Automated resume screening was supposed to save recruiting teams hours. In 2026, more than fourteen heads of recruiting at large French groups call it useless “as it stands”. The complaint is less about the algorithm than about what it is fed: a vague brief, implicit criteria, no feedback on rejections. A candidate score only helps a decision under five conditions, detailed below.
A Head of Talent Acquisition receives 300 applications for a logistics manager role. The screening tool keeps 12. Nobody can say why those 12, or what disqualified the other 288. Two candidates later hired for other roles were sitting in the rejected pile. The question is no longer “is the AI faster?” but “what did it decide on?”.
What do recruiting leaders hold against automated resume screening?
The verdict is harsh and it comes from practitioners. More than fourteen heads of recruiting at large French groups, interviewed by the business daily Les Echos and relayed by the newsletter Les talents narratifs (2 April 2026), consider scoring tools useless as they stand. One of them checked: his tool would have screened out candidates he eventually hired.
Three grievances come back in these accounts:
- Opacity: a score appears, but no readable criterion explains why one profile surfaces and another disappears.
- Rigidity: non-linear careers, career changes and gaps are read as flaws, when they are sometimes the very signal the role needs.
- No learning: when a recruiter rejects a profile, the tool retains nothing. The next shortlist repeats the same mistakes.
The debate running on LinkedIn this week follows the same thread: an AI sourcing tool that proposes ten candidates in thirty seconds without saying why does not save time, it moves the sorting somewhere else.
Why AI candidate scoring fails: the data, not the algorithm
The same accounts point to one shared cause: the raw material. A resume is not structured data. An ageing ATS makes it worse by parsing fields badly. A candidate's potential does not appear on any line.
Then comes the brief. In most cases the tool receives a job title, a city and a number of years of experience. It does exactly what it is told: it finds people who look like the job description. Not people who would succeed in the company. The most quoted comment in this week's LinkedIn debate sums it up: if an AI returns fifty wrong profiles, the problem is the brief.
Last factor: implicit criteria. The hiring manager knows she does not want someone from the medical sector, or that support functions are out. She never wrote it down. The tool cannot guess what was never stated.
Discrimination and the EU AI Act: when the risk becomes regulatory
The subject has changed in nature. “We are industrialising discrimination,” warned HR consultant Tania Ocana in an interview published by Les talents narratifs on 14 July 2026. A biased criterion applied by hand affects a few applications; applied by a system to thousands of resumes, it becomes systemic.
The law is catching up. The EU AI Act classifies AI systems used for recruitment and candidate selection as high-risk, with obligations of human oversight, traceability and documentation (Apec, the French executive employment agency, 3 September 2026). GDPR, for its part, has framed since 2018 any decision with legal or similarly significant effects based solely on automated processing (Article 22).
For a buyer the consequence is concrete: a tool that eliminates candidates without a human being able to explain and challenge the decision is a legal risk, not only a quality risk.
📄 A score your DPO and your recruiters can both read. The white paper on context enrichment: visible criteria, human validation, learning from rejections. PDF format.
Read the Hiring Graph white paper
What works instead: AI prepares, people decide
The companies that do well have not given up on AI. They have changed its place in the process. The accounts gathered by Les talents narratifs (2 April 2026) describe the uses that hold: drafting job ads, preparing structured interviews, keyword sourcing, handling volume with voice AI. A retailer hiring more than a thousand permanent employees a year uses no automated pre-screening. An HR director at a small company ranks 78 applications with AI, then calls the candidates herself.
The common point: the tool orders and documents, the person decides. That is the difference between automated candidate rejection and assisted ranking. The first decides for you. The second saves you reading time, not decision time.
AI candidate scoring: the 5 conditions for it to be useful
1. Criteria the recruiter can read and edit
The score must break down criterion by criterion, with visible weights. If the hiring manager wants to raise “experience in an industrial environment” from 15 to 30 percent, she must be able to do it herself, without a ticket to the vendor.
2. No rejection without human validation
A low-scored profile stays visible. It is ranked lower, not deleted. The recruiter can move it up in one click if she sees something the system missed.
3. A system that learns from rejections
Every reasoned accept or reject must recalibrate the next search. Rejecting three profiles as “too senior” should make an experience ceiling appear in the criteria. That is the principle of context enrichment: the system accumulates context about your company instead of staying frozen on day one.
4. A trace for every decision
For each candidate: which criteria weighed in, what evidence in the career path, who validated, when. That is what an audit will ask for, and what lets you answer a candidate who challenges the outcome.
5. A score that ranks, never eliminates
The shortlist is a reading order. The last name on the list still exists. This rule alone resolves most of the grievances voiced by the recruiting leaders quoted above.
Umamy's Dynamic Fit Score applies these five conditions: context-enriched scoring, readable criterion by criterion, recalibrated by every recruiter decision, and never dropping a profile on its own. For SMBs comparing tools, see the LinkedIn Recruiter alternatives compared.
What we've seen
At a large consulting firm, we calibrated a search for consultants with an industrial background in late October. The first reviews surfaced several profiles rejected for the same reason: too much experience, 23 to 32 years of career against a 10 to 20 year target. The score did not see the problem because nobody had written it down: the scorecard had an experience floor, no ceiling. We added the range, excluded support functions and set a minimum number of years in industry. The client-side lead reviewed his profiles in 10 to 15 minutes and compared the exercise to the continuous briefing he runs with a headhunting firm. The reasoned rejection fixed the criterion, not the algorithm.
Frequently asked questions
Is automated resume screening legal in Europe?
It is regulated rather than banned. GDPR (Article 22) gives candidates the right not to be subject to a decision based solely on automated processing that produces legal or similarly significant effects. The EU AI Act adds obligations for high-risk systems, which include recruitment. A ranking that a human validates remains possible; a fully automatic rejection exposes the employer. A legal review is recommended before any rollout.
Can a scoring tool reject a candidate?
Technically yes, and that is precisely what to avoid. A sound tool ranks applications and leaves the decision to reject to the recruiter, with a recorded reason. If the vendor cannot switch off automatic rejection, the regulatory risk and the risk of missing a strong profile add up.
How do you audit a resume screening tool?
Take a role you have already filled, give the tool the brief, and compare its shortlist with the candidates you actually interviewed and hired. Then ask, for three rejected profiles, for the justification criterion by criterion. If the tool cannot produce it, the audit is over.
Where Umamy stands
We built the Dynamic Fit Score to answer “why this candidate?” before “how fast?”. The score reads criterion by criterion, with evidence drawn from the career path, it recalibrates on every accept or reject, and it never removes a profile from the list. Umamy is based in Paris and currently filing for AI Act and GDPR compliance. To see how ranking fits into a search, the AI sourcing agents page walks through the workflow.
📄 Understand a score that explains itself. The Hiring Graph and context enrichment white paper: how a system builds context about your company instead of staying frozen.
Read the Hiring Graph white paper
Sources: Les talents narratifs, “IA et recrutement : un échec ?”, 2 April 2026, citing Les Echos ; Apec, “IA et RH : le virage est déjà pris, l'accélération est à venir”, 3 September 2026 ; Les talents narratifs, interview with Tania Ocana, 14 July 2026 ; Regulation (EU) 2016/679 (GDPR), Article 22.
