What is resume parsing? How it works, types and accuracy (2027)
Resume parsing turns CVs into searchable ATS fields. Learn how parsers work, the main types, how accurate they are and where they fail.

Every recruiter has opened a candidate record where the job title sits in the name field. Resume parsing is the step that decides whether your ATS data looks like that.
Resume parsing is the automated extraction of information from a resume or CV into structured fields, such as name, contact details, work history, education and skills. A parser organizes what the candidate wrote and does not decide whether the candidate is qualified.
Below, you will see how parsers work, which types exist, how accurate they are and where they fail. You will also see which rules apply and how to verify skills once parsing is done.
TL;DR:
- Resume parsing automatically extracts resume data such as contact details, work history, education and skills into structured ATS fields. It organizes claims; it does not judge qualifications.
- A parser runs five stages: ingest, extract, segment, normalize and output. An early error, such as merged two-column text, carries into every later stage.
- Contact details parse most reliably. Skills, certifications and dates in unusual layouts parse worst.
- Vendor figures put keyword parsers near 70% accuracy and grammar-based parsers above 90%. Test 30 to 50 of your own resumes before you trust any number.
- For design, marketing or international roles, choose an ML, cloud-based or LLM-based parser over a rule-based one.
- Once parsed fields feed a score, hiring rules apply. NYC Local Law 144 requires a yearly bias audit, and EU AI Act Annex III rules apply from 2 December 2027.
- Keep filters loose and never rank on keyword match. Send candidates who pass must-haves to a skills test, and audit 20 to 30 records each quarter.
What is resume parsing?
Resume parsing is the process of converting an unstructured resume file into a structured candidate record. A resume parser reads a PDF, DOCX, HTML, plain-text or image file, finds each piece of information, labels it, and returns it as JSON or XML that your applicant tracking system (ATS) or HRIS can store. CV parsing means the same thing; the name changes with the region.

Parsers are built into most ATS platforms and are also sold as standalone APIs for teams that need more accuracy, more languages or custom fields. A parser extracts the following data, and some fields parse far more reliably than others.
Field | What the parser extracts | How reliably it parses |
|---|---|---|
Contact details | Name, email, phone, location, profile links | High |
Work history | Employer, job title, start and end dates, duties | Medium: dates and titles vary by layout |
Education | Institution, degree, field of study, graduation date | Medium to high |
Skills | Hard skills, tools, languages spoken | Medium to low: synonyms and abbreviations |
Certifications | Licences and certificates | Medium to low: depends on the parser's list |
Other | Summary, availability, salary expectation, work authorization | Varies |
Enterprise parsers also infer attributes, such as total years of experience, seniority level and career trajectory. These inferred values help ranking but can act as proxies for age or other protected traits, so review them before they feed any score.
Key takeaway: Resume parsing does not evaluate candidates. The output is only as accurate as the resume and the parser, so treat parsed data as a list of claims to check, not as evidence of skill.

How does resume parsing work?
Resume parsing works in five stages: the parser ingests the file, extracts the text, segments it into sections and entities, normalizes the values, and writes a structured record to the ATS. An error at an early stage carries through to every later stage, which is why layout problems hurt so much.
Ingest
The parser receives the file, detects the format and routes it to the right engine. Scanned PDFs and photos need optical character recognition (OCR) first, which adds errors when the scan is poor.
Extract
The engine pulls all raw text and keeps a rough idea of position, such as header, left column or right column. Two-column layouts and tables often break here because text from both columns gets merged into one line.
Segment
The parser labels sections (contact, summary, experience, education, skills) and entities inside them (employer, job title, date range, degree). Rule-based parsers look for headings like "Work Experience"; machine learning parsers use natural language processing and named entity recognition trained on large sets of resumes.
Normalize
Raw values are converted to standard formats. "Sep 2021", "09/21" and "September 2021" all become 2021-09. Job titles and skills are mapped to a taxonomy such as O*NET, so "JS" and "JavaScript" count as the same skill.
Output
The parser returns a JSON or XML object, or writes fields straight into the ATS. Recruiters can then search and filter, for example all candidates with three or more years of Python and a degree in computer science.
Many parsers also return a confidence score per field. Send low-confidence fields to a short human review queue instead of letting them flow straight into filters.
What are the types of resume parsers?
Resume parsers come in six main types: keyword, grammar-based, statistical, rule-based, machine learning and cloud-based. Most commercial products now mix several methods, and a newer group uses large language models (LLMs) to read unusual layouts.
The type decides how well a parser copes with non-standard resumes, informal wording and other languages.

Keyword parsers
These are the fastest and cheapest but miss "Python 3.12" or "PyTorch" if the list only says "Python". Grammar-based parsers: These read how words relate, so they handle varied phrasing and non-US section headings.
Statistical parsers
These assign probabilities to each field instead of following fixed rules. Rule-based parsers: These are predictable and easy to customize, but someone has to update the rules whenever resume styles change.
Machine learning parsers
These learn from labelled examples and handle informal language, creative layouts and multilingual resumes better than older methods. Cloud-based parsers: These deliver those models as a service and usually plug into an ATS through a REST API.
LLM-based parsers
Now offered by several parsing vendors for complex documents, these read context well and can pull custom fields. They still depend on clean text input and still need testing.
Pro tip: If you hire for design, marketing or international roles, choose an ML, cloud-based or LLM-based parser over a rule-based one. Visual CVs with columns make rule-based parsers merge unrelated fields, and those broken records quietly drop out of searches.
Why is resume parsing important for recruiters?
Resume parsing matters because it turns a pile of differently formatted files into one searchable candidate database, so recruiters stop copying data by hand and start comparing candidates on the same fields. The time it frees goes back into talking to candidates and hiring managers.

Recruiting is where most HR teams use automation first. In SHRM's State of AI in HR 2026 research, 27% of organizations apply AI in talent acquisition, more than in any other HR function, and screening and matching resumes is one of the most common uses.
Adoption reaches 60% at organizations with 5,000 or more employees. SHRM's 2024 Talent Trends survey found that 64% of employers using AI in HR use it for talent acquisition.
Resume parsing helps hiring teams in four ways:
- Time: A parser processes a resume in seconds, so a high-volume role no longer means days of data entry before the first interview.
- Fewer errors: Retyping resume data into an HRIS introduces typos and missed fields. Parsing removes the manual copy step.
- Consistency: PDF, DOCX and HTML resumes all become the same fields, so candidates can be filtered and compared on equal terms.
- Languages: Many commercial parsers read resumes in dozens of languages. Textkernel, for example, lists 29 resume languages on its parser page.
How accurate is resume parsing software?
Resume parsing accuracy depends on the parser type, the resume format and the field. Contact details parse best; skills, certifications, and dates in unusual layouts parse worst. Most published accuracy figures come from parsing vendors, so treat them as upper limits and test on your own resumes.
The figures vendors publish give a rough range. Daxtra's parsing guide puts keyword parsers at about 70% accuracy and grammar-based parsers above 90%, and notes that human accuracy is rarely above 96%.
Affinda reports 92% overall on its own strict benchmark and 97% or more on contact fields. Daxtra and several buyer guides treat about 90% as the minimum before manual correction eats the time saved.
Accuracy drops on scanned PDFs, two-column visual CVs, non-English resumes read by English-trained models, and resumes heavy with abbreviations. It also drifts over time as resume styles change. To measure accuracy yourself, run 30 to 50 real resumes from your own applicant pool through the parser and check three numbers:
Measure | How to calculate it | What it tells you |
|---|---|---|
Field accuracy | Correct fields divided by fields the parser filled | How often parsed data is right |
Field completeness | Fields captured divided by fields present in the resume | How much the parser misses |
Correction time | Minutes a recruiter spends fixing one record | Whether parsing actually saves time |
Split the results by format (PDF, DOCX, scan) and by language, because an overall average hides the weak spots.
What are the challenges of resume parsing?
Resume parsing has five main challenges: layout and language errors, keyword stuffing, unverified self-reported data, cost at scale, and bias from inferred attributes. Each one can remove good candidates or let weak ones through before any person reads the resume.
Layout and language errors
Even ML parsers misread creative layouts, dense abbreviations and the many date formats candidates use ("Present", "current", "to date"). One wrong end date changes calculated years of experience and every ranking built on it.
Keyword stuffing
Many ATS platforms rank by keyword match, so some candidates repeat target terms or paste the job description into the resume. A keyword parser cannot tell a real project manager from someone who typed "project management" 15 times.
Unverified claims
Parsers extract what candidates wrote and never check it. Inflated titles, stretched dates and skills listed without real ability all pass through. Our guide on how to detect fake resumes covers the warning signs.
Cost
Enterprise parsers are priced per parse, per document bundle or per seat, so at tens of thousands of applications a month, parsing becomes a budget line to justify.
Bias and data use
Parsers and ranking models trained on past hires can favor proxies such as ZIP code, school name or employment gaps. The UK regulator found real cases in its audit of AI recruitment tools. The ICO saw tools that let recruiters filter out candidates by protected characteristics, tools that guessed gender and ethnicity from names, and tools that kept far more data than needed, indefinitely. The providers accepted or partly accepted all of its almost 300 recommendations.
What do privacy and AI laws require for resume parsing?
Privacy and AI laws regulate how parsed candidate data is collected, used and kept, and how automated screening built on it is audited. Parsing alone is a data step, but once parsed fields feed a score or an auto-reject rule, hiring rules for automated decisions apply.
New York City
Under Local Law 144, enforced by NYC's Department of Consumer and Worker Protection since July 5, 2023, an employer can use an automated employment decision tool only if it had a bias audit within the past year, the audit summary is public, and candidates get notice. Testlify publishes its own NYC bias audit.
European Union
The EU AI Act lists AI used in employment, including tools that filter applications and evaluate candidates, as high-risk under Annex III. After the AI Omnibus, the European Commission's AI Act timeline shows the Annex III rules applying from 2 December 2027, so EU hiring teams have a short window to document risk controls, human oversight and logs.
GDPR and UK GDPR
You need a lawful basis to process candidate data, clear notice that resumes are parsed and why, a defined retention period (Article 5(1)(e)), and a way to honor deletion requests.
United States, federal
No federal rule requires bias audits of resume parsers. Federal anti-discrimination law still applies to outcomes, so a parsing or ranking rule that screens out a protected group at a higher rate can create disparate impact risk.
What is the difference between resume parsing and resume screening?
Resume parsing extracts and structures data; resume screening evaluates that data against job criteria. Parsing comes first, so a parsing error, such as a dropped certification or a misread date, flows into every screening decision built on that record.
The two often live in the same ATS but are separate steps, technically and legally.
Step | What it does | Output | Typical tool |
|---|---|---|---|
Resume parsing | Extracts and structures resume data | Candidate record with fields | ATS parser or parsing API |
Resume screening | Applies criteria, filters or scores to the record | Pass, reject or a fit score | ATS rules, AI resume screener |
Candidate matching | Compares the profile to a job or to other profiles | Ranked list or relevance score | ATS search, matching engine |
Skills assessment | Tests what the candidate can actually do | Verified skill scores | Assessment platform |
For more on the evaluation step, see these resume screening techniques.
How do you choose a resume parser?
Choosing a resume parser starts with your own applicant pool: the formats, languages and roles you actually receive. Ask every vendor to parse a sample of your real resumes, then compare field accuracy and correction time, not marketing percentages.
Check these points before you buy:
- ATS fit: Does it plug into your ATS natively or through an API, and does parsed data sync both ways?
- Accuracy method: Does the vendor report accuracy by field, and does a partly correct field count as wrong?
- Formats and layouts: Test scans, photos, two-column CVs and DOCX files with tables.
- Languages: List the markets you hire in and confirm each language for resumes, not only for the interface.
- Taxonomy: Can skills and titles map to O*NET, ESCO or your own skills framework?
- Data handling: Ask about data residency, retention controls, whether your data trains their models, and security certifications.
- Bias controls: Can you switch off or redact fields such as name, photo, graduation year and address before scoring?
How do you implement resume parsing in your organization?
Implementing resume parsing well takes five steps: audit what your ATS already does, define the fields you need, set up bias and data controls, then add a verification step so parsed claims are checked before the shortlist.
- Audit your current ATS parser: Workday, Greenhouse and Lever all include built-in parsing, and results differ by layout. Run 20 to 30 sample resumes in your common formats through it to set a baseline before you pay for a third-party parser.
- Define the fields your workflow needs: Map parsed fields to your screening criteria and decide which fields, if any, trigger an automatic invitation to pre-employment tests.
- Set a bias audit routine: Track selection rates by group at each automated step, suppress proxy fields such as ZIP code and graduation year, and name an owner for the results.
- Configure data retention: Set a retention window per requisition, automate deletion, and document the lawful basis for EU and UK candidates before parsing starts.
- Add a verification layer: Send candidates who pass must-have filters to a role-specific skills test, then review parsed data and test scores together.
What are best practices for more accurate resume parsing?
Better resume parsing comes from two places: how you set up the parser and what you ask candidates to submit. The practices below fix the most common failures at little or no cost.

Seven habits improve parsing results without a new tool. The first two fix file quality and setup, and the rest protect your screening decisions and keep errors visible over time.
- Ask for machine-readable files: Request PDF (not scanned) or DOCX in the application form. Image-only files force OCR and lower accuracy.
- Tune the skills taxonomy to the role: Generic skill lists miss niche terms. For technical roles, map the parser to the exact stack in the job description, versions included where they matter.
- Audit a sample every quarter: Check 20 to 30 parsed records against the source files and track errors by field. Dates and non-English characters fail most often.
- Keep filters loose: Strict keyword filters reject people with non-standard job titles or career paths. Filter only on true must-haves and rank the rest on verified skills.
- Do not rank on keyword match: Keyword scores reward resume tuning more than ability. Use them to check minimum requirements, then rank on structured interviews or skills tests.
- Pair parsing with ability tests where learning matters: Parsed history shows what someone did, not how fast they learn. For roles that will change, add cognitive ability tests.
- Track parse error rate as a KPI: A 5% error rate across 10,000 applications means 500 distorted candidate records entering your pipeline every cycle.
How does Testlify fit with resume parsing?
Testlify adds the verification step that parsing leaves out. The Testlify AI resume screener pulls resumes from your ATS and scores each candidate against role-specific criteria.
Candidates then take a test from the Testlify test library of 3,500+ tests. Assessment results sync back to your ATS through 100+ ATS and HRIS integrations.
Recruiters see parsed history and proven skills on one record.
Ready to verify skills after resume parsing?
Resume parsing gets candidates into your system, and skills tests show who can do the job. Book a demo to see how Testlify adds a verification step, or start a 7-day free trial, no credit card needed.
Frequently asked questions (FAQs)
Co-Founder and CTO
Jiten Modi is Testlify's CTO and leads a 50-person engineering organization. He writes on technical hiring, evaluating developers with real-world assessments, and building engineering teams.
LinkedInRelated resources
View all
HR & recruitment
What is 360 recruitment? A guide for recruiters and TA teams

HR & recruitment
What is a competency framework and how to develop one?

HR & recruitment
HR digital transformation: a 2026 guide for HR leaders

HR & recruitment
Talent Lifecycle: 5 Stages and HR’s Role (2026 Guide)

HR & recruitment
The five main responsibilities of a human resources director

HR & recruitment
Recruitment statistics 2026: 50+ numbers every hiring team should know
Get started.
Hire on proof, not resumes.
Run your first skills-based assessment free — no credit card required.