Use case assessment
Is AI resume parsing worth building?
Parsing and screening are different things with different obligations, and most tools do both.
Extracting structured fields from a CV is ordinary extraction. Ranking candidates is Annex III high-risk. The same product usually does both, and the toggle is not obvious.
What it usually means
Reading a CV and populating structured fields: name, contact details, roles, dates, skills, education.
Is it genuinely AI?
Yes, and it is well solved. CVs arrive in every format anyone has ever invented, and extracting a consistent structure from that variety is a real model problem with mature products.
The distinction that decides your obligations
Parsing extracts. Screening evaluates. They are different activities and the EU AI Act treats them differently.
Populating an applicant record with the roles someone listed is administrative. Ranking that applicant against others, scoring their fit, or filtering them out is recruitment and selection — explicitly inside the Annex III high-risk area covering employment and access to self-employment.
The problem is that almost every applicant tracking system ships both, and the scoring feature is often on by default or one configuration away. A parsing procurement can become a high-risk deployment without a decision being recorded.
Check what your system actually does with the extracted fields.
What the simpler version looks like
Structured application forms. If you need the last three roles with dates, ask for them as fields rather than extracting them from a document the candidate formatted themselves.
Candidates prefer it, the data is cleaner, and parsing becomes unnecessary rather than automated.
What it costs to run
Extraction accuracy degrades on unusual formats and non-Latin scripts, and those failures are not random — they cluster on particular candidate populations, which is a fairness problem before it is a technical one.
Budget for review of low-confidence extractions, and check error rates across candidate groups rather than in aggregate.
Where it sits under the EU AI Act
Parsing alone: not Annex III. Parsing plus any evaluation, ranking or filtering: Annex III high-risk, with obligations applying from 2 December 2027 following the Digital Omnibus amendments.
And under Article 25, branding a purchased tool as your own or changing its intended purpose can transfer provider obligations to you.
When it is worth building
Buy it, and configure it deliberately. Worth it at volume where structured forms are not practical — senior hiring, agency submissions.
Not without knowing exactly which features are active. The parsing decision is low risk. The screening decision that arrives bundled with it is not.
Common questions
Is resume parsing the same as CV screening under the EU AI Act?
No. Parsing extracts structured fields and is administrative. Screening evaluates, ranks or filters candidates, which falls inside the Annex III high-risk area covering employment and access to self-employment.
Why does the distinction matter?
Because almost every applicant tracking system ships both, and scoring is often on by default or one setting away. A low-risk parsing procurement can become a high-risk deployment without any decision being recorded.
What is the alternative to resume parsing?
Structured application forms. If you need the last three roles with dates, ask for them as fields rather than extracting them from a document the candidate formatted themselves. Cleaner data, and parsing becomes unnecessary.
Scoring 40 AI use cases in a regulated enterprise — where a default-on feature changed the risk classification of an approved project.