AI in recruitment: who really reads your CV?
You send an application and, three days later, a polite, impersonal rejection arrives. Did a human read it? More and more candidates doubt it, and rumour has it that an artificial intelligence sorts everything before a recruiter opens the file. The Swiss reality is more nuanced, and it has been measured: in 2026, 13% of companies use AI somewhere in their recruitment, and 3.6% to pre-screen applications.
This guide explains what software really does with your file, what Swiss and European law guarantee you, the documented limits of these tools, and how to apply for two readers at once, the human and the machine. Along the way, the right way to use AI yourself without it backfiring.
What software does with your application
The ATS, the mailbox
The applicant tracking system centralises applications, stores them and lets the recruiter search them by keyword. Almost every large company has one. It is not AI: it evaluates nothing, it files and retrieves.
Automatic screening
Some tools give each file a score based on the keywords of the ad, years of experience or diplomas, and suggest a reading order. In Switzerland, 3.6% of companies use them for pre-screening.
The analysed video interview
Some services analyse the voice, vocabulary or expressions in a recorded video interview. Use remains marginal in Switzerland and controversial everywhere, because of biases that are hard to control.
AI that writes the job ad
This is the most widespread use: 10.9% of companies use it to write or improve their job offers. If an ad looks smooth and generic, this is often why.
What Swiss law says, and European law
Switzerland has no law on artificial intelligence. But the Data Protection Act, in force since 1 September 2023, precisely covers the case that concerns you: a decision taken by a machine alone.
- Automated individual decision. When a decision is taken exclusively through automated processing and has legal effects for you or significantly affects you, which includes pre-screening or rejecting an application, the controller must inform you (art. 21 para. 1 FADP).
- Your point of view, and a human. You may ask to state your point of view and demand that the decision be reviewed by a natural person (art. 21 para. 2 FADP).
- The logic of the decision. By exercising your right of access, you can obtain confirmation that an automated individual decision exists and the logic on which it is based (art. 25 para. 2 let. f FADP). The information is free of charge and generally provided within 30 days.
- The limits. These rights do not apply if you expressly consented to the automated decision, nor when the decision is directly linked to a contract and your request is granted, which is never the case for a rejection (art. 21 para. 3 FADP). And a tool that merely suggests a reading order to a recruiter who decides for themselves is not an automated decision.
On the European side, the AI Act (Regulation (EU) 2024/1689) classifies systems for recruiting, selecting and evaluating candidates as high-risk. Employers using them will have to ensure human oversight, inform candidates and document the tool. The deadline for these obligations was postponed to 2 December 2027 by Regulation 2026/1744, in force since 27 July 2026. The transparency obligation, telling people that they are interacting with an AI, has applied since 2 August 2026.
This law also concerns Swiss companies: it applies as soon as the output of a system is used in the Union, for instance by a Swiss group recruiting for its European sites with the same tool.
The real limits of the tools
Learned biases
A tool trained on past hires reproduces their flaws. The best-known case is Amazon, which in 2018 abandoned a screening engine after finding that it penalised CVs containing the word “women’s”, because it had learned from ten years of mostly male applications.
Misread CVs
Tables, icons, text in images, scanned PDFs, badly built columns whose text reads out of order: everything that pleases the eye can be unreadable for a text extractor. The software then sees neither your title nor your dates, and you vanish from the search without any human deciding it.
False positives
Keyword screening rewards whoever repeats the right terms, not whoever can do the job. Recruiters know this, which is why the score serves as a reading order rather than a verdict.
Mass-produced applications
Companies estimate that 39% of the files they receive are at least partly written with AI, and 55% of them say they can recognise them by their form or language. An interchangeable letter gets spotted, and it does not work in your favour.
Applying for two readers: the human and the machine
A clear structure
Standard headings (Experience, Education, Skills, Languages), real text rather than images, a PDF generated from a word processor rather than scanned. A column layout is fine as long as the text extracts in the right order: copy and paste your PDF into a text editor to check. Your name, position and dates must be readable as plain text.
The words of the ad
Use the job titles and skills as written in the offer, when they match what you did. No stuffing: a list of keywords with no facts behind them is spotted immediately.
Dated, quantified facts
A result, a scope, a duration. That is what a human remembers, and also what a tool can extract.
No guarantee, and that is normal
A readable file can be rejected on the criteria of the position, and that is legitimate. Readability only prevents you from being rejected for the wrong reason. Beware of anyone who promises to “beat the filters”.
Using AI yourself, without getting caught out
Where it helps
Structuring a CV, rewording a heavy sentence, summarising an ad, preparing interview questions, practising answers. Everything that is form and preparation.
Where it hurts
Interchangeable cover letters, hollow phrases, invented skills, mistakes about the company. A recruiter reading thirty files recognises the generic tone by the third line.
The rule: your facts, its words
Give it your figures, your projects, your results, and let it work on the form. Then reread every line: a wrong date or an inflated title will follow you into the interview.
In a video interview, be yourself
Do not read a generated text. Recruiters see it, and analysis tools do too. Prepare your examples, not your sentences.
The Licorn method: matching replaces keyword screening
You declare, the algorithm compares
You state your skills and requirements, the recruiter states the position and its conditions. Matching compares criteria declared on both sides, not keywords guessed from a document. You only receive offers that meet what you set.
A standard CV, readable by everyone
Licorn generates your CV in a single, structured format that any recruiter and any software reads effortlessly. No more layout to guess.
A human always decides
On Licorn, no application is rejected by a machine. Recruiters contact you, based on your criteria, and you choose whom to answer.
Frequently asked questions about AI in recruitment
If the rejection is decided exclusively by automated processing, the FADP requires the company to inform you and gives you the right to state your point of view and demand a review by a natural person (art. 21 FADP). In practice, this is rare in Switzerland: 3.6% of companies use AI for pre-screening, and most use it as a reading order, not a decision.
Yes. You can exercise your right of access (art. 25 FADP) with the company and ask which data it processed and, in case of an automated decision, the logic on which it was based. The reply is free of charge and generally arrives within 30 days. In the European Union, the AI Act’s transparency obligation has required since 2 August 2026 that people be told when they interact with an AI.
They are not prohibited, but they are regulated: the employer may only process data needed to assess your suitability for the job (art. 328b CO), must inform you of the processing, and a decision taken by the tool alone falls under art. 21 FADP. In the EU, these systems will be subject to the high-risk obligations from 2 December 2027.
No law requires it. What matters is that the facts are true and that you can defend them in an interview. If a recruiter asks, answer simply: the AMOSA survey shows that companies want this transparency and do not penalise the use as such, but hollow files.
No. A clear CV with standard headings, real text, an unscanned PDF and a clean reading order is readable by every tool on the market. Services that promise to “beat the filters” sell fear. On Licorn, the CV is generated in a standard format, two structured columns and real text, with nothing to optimise.
Nothing suggests it in Switzerland: 13% of companies use AI in recruitment, mainly to write ads, and the decision remains human almost everywhere. What changes is the volume of generated applications, and therefore the premium on precise, sincere files.
In short
In Switzerland, your application is almost always read by a human, sometimes helped by software that ranks more than it decides. The law protects you against a decision taken by a machine alone, and European law will impose human oversight on recruitment tools. Make your file readable for both readers, use AI for the form and never for the facts, and remember that no layout replaces experience that matches the job.
Or change the logic: on Licorn, your declared criteria do the sorting, and it is a recruiter, not an algorithm, who writes to you.
Sources
- AMOSA, “KI in Jobsuche und Recruiting”, survey of some 6,000 jobseekers and 700 companies, April 2026 (in German)
- Netzwoche, “KI öffnet Stellensuchenden mehr Türen zu Bewerbungsgesprächen”, 16 April 2026
- Federal Act on Data Protection (FADP), art. 21 and 25, full text on Fedlex
- Federal Data Protection Commissioner, data processing by the employer (in French)
- ICTjournal, “Décisions individuelles automatisées : quel cadre légal ?”, S. Métille and D. Pressouyre, September 2024
- Regulation (EU) 2024/1689 on artificial intelligence, EUR-Lex
- Cooley, “Digital AI Omnibus delays key deadlines”, Regulation (EU) 2026/1744, August 2026
- Reuters, “Amazon scraps secret AI recruiting tool that showed bias against women”, October 2018
