Do applicant tracking systems really reject your CV?

The number everybody repeats

You have read it a hundred times. Seventy five percent of CVs are never seen by a human being. The robot eats them.

It is worth knowing where that number came from, because the answer is short.

It comes from an article in Computerworld published on 4 March 2012. The article says applicant tracking systems "kill 75 percent of candidates' chances of landing an interview as soon as they submit their resumes, according to job search services provider Preptel."

Two things about that sentence.

Preptel sold a service that optimised your CV for applicant tracking systems. The figure describing the problem came from the company selling the cure. No study, no sample, no method was ever published, and the company shut down around 2013. Nothing of the research survives because there does not appear to have been any.

And read what it actually claims. It says the systems reduce your chances of landing an interview by 75 percent. That is not the same claim as "75 percent of CVs are never read by a human", which is what the internet now says. The claim got stronger every time it was retold, and the stronger version has no source at all, not even a vendor blog.

What the software companies say it does

The useful thing about applicant tracking systems is that the companies that build them publish documentation, and that documentation is specific.

Greenhouse has an auto-reject feature. It fires on "an applicant's answer to a question", meaning the yes/no and multiple-choice questions the employer wrote into the application form. Greenhouse also tells candidates directly that applications "are reviewed by real people, one by one, in the order received", that the filters "are set and controlled directly by recruiters (AKA, living, breathing humans) not outsourced to a bot", and that its AI "doesn't score or rank applications, nor does it make any decisions about whether or not you move forward."

Lever has auto-screening, which archives candidates "based on how they respond to custom questions".

Oracle Taleo has disqualification questions: "A candidate not meeting the required response can be instantly exited from the application process." Everything else it does with your application, including its scoring, ranks candidates for a recruiter rather than removing them.

Oracle Recruiting Cloud is the same shape: a single wrong answer to a disqualification question disqualifies you, and the other scores are used to rank.

Workday documents a Screen step that routes a task to a named human who has to choose Move Forward or Decline.

Notice what is missing from all of it. Not one of these companies documents rejecting anybody because their CV did not contain the right words. The auto-rejection that exists is real, and it fires on the form, not the document.

The Harvard study, read properly

The other thing people cite is a Harvard report. It exists, and it does not say what it is quoted as saying.

Hidden Workers: Untapped Talent, by Joseph Fuller and Manjari Raman at Harvard Business School with Accenture, September 2021. It is a white paper, not peer reviewed.

Its method is the part that matters: it surveyed 2,275 executives and asked them what they thought. It did not audit, instrument, test or measure a single applicant tracking system.

The famous figure is that 88 percent of employers agreed that qualified candidates get "vetted out of the process because they did not match the exact criteria". That is 88 percent of executives holding an opinion about their own hiring. It is quoted as though it were 88 percent of CVs being rejected, which is a completely different statement, and one the report does not contain anywhere.

The report is still interesting. A survey of what hiring managers believe about their own process is worth having. But it cannot tell you what the software did, and it cannot separate the software from the people configuring it, because it never looked at either.

Where applications genuinely do die automatically

There is real automatic rejection, and knowing exactly where it lives is worth more than any amount of keyword advice.

It lives in the questions on the application form. Work authorisation. Location and willingness to relocate. A required licence or certification. Years of experience, chosen from a dropdown. Willingness to meet a stated condition. Those are the fields wired to the auto-reject rules, and when one of them rejects you, it is silent: Greenhouse's documentation notes that people set up to receive notifications about new applicants "will not be notified of auto-rejected applicants". The employer never sees it either.

So the five minutes you spend agonising over whether to write "managed" or "led" in a bullet point is five minutes spent on the part of the application that has no documented automatic consequence. The dropdown you clicked through in two seconds is the part that does.

The advice nobody can evidence

While we are here: the standard formatting rules. No columns, no tables, no text in headers, save as .docx not .pdf, because the parser will choke.

I went looking for the evidence and could not find any. No study. No vendor statement. Greenhouse's official page on supported formats accepts .doc, .docx, .pdf, .rtf and .txt, and offers exactly one caveat, which is that externally linked images do not load in its preview. It says nothing about columns or tables.

That does not prove the advice is wrong. Parsers do fail, and a simpler document is a safer document. But it is currently folk wisdom, not a finding, and it is repeated with a confidence nobody has earned.

Be suspicious of the correction too

There is now a counter-claim going around: that 92 percent of recruiters confirm applicant tracking systems do not auto-reject. It comes from research run by a company that sells CV tools, based on interviews with 25 recruiters.

That is sourced about as thinly as the myth it corrects. Replacing one vendor statistic with another vendor statistic is not the same as knowing something.

What this leaves you with

• Answer the form questions with as much care as you write the CV. That is where documented automatic rejection actually happens, and it is the part most people click through.
• Stop optimising for a keyword filter that no vendor documents. Write clearly for the recruiter who will read it, because the documentation says a recruiter is who reads it.
• Assume the real filter is a human being under time pressure, searching and skimming a long list. Greenhouse admits that on high-volume roles, qualified applications get "skipped over". That is a very different problem from being eaten by a robot, and it has a different answer: being obviously relevant in the first few lines.
• Treat every confident number in CV advice as unsourced until you have seen the source. Including the ones in this article, all of which are linked.

The question of how much of your CV even gets read is its own can of worms, and the numbers there are no better. That is covered here.

This is one chapter of The CV That Gets Read.