacing in recruiter searches. And I learned something that most resume advice websites get fundamentally wrong.

ATS systems don't reject resumes. People do.

The widely-cited claim that "75% of resumes are rejected by ATS before a human sees them" is a misunderstanding of how these systems actually work.[1] What actually happens is more nuanced, more technical, and — once you understand it — more fixable. This guide explains the real mechanics, based on what I saw working inside one of the largest hiring platforms in the world.

What an ATS Actually Is (And Isn't)

An applicant tracking system is a database with a workflow layer on top. It stores candidate information, tracks where each applicant is in the hiring pipeline, and gives recruiters tools to search, filter, and communicate with candidates. That's it. It's not an AI gatekeeper. It's not a resume-reading robot that decides your fate.

Think of it like a CRM for hiring. Salesforce doesn't "reject" potential customers — it stores their data and lets salespeople search for leads. An ATS does the same thing for job candidates.[2]

The ATS market has exploded in the past decade. In 2026, 97.8% of Fortune 500 companies use an ATS, and 93% of all recruiters report relying on one.[3][4] Even 60% of small businesses with under 50 employees now use one, up from roughly 30% in 2018.[5] The global ATS market is worth $3.28 billion and growing at 8.2% annually.[6]

Here's the landscape of major ATS platforms and who uses them:

The Three-Stage Resume Parsing Pipeline

When you submit a resume, it passes through a three-stage technical pipeline before a recruiter ever sees your data. Understanding each stage reveals where information gets lost — and what you can actually control.

Stage 1: Document Conversion

The first thing an ATS does is convert your file into machine-readable text. If you submit a Word document (.docx), this is straightforward — the system reads the XML structure that Word uses internally. PDF files require more work. The parser must determine whether the PDF contains selectable text (a "native" PDF) or is a scanned image (which requires OCR).[7]

Here's where the first data loss can happen. When I was at ZipRecruiter, our engineers tracked conversion accuracy rates across file formats:

• .docx files: 98%+ text extraction accuracy. The XML structure preserves formatting metadata.
• Native PDF files: 95%+ accuracy on modern ATS platforms (Greenhouse, Lever, Ashby). Older systems (Taleo) drop to 85-90% because they strip layout information inconsistently.
• Scanned PDFs: 80-90% accuracy with OCR, dropping significantly for low-resolution scans or unusual fonts.
• Image files (.png, .jpg): Below 80% accuracy. Just don't do this.

The modern trend is positive. In 2020, "always submit .docx" was universally correct advice. In 2026, most major ATS platforms handle native PDFs reliably. The exception is Taleo and some Workday configurations, where .docx still has a measurable advantage.[8]

Stage 2: NLP Extraction

Once the parser has raw text, it uses natural language processing to identify what each section of your resume contains. This is the most technically complex stage and the one most people misunderstand.

The parser doesn't "read" your resume the way a human would, top to bottom. It uses a combination of techniques:[9]

1. Section detection: The parser identifies section boundaries using headers, whitespace patterns, and contextual clues. It's looking for regions like "Experience," "Education," "Skills," and "Contact Information."
2. Named Entity Recognition (NER): Within each section, the parser identifies entities — company names, job titles, dates, degree types, school names, certifications, and skills. Modern parsers use machine learning models trained on millions of resumes.[10]
3. Relationship mapping: The parser connects entities to each other. "Software Engineer" + "Google" + "2019-2023" become a single work experience record. "B.S. Computer Science" + "Stanford" + "2019" become an education record.
4. Skill extraction: The parser identifies both explicit skills (listed in a "Skills" section) and implicit skills (mentioned in bullet points). Modern AI-powered parsers understand that "built microservices architecture" implies skills in distributed systems, API design, and backend engineering.

The critical insight from my time at ZipRecruiter: section detection is where most parsing failures originate. If the parser misidentifies where your "Experience" section starts and ends, every entity extraction within that section cascades into errors. Creative headers like "Where I've Made an Impact" instead of "Experience" can confuse older parsers. Non-standard section ordering (skills before experience, education split across multiple locations) increases error rates.

Stage 3: Field Mapping

The final stage maps extracted data into the ATS database schema. Every ATS has a structured database with specific fields: first_name, last_name, email, phone, current_title, current_company, years_of_experience, skills[], education[], work_history[].

This is where the "structured vs. unstructured" distinction matters. Your resume is an unstructured document. The ATS needs structured data. The mapping process is inherently lossy — some information in your resume simply doesn't fit into the database fields.[11]

Common mapping failures I saw at ZipRecruiter:

• Job title ambiguity: "VP of Design" maps cleanly. "Design Lead / UX Manager / Product Design" confuses the parser about which is your actual title.
• Date format inconsistency: Mixing "Jan 2020" with "2020-01" with "January 2020" in the same resume increases parsing errors for date fields.
• Multi-role experiences: If you held three positions at the same company, some parsers create three separate entries (correct), while others merge them into one (losing your progression).
• Skills vs. tools vs. frameworks: "Python" is clearly a skill. "Built a recommendation engine using collaborative filtering" contains skills, but the parser may not extract "collaborative filtering" as a searchable skill.

Why Your Resume Doesn't Surface in Recruiter Searches

The real reason most applications fail isn't formatting — it's keyword mismatch. Once your resume is parsed and stored, it sits in the database alongside hundreds or thousands of other applications for the same role. A recruiter then runs a search.

Here's what a typical recruiter search looks like inside an ATS:

Recruiters typically search using:

• Job title keywords: "data engineer," "product manager," "registered nurse"
• Required skills: "Python," "SQL," "Figma," "Salesforce"
• Certifications: "PMP," "CPA," "AWS Solutions Architect"
• Location filters: City, state, or "remote"
• Experience level: Sometimes filtered by years, sometimes inferred from title seniority

The ATS runs this search against its database and returns results ranked by relevance. Different platforms use different ranking algorithms. Greenhouse uses a combination of keyword match density and recency. Workday weighs exact title matches heavily. Taleo is the only major ATS that actually assigns a "requisition rank" — a percentage score based on how well your parsed data matches the job requirements.[12]

This is where the "75% rejection" myth comes from. Your resume wasn't rejected by the ATS. It was stored perfectly fine. But when the recruiter searched for "data engineer" + "Spark" + "Airflow," your resume said "data analyst" + "pandas" + "cron jobs." Same skill family, different vocabulary. The recruiter never saw you.

The 6 Actual Failure Modes (And How to Fix Each One)

After years of analyzing why qualified candidates failed to surface, I identified six distinct failure modes. Here's each one with its technical cause and practical fix:

1. Parsing Failure (5% of cases)

Your resume file couldn't be properly converted or parsed. This is the failure mode that most ATS advice focuses on, but it's actually the least common.

Technical cause: Multi-column layouts that split a single job entry across columns. Tables that cause section detection to fail. Headers embedded as images. Text in text boxes (which some parsers skip entirely).

Fix: Use a single-column layout. Standard section headers. No text boxes, no header/footer text for critical information, no tables for layout (tables for data within content are fine). Test your resume with a free ATS checker to verify parsing accuracy.

2. Keyword Vocabulary Mismatch (40% of cases)

Your resume uses different terminology than the recruiter's search query. This is by far the most common failure mode.

Technical cause: You wrote "people management" but the recruiter searched "team leadership." You listed "React.js" but they searched "React." You said "cost reduction" but they wanted "budget optimization."

Fix: Read the job description carefully and mirror its specific terminology. If the posting says "stakeholder management," use that exact phrase — not a synonym. Include both common abbreviations and full terms: "Search Engine Optimization (SEO)."

3. Missing Skills Section (15% of cases)

Your resume relies entirely on skills mentioned in context within bullet points, without a dedicated skills section that the parser can easily extract.

Technical cause: Most ATS platforms index a dedicated "Skills" section more heavily than skills mentioned in work experience descriptions. If you don't have a skills section, you're relying on the parser's NER accuracy to extract skills from natural language — which works about 70% of the time for explicit skill mentions and much less for implied skills.[13]

Fix: Include a dedicated "Skills" or "Technical Skills" section with your key competencies listed explicitly. Keep skills that are in the job description near the top of the list.

4. Title Mismatch (20% of cases)

Your actual job title doesn't match what the recruiter is searching for, even though you did the same work.

Technical cause: Company-specific titles ("Digital Experience Lead") don't match industry-standard searches ("UX Design Manager"). Some companies use inflated titles, others use quirky internal naming. The ATS doesn't know that "Ninja Developer" means "Senior Software Engineer."

Fix: If your official title is non-standard, add the industry-standard equivalent in parentheses: "Digital Experience Lead (UX Design Manager)." This preserves accuracy while ensuring searchability.

5. Recency Bias (10% of cases)

Your resume was submitted after the recruiter had already reviewed enough candidates. Most ATS search results default to "most recently applied" sorting.

Technical cause: Recruiters typically review applications in batches. The first batch of candidates (days 1-3 after posting) gets the most attention. By day 7, most recruiters have their shortlist and stop actively searching. Ladders research found that recruiters spend an average of 7.4 seconds on initial resume review.[14]

Fix: Apply within 48 hours of a job posting going live. Set up job alerts for your target roles. Early applications get more screen time.

6. Qualification Screens (10% of cases)

You answered a knockout question incorrectly or didn't meet a hard filter requirement.

Technical cause: Some ATS platforms support "knockout questions" — mandatory screening questions like "Do you have a valid nursing license?" or "Are you authorized to work in the US?" If you answer "No" to a knockout question, the ATS flags your application as disqualified. This is the one case where the ATS does perform an automated screen. But it's based on your explicit answers, not your resume content.

Fix: Read every application question carefully. If a question asks about a requirement you genuinely don't meet, be honest — but understand that a "No" answer to a knockout question will likely disqualify your application automatically.

What I'd Do Differently Knowing What I Know

After 12 years on the other side of the ATS, here's what I'd prioritize if I were job searching today: . After 12 years on the other side of the ATS, here's what I'd prioritize if I were job searching today.

After 12 years on the other side of the ATS, here's what I'd prioritize if I were job searching today: .

After 12 years on the other side of the ATS, here's what I'd prioritize if I were job searching today:

1. Study the job description like a legal document. Every keyword in the "requirements" section is a potential search term. Mirror the exact phrasing in your resume.
2. Include both acronyms and full terms. Write "Customer Relationship Management (CRM)" so you match either search variation.
3. Use a dedicated skills section. Don't rely on the parser to extract skills from your bullet points. List them explicitly.
4. Apply early. The first 48 hours matter more than the perfect resume. A good resume submitted on day 1 beats a perfect resume submitted on day 10.
5. Don't obsess over formatting. A clean, single-column layout with standard headers is sufficient. The marginal difference between "ATS-optimized" templates is negligible compared to the keyword match issue.
6. Tailor every application. The single biggest improvement you can make is customizing your resume's keywords and skills section for each job posting. Yes, it takes more time. It's the only thing that consistently moves the needle.

The Future: AI Is Changing Everything

The ATS industry is in the middle of its biggest transformation since the shift from paper resumes to digital applications. AI-powered parsing is replacing rule-based parsing. Instead of matching exact keywords, modern systems are beginning to understand semantic similarity — recognizing that "managed a team of 12 engineers" and "engineering leadership" describe the same capability.[15]

At ZipRecruiter, we were already building these semantic matching capabilities in 2022-2023. The technology works, but adoption is slow because recruiters need to trust the results. When you switch from keyword matching to AI matching, the ranking of candidates changes dramatically — and recruiters are skeptical of systems that surface different people than they're used to seeing.

What this means for job seekers: the advice in this guide will remain relevant for at least the next 3-5 years. Even as AI parsing improves, the fundamental architecture of ATS systems — parse, store, search — isn't changing. And as long as recruiters are the ones constructing search queries, keyword alignment will remain the primary determinant of whether your resume surfaces.

The ATS isn't your enemy. It's a tool that does exactly what it's designed to do. Once you understand how it works, you can work with it instead of against it.

Frequently Asked Questions

Does an ATS automatically reject my resume?

No. ATS systems parse and store your resume in a database — they do not auto-reject applications based on content. The "75% rejection rate" statistic is misleading. What actually happens is that recruiters search the database using keywords, and resumes that don't match those keywords never appear in search results. Your resume is still stored; it just isn't surfaced to the recruiter.[1]

Should I submit my resume as a PDF or Word document?

In 2026, either format works on most modern ATS platforms. Greenhouse, Lever, Ashby, and SmartRecruiters all handle native PDFs reliably. The exceptions are older systems like Taleo and some Workday configurations, where .docx has a slight parsing accuracy advantage. When in doubt, submit .docx for maximum compatibility across all platforms.[8]

Do I need a special ATS-friendly resume template?

Not really. Any clean, single-column layout with standard section headers will parse correctly. Avoid multi-column layouts, text boxes, embedded images for text, and creative section headers that a parser won't recognize. The marginal difference between "ATS-optimized" templates and a standard professional template is far less important than matching the right keywords.

How do recruiters actually search inside an ATS?

Most recruiters type 3-5 keywords (job title, 2-3 key skills, sometimes a certification), apply a location filter, and review the top 20-30 results. The search takes about 2 minutes to construct. If your resume doesn't match those specific terms, you won't appear in results regardless of your qualifications.

Which ATS platform does my target company use?

You can often identify the ATS from the job application URL. Greenhouse uses "boards.greenhouse.io," Lever uses "jobs.lever.co," Workday uses "myworkdayjobs.com," and iCIMS uses a numeric job ID pattern. Knowing the platform helps you understand how strictly to follow formatting guidelines — Greenhouse is forgiving, Taleo is strict.

References

1. The Tech Resume Inside Out — ATS Myths Busted
2. SmartRecruiters — What is Resume Parsing?
3. Jobscan — 2025 Fortune 500 ATS Usage Report
4. RecruitCRM — ATS Statistics 2026
5. Select Software Reviews — ATS Statistics 2026
6. MarketsandMarkets — ATS Market Worth $4.88B by 2030
7. Affinda — How OCR Resume Scanning Transforms HR Tech
8. Workable — How an ATS Reads a Resume
9. MagicalAPI — Resume Parsing with ML and NLP
10. X0PA — AI Resume Parsing: Definition, Benefits, Types
11. Bullhorn — What is Resume Parsing?
12. The Tech Resume Inside Out — Only Taleo Assigns Requisition Ranks
13. MagicalAPI — NER Accuracy in Resume Parsing
14. Ladders — Recruiter Eye-Tracking Study
15. X0PA — AI-Powered Semantic Matching in Parsing