Job search in the AI era

What Really Happens to Your CV in an ATS in Israel

Published: 2026-07-09

Nir Kosover · Founder of JobFox. Building the job-search engine for Israeli tech.

There is one number that dominates every conversation about CVs: “75% of CVs are rejected by a robot before a human even sees them.” It shows up in every LinkedIn post and every job-search TikTok, and it makes people design their CV out of fear of an imaginary machine. The problem: that number simply is not true. Let us walk through what actually happens to your file the moment you hit “submit,” and what is worth doing with the energy you get back once you stop being afraid.

Does a robot really reject your CV?

Short answer: in most cases, no. The “75% auto-rejected” figure traces back to a software company called Preptel that pitched a version of it around 2012 and went out of business in 2013, without ever publishing a measurement method. It has survived on shares, not on research.

What the research does show is very different. In Enhancv’s 2025 survey of 25 US recruiters, only 8% (two of the 25) enabled content-based auto-rejection of CVs. The rest read manually. What actually filters you is two thoroughly human things: knockout questions the employer set (work authorization, minimum years of experience, location), and a recruiter drowning in applications who has no time to read every one in depth.

And here is the subtle point. The Hidden Workers study from Harvard Business School and Accenture, 2021 found that 88% of employers admit their screening systems filter out qualified candidates. But the cause is not a malicious AI, it is rigid criteria a person defined: a hard degree requirement, an employment-gap cutoff, a demand for one specific certification. In other words, what knocks you out is usually a human, acting through the system, not the system acting instead of a human.

How an ATS actually works: 4 steps

An ATS (applicant tracking system) is not a gatekeeper with a “reject” button. It is a four-step pipeline, and the machine makes a real decision in only one of those steps:

1. Parsing, the real gate. The system extracts the text from your file, recognizes entities (name, roles, dates, skills), and arranges them into structured fields. If the text does not extract cleanly, nothing downstream can evaluate you. This is where broken formatting truly hurts you, not because someone rejected you, but because you became unreadable.
2. Search and filter. The recruiter searches and filters the parsed profiles, not your file. This is where keywords come in (do the words they searched for appear) and knockout questions. This step filters the most people, and it is defined by the employer.
3. AI ranking. Systems in 2026 increasingly add a semantic ranking layer that summarizes and ranks the survivors. It understands context and synonyms, which is why natural, specific language beats keyword-cramming.
4. Human review. A real recruiter goes through the shortlist. This is where the decision is made, and this is where the critical difference between system types comes in.

That difference is worth explaining, because it determines whether your design ever reaches human eyes at all. There are two families. Greenhouse is “file-forward”: the recruiter sees a split view with your original PDF next to the parsed fields and a summary. Your beautiful file genuinely reaches the eyes. Workday, by contrast, is “parse-forward”: the recruiter mostly sees Workday’s reconstruction in its own layout, and the original file is a download link that often never gets opened. Lever sits closer to Greenhouse. In Israel you will meet all of them, but pay special attention to Comeet, an Israeli system that is very common in local startups.

What really filters candidates: the structural problems

Now that it is clear the machine is not maliciously rejecting you, the real problem remains: broken parsing. If the system cannot extract your text cleanly, you become unfindable in a search, even if you are perfect for the role. These are the structural problems that actually knock people out:

1. Multi-column layout. This is the number-one killer. Parsers read left to right, top to bottom, and blend the two columns into gibberish.
2. Tables, text boxes, icons, charts, and logos. Any floating object can lose the content inside it.
3. Contact details in a header or footer. Many parsers skip the header and footer region, so your email and phone disappear.
4. Non-standard section headings. “My Journey” instead of “Work Experience” confuses the recognizer. Use the familiar headings.
5. Unclear date formats. The system reads dates to compute tenure. “Summer 2023” or “‘23” break it. Write “Month Year,” and “Present” for a current role.
6. PDFs from design tools. A file exported from Canva or Figma with image layers is nearly unreadable, because the text is effectively a picture.

And here it is worth breaking a second myth: an ATS-friendly CV does not have to be ugly. The constraint is structural, not aesthetic. No columns, no tables, no graphics. But a good font, size, clear hierarchy, generous whitespace, dividers, and restrained color are all completely safe. A single-column CV can look great.

How to check that your CV reads correctly

The best way to know whether your CV reads correctly is to see what the machine sees. Here are three practical checks:

1. The copy test. Open the PDF, select all, copy, and paste into a blank document. If the order scrambles, words stick together, or things go missing, that is exactly what the parser receives.
2. The parse preview. Some systems, Greenhouse chief among them, show the candidate a parse preview before submission. If an employer offers it, do not skip it, that is your chance to fix a field that was read incorrectly.
3. Two versions, by channel. Keep a clean structural version for portals and ATS systems, and a more impressive version for a warm intro, a direct email, and an interview, where a human sees the file itself. In Israeli tech, where warm introductions close a large share of the good roles, the impressive version pays off exactly where it matters most.

The bottom line: stop designing out of fear of a machine that, for the most part, is not making decisions. Make sure your text extracts cleanly, mirror the wording of the posting, and keep two versions for the two channels. If you want to see exactly how your CV parses, JobFox’s CV tools show you the fields an ATS will extract from your file, before you submit.