Do applicant tracking systems automatically reject or filter out resumes before a human sees them?

An applicant tracking system is a database and a workflow tool. It stores the application, attaches your resume to a candidate record, pulls what text it can into structured fields, and gives recruiters somewhere to review and advance candidates [1][5]. Rejection is a status inside that workflow, and in most configurations a person sets it.

Greenhouse tells candidates that applications go into a queue on a recruiter's dashboard and are usually reviewed by real people in the order received, and that its AI "doesn't score or rank applications, nor does it make any decisions about whether or not you move forward" [1]. Workday takes a different position on scoring and sells HiredScore AI for Recruiting, which grades candidates against the role and surfaces the strongest matches to the recruiter [6]. Both statements are accurate about their own products, which is why any sentence beginning "the ATS does" is too broad to be useful.

Four different things get called rejection

Most of the confusion here comes from four mechanisms sharing one word. Parsing extracts text from your file into database fields. Screening is a rule acting on an answer you typed into the form. Ranking is a sort order deciding whose application sits near the top of a recruiter's list. Rejection is the status change that ends your candidacy. A failed parse leaves your record thin without touching your status, and a low rank can leave you unread while your application stays open, so neither is a rejection even though both feel like one. Automatic rejection is real but narrow and employer-controlled, and the more common outcome is that nobody opened your application [7].

Parsing copies text out of your resume file into structured database fields, and when it fails the application still exists with empty or wrong fields. Greenhouse documents the causes: files larger than 2.5MB, resumes uploaded as an image rather than a document, spaces between letters, graphics, word art, tables, headers and footers, columned layouts, and contact details inside a header, footer, or text box [5]. When the parse fails, the resume "has only been attached to the candidate" and someone has to type your details in by hand [5].

None of that rejects you. The record simply carries less searchable text, which matters because a recruiter working through several hundred applications searches and filters rather than reading every file [7][8]. If your skills never made it into the indexed text, you are harder to surface in that search, and that is a quieter failure than being rejected.

The useful action is narrow: submit a single-column document with your contact details in the body instead of the header, keep the file well under 2.5MB, and skip scanned images [5]. A sibling answer covers resume formatting in more depth.

Knockout questions are the real automatic rejection, and they act on the answers you type into the form rather than on anything in your resume. In Greenhouse an administrator picks a yes/no, single-select, or multi-select question on the job post, chooses which response should disqualify, and turns on auto-reject; candidates who give that response "will now be auto-rejected," with an optional rejection reason and email attached [2]. The feature sits on the Plus and Pro tiers and is set up one question at a time, so it applies only where an employer built the rule [2].

Workable works the same way with a narrower trigger. Knock-out questions there are supported only for yes/no questions and fire on a "No" answer, and a disqualified applicant still gets the usual confirmation email while being filed under a Disqualified tab where the recruiter can see which question caused it [3]. Workable's documentation states that "the candidate does not know that they have been disqualified" [3], so an application can be closed while looking, from your side, exactly like one still under review.

LinkedIn applies the same pattern, where a screening question can be marked as a must-have qualification and the poster can choose to automatically archive candidates who fail it and send an automatic rejection email [4].

The questions that usually carry these rules are hard requirements: work authorization and visa sponsorship, a professional license, a minimum number of years in a named skill, willingness to work onsite, and ability to pass a background check. Answer them accurately even when the answer costs you the role, because an inflated answer tends to surface later during verification, and a genuine "no" on a legal requirement was never something a better resume would have fixed.

Ranking and filtering change the order in which a recruiter meets your application, and one near the bottom of a long queue often gets no review while its status stays open. Harvard Business School's Project on Managing the Future of Work, with Accenture, surveyed 2,275 executives across the United States, United Kingdom, and Germany and found more than 90 percent used their recruiting system to initially filter or rank candidates, 94 percent for middle-skills roles and 92 percent for high-skills roles [8].

That research also shows where the filters cut hardest. About 48 percent of employers filtered middle-skills candidates on employment gaps longer than six months, so an applicant who stepped away for illness or caregiving could be screened out on that single variable, and in the report's words "a recruiter will never see that candidate's application" [8]. The same survey found 88 percent of employers agreed qualified high-skills candidates were vetted out for not matching the exact criteria in the job description, rising to 94 percent for middle-skills workers [8]. That survey ran in early 2020 and captures employer belief and configuration rather than a measured rejection rate, so read it as evidence that aggressive filtering is normal, not as a percentage of resumes discarded.

AI grading belongs in this category rather than in the rejection category. Workday's HiredScore AI for Recruiting grades candidates against the role and places top matches in a spotlight inbox [6]. A low grade is not a rejection, but on a requisition with several hundred applicants the effect is similar, because attention runs out before the list does.

Application volume grew much faster than recruiting capacity, and that gap explains most of the silence. Greenhouse analyzed more than 640 million applications across over 6,000 companies from 2022 through 2025 and found the average job opening went from 116 applications in 2022 to 244 in 2025, an increase of 111 percent [7]. Over the same stretch the average recruiter's annual load climbed from 146 applications to 746, up 412 percent, while the average recruiting team shrank from 10.43 people to 4.62 [7].

A recruiter carrying 746 applications a year with less than half their 2022 team is not opening all of them. Greenhouse says as much to candidates, noting that when a role is popular, recruiters may filter the queue rather than work through it [1]. That is a human decision made under a workload constraint, and it produces the same experience as an automated rejection while being a different thing entirely.

The distinction changes what you do next. A software rejection would call for a technical fix, something about your file or your phrasing. An application a recruiter never reached calls for timing and access instead, meaning applying early in a posting's life while the queue is short and finding someone inside the company who can put your record in front of the hiring manager.

The claim that 75 percent of resumes are rejected before a human sees them

No published study supports this figure. It traces to a 2012 sales pitch from a company called Preptel, which sold resume optimization services and dissolved the following year without ever publishing a methodology, a sample size, or a dataset [9]. Searches of the academic literature for an ATS rejection rate turn up nothing supporting it [9]. The number survives because it now cites itself, and because the businesses repeating it sell resume services.

The claim that hiding keywords in white text gets you through

Hidden text is one of the easier things for a recruiter to catch, because parsing strips the formatting and drops the extracted words into the candidate record. Jobscan's walkthrough shows repeated hidden instances of a keyword appearing plainly once the resume has been parsed into the system, and calls the tactic "one of the quickest ways to burn a bridge with a recruiter" [10]. Parsing is exactly what makes the invisible words visible.

The claim that any design or unusual formatting kills your resume

Formatting problems degrade parsing rather than trigger rejection, and the list of things that actually break a parse is short and published [5]. A restrained color header or a nonstandard font is not on it, and a resume that fails to parse still reaches the employer as an attached file [5].

What an applicant tracking system is not

It is not, in most deployments, an algorithm that reads your resume and judges you [1]. It is not a shared blacklist, since each employer runs its own instance with its own data. It is not the same thing as an AI interview or an online assessment, which are separate products attached later, and it is not the reason most applications go unanswered [7].

Automated hiring tools are regulated in a growing number of jurisdictions, though the rules govern an employer's disclosure and testing obligations rather than giving you a right to a human reviewer. This is general information and not legal advice.

New York City's Local Law 144 covers automated employment decision tools, which the city describes as a computer-based tool used to screen job candidates or assess employees [11]. An employer using one on New York City candidates has to make sure a bias audit was done first, post a summary of the results on its website, notify candidates that the tool will be used to assess them, and give instructions for requesting a reasonable accommodation [11]. If you applied to a New York City role, that published audit summary is something you can go read.

Illinois amended its Human Rights Act effective January 1, 2026. Section 2-102(L) makes it a violation for an employer "to use artificial intelligence that has the effect of subjecting employees to discrimination on the basis of protected classes under this Article or to use zip codes as a proxy for protected classes," and separately to fail to notify an employee that artificial intelligence is being used for those purposes [12]. Illinois also has an older, narrower statute, the Artificial Intelligence Video Interview Act, requiring notice, an explanation of how the artificial intelligence works, and written consent before AI analyzes a recorded video interview [13].

Colorado's position is unsettled. The Colorado AI Act, SB 24-205, would have covered artificial intelligence used in consequential decisions including employment, but on April 27, 2026 a federal court granted a joint motion temporarily suspending enforcement after xAI sued the state and the U.S. Department of Justice moved to intervene [14]. Verify its status before relying on it, since amendment and replacement were both live possibilities [14].