NewKalent + Marvin: AI-powered sourcing in your ATS
Book a demoDemo

GuidesAI8 min read

CV parsing:
what a parser does, the formats it reads and its limits.

CV parsing, or automatic CV analysis: what a parser extracts (identity, experience, skills), how it reads a PDF, a Word file or a scan, its most common errors, the difference with screening applications, and what the GDPR requires.

Christophe Hébert, CEO and founderUpdated
Contents

Contents

0% read

CV parsing is software reading a CV automatically: it extracts the candidate's identity, contact details, experience, education, skills and languages, and fills in a structured candidate record. The recruiter no longer retypes the CV, and the text of the CV becomes searchable in their database.

This guide explains what a parser does, how it reads a PDF, a Word file or a scan, where it gets things wrong and what the GDPR requires.

Guide updated on 11 November 2026.

What a CV parser does

A CV parser turns a free-form document into a structured record. It picks out the candidate's name, email, phone number and city, splits their career into dated roles, and lists their education, skills, languages and certifications. It files each piece of information in the matching field of the record, with no retyping.

What a complete parser extracts:

Block What it takes from it
Identity and contact First name, last name, email, phone, city, LinkedIn profile
Headline The current role, or failing that the latest role held
Experience Job title, company, location, start and end dates, description
Education School, degree, field, dates
Skills The skills mentioned, often reused as tags on the record
Languages and certifications Language and level, certification title and issuer

Some parsers also infer information that is not written as such: years of experience, education level, the type of contract sought.

How CV parsing works

Parsing happens in two steps. The software first extracts the text from the file: the text layer of a PDF, the content of a Word document, or character recognition on an image. It then files that text into fields, using rules or a language model, and normalises dates and phone numbers.

Text extraction. A PDF produced by a word processor has a text layer that the software reads directly. A scanned PDF or a photo has none: it takes optical character recognition (OCR) or a model able to read images.

Structuring. Early parsers looked for keywords and patterns ("Experience", an email address, a run of dates). Recent parsers hand the text to a language model, which copes better with an unusual layout or a free-form job title.

Normalisation. A phone number written "06 12 34 56 78" is converted to international format, a date such as "March 2021" becomes a usable date, and a skill spelled three ways can be mapped to the same tag.

Formats: PDF, Word, scans and photos

PDF and Word are the most common formats and the best read, because they carry their own text. The older .doc format is still frequent in CVs coming from job boards. A scan or a photo can only be read if the parser recognises text in an image, and a password-protected document cannot be read at all.

Format Reading Watch out for
PDF with text Direct Columns and tables can scramble the order of the text
Word (.docx) Direct Text boxes and headers are sometimes ignored
Older Word (.doc) Depends on the parser A password-protected file will not open
Scanned PDF, photo (JPG, PNG) Image reading (OCR or vision) One misread character breaks an email or a phone number
Protected file Not possible Ask the candidate for an unprotected version

The limits of CV parsing

A parser mostly gets the form wrong: two-column CVs, tables, contact details in a page header, blurry scans. It also gets the substance wrong when the document holds several names, such as an agency's candidate presentation that carries the consultant's contact details. The record needs checking, starting with the email and phone number.

The most frequent errors:

  • The presenter's contact details. A CV formatted by an agency or a consultancy often carries the email and phone of the consultant presenting the candidate. A naive parser takes them for the candidate's.
  • A misread character on a scan. An "o" read as "0", a "5" read as "S": the email looks valid but reaches nobody. An empty field is better than a wrong address.
  • Ambiguous dates. "2019 - 2021" gives no month; "for 3 years" depends on when the CV was written.
  • Free-form job titles. "Head of Growth" or "Biz Dev Lead" match no job taxonomy without interpretation.
  • A name left over from a template. A downloaded CV template sometimes keeps the sample name in its header.

Parsing, screening and assessment: the difference

Parsing reads the CV and fills in the record; it does not say whether the candidate is a fit. Screening, scoring or automatically ranking applications are separate steps that raise other questions: on which criteria, with which biases, and who decides. The GDPR strictly limits decisions taken without human involvement.

In principle, Article 22 of the GDPR prohibits a decision with a significant effect on a person, such as being rejected from a hiring process, from being based solely on automated processing. A score can help a recruiter prioritise; it must not reject in their place. Our article on CV retention and the GDPR covers this point, and our guide to recruitment scoring explains how to build a scorecard.

The GDPR and CV parsing

A CV contains personal data, and parsing is processing within the meaning of the GDPR. The agency or company remains the controller: it informs the candidate, keeps only the data useful for recruitment and binds its parsing provider, which acts as a processor, by a contract that sets what it may do with the CVs.

Three questions to ask a parsing provider:

  1. Where are CVs processed, and by which sub-processors, AI providers included?
  2. Are CVs used to train models? The answer must be written into the contract.
  3. What happens to the files after parsing, and how long are they kept?

On the candidate's side, the information owed is the same as for any hiring process: who processes their data, why, for how long, and how to exercise their rights.

In Marvin

Marvin reads CVs in PDF, Word (.docx and the older .doc) and image formats (JPG, PNG, WebP, GIF), up to 10 MB; a scanned PDF or a photo goes through image reading. Reading fills in the record: identity, contact details, role, experience, education, skills, languages and certifications. The text of the CV becomes searchable in your database.

You can also have AI fill in a custom field from an instruction, for example experience outside the European Union. See also Marvin's cv database software.

CV reading, like Marvin's other AI features, runs by default on OpenAI models, with providers listed in the DPA, and your data trains no model. Hosting the AI models at Scaleway, in Paris, is planned: the details are on the responsible AI page.

Frequently asked questions

What is CV parsing?

It is software reading a CV automatically and extracting the candidate's identity, contact details, experience, education, skills and languages to fill in a candidate record. The recruiter no longer retypes the CV, and its text becomes searchable in the database.

Can a CV parser read scanned CVs?

Only if it can recognise text in an image, through optical character recognition or with a model able to read images. A scan has no text layer: a parser that only reads the text of a PDF returns nothing. On a blurry scan, check the extracted email and phone number.

Which CV formats can a parser read?

PDF and Word (.docx, and often the older .doc) are the usual formats. Images (JPG, PNG) and scanned PDFs require image reading. A password-protected document cannot be read: ask the candidate for an unprotected version.

Does CV parsing screen candidates?

No. Parsing reads the CV and fills in the record; it decides nothing. Screening, scoring or ranking applications are separate steps. If software rejects a candidate without human involvement, Article 22 of the GDPR applies: in principle, it is prohibited.

Is CV parsing GDPR compliant?

It can be. The employer or agency informs the candidate that their data is processed, the parsing provider is a processor bound by a contract, and the data is used only for recruitment. Also check where CVs are processed and whether your provider commits not to use them to train its models.

Christophe HébertCEO and founder

CEO and founder of Marvin. A former recruiter turned tech entrepreneur, he is building the operating system for modern recruiting.

Never lose an application again. See how, in 30 minutes.