Most advice about ATS systems treats them like a black box. "Optimize for the ATS" people say, without explaining what the system actually does. That is like telling someone to optimize for Google without explaining what a search engine is. So let me open the box.
The parsing pipeline, step by step
When you upload a resume to a job application portal, the ATS runs it through a parsing pipeline. The specific implementation varies by vendor, but the general process looks like this:
- File ingestion. The system accepts your file (DOCX, PDF, or plain text) and determines its type.
- Text extraction. For DOCX files, the parser reads the XML structure directly. For PDFs, it uses a text extraction library to pull out the content. For scanned documents, some systems attempt OCR (optical character recognition), though many skip this step entirely.
- Section identification. The parser looks for known section headers - Experience, Education, Skills, Certifications, Summary - and uses those as structural markers to divide the document into zones.
- Entity extraction. Within each zone, the parser identifies specific entities: names, email addresses, phone numbers, company names, job titles, dates, degrees, skill keywords.
- Field mapping. Each extracted entity gets mapped to a structured database field. Your name goes into the "name" field, your most recent job title into "current title," your listed skills into a searchable skills array.
- Indexing. The structured data is indexed so recruiters can search, filter, and sort candidates.
Every step in this pipeline is a potential point of failure. If text extraction scrambles your content (common with complex PDFs), everything downstream breaks. If section identification fails because you used creative headings like "Where I've Made an Impact" instead of "Experience," the parser might dump your work history into the wrong field. If entity extraction confuses a company name with a job title, the recruiter sees garbled results.
What major ATS platforms look for
While exact algorithms are proprietary, the major platforms share common expectations:
| ATS Platform | Notable parsing behavior |
|---|---|
| Workday | Strict section header matching; struggles with non-standard layouts; prefers DOCX |
| Greenhouse | Generally flexible; handles well-structured PDFs; good at date extraction |
| Lever | Strong parsing for standard formats; text box content may be missed |
| iCIMS | Handles multiple formats; can struggle with complex table layouts |
| Taleo (Oracle) | One of the oldest platforms; stricter parsing; DOCX strongly recommended |
How keyword matching actually works
Recruiters search their ATS the same way you search Google - with keywords. "Python developer San Francisco 5 years" might return 200 candidates. The system matches those terms against the structured fields it extracted from each resume.
This is why your wording matters. If a job posting says "stakeholder management" and your resume says "working with internal teams," you are describing the same skill but using different words. The recruiter searching for "stakeholder management" will not find you.
However - and this is important - the solution is not to stuff your resume with keywords. Modern ATS platforms and recruiters both penalize obvious stuffing. The solution is to use industry-standard terminology for skills you genuinely have. Read the job posting, note the key terms, and check whether your resume uses those same terms naturally. If it does, great. If not, adjust the wording to match - but only for skills you actually possess.
The section header problem
Section headers are surprisingly important. The parser uses them to understand the structure of your document. Standard headers that work everywhere: "Experience" or "Work Experience," "Education," "Skills," "Certifications," "Summary" or "Professional Summary," "Projects."
Headers that cause parsing confusion: "My Journey," "Toolbox," "What I Bring," "Career Highlights" (could be confused with Summary), "Professional DNA." These are not inherently bad labels - they just do not match the patterns the parser expects, which can result in content being miscategorized or ignored.
How dates get extracted
Date extraction is one of the most failure-prone steps. ATS parsers expect dates in recognizable formats: "January 2020 - March 2023," "Jan 2020 - Mar 2023," or "01/2020 - 03/2023." Formats like "2020-23" or just year ranges without months can lead to incorrect duration calculations.
This matters because some recruiters filter by years of experience. If the parser miscalculates your tenure because the dates were ambiguous, you might be filtered out of searches you should appear in.
Always use explicit month-year formats, placed consistently in the same position for every role. Right-aligned or on the same line as the job title works best.
What happens after parsing
Once your resume is parsed into structured data, it enters the recruiter's workflow. At this point, the ATS is no longer making decisions - the recruiter is. They search, filter, and scan the list of candidates. Your parsed profile competes with potentially hundreds of others.
This is where the "ATS rejection" myth falls apart. The system does not reject you. It organizes you. A recruiter might filter for candidates with specific skills or location, and if your resume did not parse those correctly, you are invisible - not rejected. The distinction matters because the fix is not to "beat the ATS" but to make sure your information reaches the recruiter intact.
AI scoring - the emerging layer
Some newer ATS platforms and add-on tools now include AI-powered scoring. These systems analyze your parsed resume against the job description and generate a relevance score. Recruiters can then sort candidates by score.
The scoring algorithms are generally looking for keyword overlap, role title alignment, experience level match, and skill coverage. Some also factor in recency - giving more weight to recent experience.
What they do not (currently) do well: evaluate the quality of your writing, assess cultural fit, or understand nuance. A strong resume still needs to convince a human in the end. The AI scoring just affects whether you make it to the human's screen.
Practical takeaways
- Use standard section headers so the parser can map your content correctly.
- Keep your layout single-column and text-based - every element should be selectable text, not an image or graphic.
- Use industry-standard terms for skills and job functions. Mirror the job posting's language where it honestly describes your experience.
- Format dates consistently with month and year.
- Place contact information in the body of the document, not in headers, footers, or text boxes.
- Test your resume by pasting it into a plain text file to verify reading order.
For related technical guidance, see PDF vs Word for ATS and Can ATS Read Tables, Columns, Icons, and Graphics?. For putting it all together into a clean layout, use the best resume format guide.
Trusted external resources
Useful next steps
Understanding how the parser works is the foundation. The guides below apply that knowledge to specific decisions you will face - choosing the right file type, picking a layout, and making sure the formatting survives the system.
- The ATS-Friendly Resume: What Actually Matters in 2026
- Can ATS Read Tables, Columns, Icons, and Graphics?
- PDF vs Word for ATS: Which File Type Is Safer?
- Best Resume Format for ATS
Frequently Asked Questions
Do all companies use an ATS?
Most companies with more than 50 employees use some form of ATS. Smaller companies often manage applications through email or simpler tools. If you are applying through an online portal with a structured form, there is almost certainly an ATS behind it.
Can an ATS read handwritten resumes?
No. Handwritten or scanned image documents cannot be parsed by ATS platforms. Always submit a digital, text-based document.
Does the ATS score my resume automatically?
Not all of them. Traditional ATS platforms just parse and store - the recruiter does the evaluation. Newer platforms and add-on tools may include AI scoring, but a human still makes the final decision.