What Happens When Recruitment Stops Parsing CVs and Starts Asking AI?

The PDF CV was built for an era when recruitment software needed one predictable file. Large language models can understand a career across several sources, which means the document is losing its claim to be the centre of hiring.

By Javier Martínez, founder and CEO of Self

A PDF is excellent at preserving the layout of a document. It is less impressive at preserving the shape of a career.

The CV became the standard input for recruitment technology because it was portable, familiar and easy for software to store. Applicant tracking systems learned to extract names, dates, job titles and keywords from it. Candidates then learned to reshape their experience around whatever those systems could recognise.

That compromise now looks increasingly dated. Recruitment teams are using more capable technology, yet candidates are still asked to compress years of work into a static file designed to pass through a parser.

The CV was a technical compromise

A CV is a snapshot taken at the moment it is exported. It may already be out of date when a recruiter opens it. It rarely shows the work behind its claims, and every addition forces something else to be shortened or removed.

Projects become bullet points. Publications become titles. A changing role becomes two dates and a job description. People with varied careers are asked to flatten them into the same template, then optimise the wording for software before a person reads it.

This is a poor foundation for understanding someone. A CV can summarise a career, but it cannot contain one. Treating it as the complete candidate record confuses a convenient file format with the person it describes.

A career is already bigger than one file

Much of the evidence that gives professional experience meaning already lives elsewhere. It can appear on a personal website, portfolio, project page, professional profile, code repository, publication page or industry platform.

Older recruitment software struggled with that variety. Large language models can interpret different formats together. They can connect a project to a role requirement, recognise related skills expressed in different language and summarise evidence across several candidate-approved sources.

The future ATS should therefore accept more than an attachment. A candidate could provide a CV, if one exists, alongside a selected group of professional links. The system could build a current view of their experience and show the source behind each conclusion. A recruiter could see the claim, the evidence and any uncertainty in the same place.

At Self, we turn existing CV material into a personal website that a candidate can review, update and share. That is one example of a wider move from career documents towards career information that lives on the web.

“The CV made sense when software needed every career flattened into the same predictable file,” says Javier Martínez, founder and CEO of Self. “LLMs no longer need that compromise. They can understand someone’s work across the professional evidence that person chooses to share.”

More information requires better boundaries

This argument is not an invitation for employers to scrape everything they can find. Public information may be old, irrelevant or personal. Online visibility also varies by profession, age, income, disability and culture. A polished web presence is evidence of presentation, not proof of suitability for every job.

Candidates should decide which professional sources enter the process. They should know how those sources will be used, see what the system inferred and have a way to correct it. Someone with little public work online must still be able to provide equivalent evidence through another route.

The Information Commissioner’s Office made almost 300 recommendations after auditing providers of AI-powered recruitment tools. Its findings emphasised fair processing, data minimisation and clear explanations for candidates. The UK government’s Responsible AI in Recruitment guide similarly asks organisations to define a system’s purpose, outputs and human oversight before buying it.

These are design requirements, not reasons to keep the CV at the centre. The EU AI Act treats many recruitment and selection systems as high-risk because employment decisions have serious consequences. Better context must arrive with provenance, testing, accountability and human judgement.

What the next ATS should do

Recruitment leaders should stop asking only how accurately a system parses a CV. They should ask which sources it can understand, whether candidates control those sources, how it separates evidence from inference and how recruiters can challenge its conclusions.

A useful system would show why a candidate appears relevant instead of producing another unexplained score. It would flag conflicts and missing information. It would help recruiters handle volume by directing attention to evidence, while leaving the final judgement with a person.

The PDF CV may survive for years because existing processes still require it. That makes it a compatibility layer, not the future of candidate information.

The question is no longer whether AI can read a CV more accurately. It is why recruitment should remain organised around one ageing file when a candidate’s professional story already exists beyond it.