ATS score: what the numbers actually say

"75% of CVs are rejected by a robot before a human ever sees them." You have certainly come across that sentence. Business magazines, recruitment agencies and hundreds of blogs repeat it. It has one flaw: nobody has ever produced the study behind it.

A statistic with no original

Follow the citation chain and you arrive at Preptel, an American CV-optimisation software vendor that circulated the figure in its marketing around 2012. The company shut down in August 2013 without ever publishing a methodology, a dataset or a study. The number was then recycled from outlet to outlet, each citing the last.

One clue is enough to be sceptical without being a statistician: depending on the site, the same "fact" is 70%, 75% or 88%. A real measurement does not drift like that. It is the signature of an orphan statistic.

Note also who keeps it in circulation. The first page of search results on the topic is almost entirely occupied by CV-optimisation vendors — the sellers of the cure describing the disease.

What APEC actually measures in France

The public debate about ATS is an American import. Solid French data does exist. Each year APEC surveys a representative sample of companies that have hired at least one manager — 1,150 companies for the 2025 edition. The results are unambiguous:

• Half of companies with 250 or more employees use recruitment software. In small and mid-sized firms, that drops to 17%.
• 78% of companies hiring managers make do with standard office software — an inbox and a spreadsheet, in other words.
• Only 4% report having integrated AI tools into their process, and "in most cases limit themselves to using them for writing job ads".

Source: APEC, Pratiques de recrutement de cadres 2025. The trend is rising — 13% of mid-sized and large companies reported AI use in 2025, against 6% a year earlier — but it starts from a very low base.

If you are applying to a small or mid-sized company in France, the odds your CV is screened by an algorithm are low. It will very probably be opened by hand.

"Filtering" does not mean "rejecting"

In large American, British and German companies, usage is indeed widespread: more than 90% of surveyed employers say they use their software to filter or rank applications, according to the Harvard Business School study Hidden Workers: Untapped Talent (2021, over 2,250 executives surveyed).

Two things to keep in mind. France is not in that sample: these figures do not describe your market. And filtering or ranking is not eliminating — sorting a pile does not empty it.

The same work describes the exclusion mechanism precisely, and it has nothing to do with keyword density. The software relies on negative logic: candidates are screened out when they fail a binary criterion — no required degree, a gap in employment history, no work authorisation. These are the familiar knockout questions on the application form, not your CV's layout.

The 7.4 seconds: a human pass, not an algorithmic one

Another ubiquitous figure: recruiters supposedly spend 7.4 seconds on a CV. It comes from a Ladders press release (2018), an American job site. The release discloses neither the number of recruiters, nor the number of CVs, nor the exact protocol, and the study was never published in a peer-reviewed journal.

But note this above all: those 7.4 seconds describe a human being skimming a CV. It is nonetheless routinely invoked as proof of automated screening. It is exactly the opposite.

In France, the documented bias is human

If unfair CV screening is the worry, French data points elsewhere. A study conducted for DARES by the Institut des politiques publiques and ISM Corum sent 9,600 applications in response to 2,400 real job openings. The finding: at strictly comparable quality, applications whose identity suggests North African origin are 31.5% less likely to be contacted.

Source: IPP Note no. 76 (2021). No algorithm is involved here: these are human decisions.

What regulation changes

The European AI Act explicitly classes systems intended to "analyse and filter job applications and evaluate candidates" as high-risk (Regulation (EU) 2024/1689, Annex III, point 4). The corresponding obligations — documentation, human oversight, traceability — apply from 2 August 2026. A simple CV parser may fall outside; a tool that scores and ranks candidates does not.

So what should you actually do?

The genuinely useful advice is less spectacular than anti-robot recipes, but it holds up:

• A properly readable file. A PDF exported from a word processor, not a scanned image. If copy-pasting your CV produces gibberish, a machine will read gibberish.
• Sections with standard headings ("Work experience", "Education", "Skills"), which help automated extraction and a human skim alike.
• The posting's vocabulary, when it genuinely describes what you did. That is the principle behind tailoring your CV to each ad.
• Answer the form's questions honestly: that is where the only well-documented automatic rejection happens.
• No keyword stuffing, least of all in white text. It is detectable, and unreadable for the person who reads next.

And one point the anti-ATS guides forget: according to APEC, in 41% of cases the manager ultimately hired was already known to the company or recommended. And the top difficulty recruiters report is not filtering a flood: it is the mismatch of the applications they receive (75% in 2026, ahead of insufficient volume at 73%). An hour spent targeting your application or tending your network often beats an hour chasing an imaginary score.

In short

There is no universal "ATS score", and no serious tool can promise you one. What does exist is a clear, honest CV consistent with the posting — which serves the machine when there is one, and the human in every case.

That is exactly what Cvnetic does: structure your CV and align it with each posting, with no magic promises. Try it on your current CV.

Sources

• APEC, Pratiques de recrutement de cadres 2025 (survey of 1,150 companies, May 2025).
• Fuller, Raman, Sage-Gavin & Hines, Hidden Workers: Untapped Talent, Harvard Business School & Accenture, 2021.
• ISM Corum & Institut des politiques publiques for DARES, IPP Note no. 76, Discrimination à l'embauche selon l'origine, 2021.
• Regulation (EU) 2024/1689 on artificial intelligence, Annex III.
• Ladders Inc., eye-tracking study press release, 2018 — methodology unpublished.
• Uncharted Career, investigation into the origin of the 75% myth.