The 75 percent ATS auto-rejection figure traces to 2012 marketing from Preptel, a vendor that closed in 2013; real auto-rejects fire on application question answers, not resumes.
The short version: An applicant tracking system is, first and last, a database with a workflow attached to it. It collects applications, tries to convert your resume into structured fields, stores the result, and lets recruiters search, tag and move people between stages. The widely repeated claim that roughly three quarters of resumes are discarded by software before any human sees them traces back to marketing material from a company called Preptel in 2012, which closed the following year and never published a method. Automatic rejection does exist, but in the systems that offer it, it is triggered by your answers to the application questions, not by anything in your resume. What the software genuinely struggles with is reading unusual page layouts. The honest advice that follows is narrow and dull: one column, standard headings, a text-based file, and then put the rest of your effort into the content and into who actually sees it.
What an applicant tracking system actually does
Strip away the marketing and one of these systems does four things. It publishes job posts and collects applications. It parses uploaded resumes into structured fields so that a name, an email address and a work history land in the right database columns. It gives recruiters a way to search, filter, tag and comment on the pile. And it keeps a record of every decision, which employers in the United States need for equal employment opportunity reporting and for defending a hiring decision if it is ever challenged.
That last function explains more about these systems than any other. An applicant tracking system is partly a compliance record. It exists to show that a process happened and to preserve it. Software built to create an audit trail tends to be conservative about throwing things away, which is one reason the image of a machine silently binning your application does not fit the design goal very well.
Parsing is the part job seekers worry about, and it is worth reading what the vendors say rather than what the resume-optimisation industry says about the vendors. Workday’s administrator documentation, on its concept page for resume parsing, states that resume parsing results can vary based on resume format and order of words, and advises that for best results you should use resumes that do not contain images or image-based styles. The same page notes that certain fields, including languages and skills, are not populated automatically at all. Greenhouse’s support documentation lists the file types it accepts for resumes and cover letters as .doc, .docx, .pdf, .rtf and .txt, with a size ceiling of 100 MB.
Notice what is absent from both documents: any mention of a score, a threshold, or a rejection. Parsing is an extraction step. When it goes wrong, the usual result is a garbled or half-empty candidate record sitting in a queue, not a deletion.
The automatic rejection that does exist
There is a real auto-reject feature in these systems, and it is worth understanding precisely because it is so often confused with resume screening. Greenhouse’s published support documentation describes three application rules: auto-tag, auto-advance and auto-reject. All three are triggered by a candidate’s answers to job post questions, and only to questions of three types, being yes or no, single-select and multi-select. Up to five questions can be combined with AND logic. Where rules conflict, the documentation states that auto-reject takes precedence over auto-advance.
In plain terms: if you answer “no” to “are you legally authorised to work in this country” or “do you hold an active registered nurse licence in this state”, a rule can close your application in seconds. That is the machine rejection people describe. It reads a checkbox, not your career. This also explains the experience of being rejected four minutes after applying, which many people attribute to an artificial intelligence reading their resume and finding it wanting.
The seventy-five percent claim, and why it will not die
The single most repeated statistic in this field is that seventy-five percent of resumes are rejected by applicant tracking software before a human sees them. The number has been traced by several career writers to promotional material published in 2012 by Preptel, a resume-optimisation vendor that ceased operating in 2013. No study, sample, or method was ever published alongside it. It is a sales figure that acquired the grammar of a research finding by repetition.
This matters because a real finding exists nearby and gets conflated with it. The Harvard Business School and Accenture report Hidden Workers: Untapped Talent, by Joseph Fuller, Manjari Raman, Eva Sage-Gavin and Kristen Hines, published in September 2021, surveyed employers and found that more than ninety percent said they use their recruiting management system to filter or rank candidates. Eighty-eight percent agreed that qualified high-skills candidates are vetted out because they do not match the exact criteria in the job description, rising to ninety-four percent for middle-skills roles. The authors estimated twenty-seven million hidden workers in the United States alone.
Read that carefully, because it says something different from the Preptel number. It is employers reporting their own opinion about their own filters. The filters are criteria a human chose: a degree requirement, a minimum years-of-experience threshold, a specific job title, an unbroken employment history. The software applies them faithfully. The exclusion is a policy decision that has been automated, not a parsing accident.
Regulators treat it the same way. The Equal Employment Opportunity Commission’s technical assistance document of 18 May 2023, Select Issues: Assessing Adverse Impact in Software, Algorithms, and Artificial Intelligence Used in Employment Selection Procedures Under Title VII of the Civil Rights Act of 1964, states that employers may be held responsible for tools built by third-party vendors, and that the four-fifths rule is only a rule of thumb rather than a safe harbour. The legal exposure sits with the employer who set the criteria.
What parsing actually fails on
Parsers read a document’s underlying structure, not its visual appearance. A human eye reconstructs two columns effortlessly. A parser reading in document order may interleave them into nonsense. Everything in the table below follows from that single fact.
| Formatting choice | What the parser tends to do with it | Practical risk |
|---|---|---|
| Two or three column layout | Reads content in document order, which may interleave the columns line by line | High. Dates detach from employers, skills land inside job descriptions |
| Tables used for layout | Varies by parser; some flatten cells in the wrong order, some read them correctly | Moderate to high, and unpredictable across systems |
| Contact details in the page header or footer | Some extractors ignore header and footer regions entirely | High. A recruiter may have your resume and no way to contact you |
| Text inside an image, logo or icon | Not extracted at all unless optical character recognition is applied | High for scanned or exported-as-image files, which yield an empty record |
| Non-standard section headings such as “My Journey” | Section not recognised, so its content is not mapped to a field | Moderate. Content survives as raw text but is not searchable by field |
| Text boxes and graphic shapes | Often read out of sequence or dropped | Moderate to high |
| Unusual date formats such as “Spring 2019 to date” | Date fields fail to populate; tenure cannot be calculated | Moderate. Filters on years of experience may silently exclude you |
| Single column, plain headings, standard dates | Extracted cleanly by every mainstream parser | Low |
The decision rule: open your resume, select all, copy, and paste it into a plain text editor. If what comes out reads in the right order, keeps your contact details, and keeps each date beside its employer, the file will parse. If it does not, fix the layout rather than the wording. This test takes thirty seconds and replaces the entire category of paid resume scanning tools.
File format
The Word-versus-PDF argument is mostly settled and mostly overblown. A PDF generated by exporting from a word processor contains a real text layer and parses fine; Greenhouse and Workday both accept PDFs. A PDF produced by scanning a printed page, or by exporting a design file as a flat image, contains no text at all, and is the single worst thing you can upload. If you cannot select and copy text from your own PDF, neither can the parser.
If an application portal offers a choice and you have no other information, .docx is the marginally safer default, because it is the format parsers have been tuned against longest. If a posting specifies a format, follow the posting. This is a low-stakes decision that the advice industry has inflated.
Keywords, and how recruiters actually search
Most applicant tracking systems do not compute a match score at all. The ones that do, produce a number the recruiter can sort by and ignore. What recruiters overwhelmingly do is run searches, typically Boolean keyword searches across the parsed text and the stored fields, and then read what comes back.
The practical consequence is not that you should stuff keywords. It is that the words a recruiter would plausibly type must appear somewhere in your resume in a form the parser captured. If the posting says “accounts payable” and your resume says only “AP”, you will not appear in that search. Write both the spelled-out term and the abbreviation the first time each appears, and let repetition take care of itself.
White text keyword stuffing, the trick of pasting the job description in invisible font, is a bad idea for a plain reason: parsers extract it, and the recruiter sees it when the text is displayed. It reads as an attempt to deceive.
What matters far more than any of this
Formatting is a hygiene factor. Getting it wrong can cost you; getting it right wins you nothing. Three things dominate it.
The first is whether you actually meet the stated requirements, because those requirements are what the filters encode. The second is specificity. A bullet reading “responsible for reporting” and a bullet reading “rebuilt the monthly close reporting pack for a fourteen-person finance team, cutting close from nine days to five” parse identically; only one gives a recruiter a reason to call. The third is route of entry. A referral enters through a different door and is read with a different disposition, and no amount of formatting compensates for its absence.
For scale, the Bureau of Labor Statistics reported in its Job Openings and Labor Turnover Survey release of 1 September 2026 that there were 7.3 million job openings and 5.1 million hires in July 2026. Hiring at that volume is not done by software reading prose. It is done by people making fast decisions about a queue, and your resume’s job is to survive the extraction and then give one of them something concrete.
Where these figures come from
Parsing behaviour and file support are taken from Workday’s administrator documentation concept page on resume parsing and from Greenhouse’s support documentation on supported file types and on application rules and auto-reject, all current at the time of writing. The origin of the seventy-five percent claim in 2012 Preptel promotional material, and that company’s closure in 2013, has been documented by several career-research writers; no primary study supporting the figure has ever been produced. Employer survey figures are from Hidden Workers: Untapped Talent by Joseph Fuller, Manjari Raman, Eva Sage-Gavin and Kristen Hines, Harvard Business School and Accenture, September 2021. The regulatory position is from the Equal Employment Opportunity Commission technical assistance document of 18 May 2023 on assessing adverse impact in software, algorithms and artificial intelligence under Title VII. Labour market volumes are from the Bureau of Labor Statistics Job Openings and Labor Turnover Survey, released 1 September 2026. The widely quoted figure that recruiters spend about 7.4 seconds on a resume comes from a 2018 eye-tracking study published by Ladders Inc., a commercial job board; its sample and method were not published in full, and it is repeated here only as a commonly cited claim rather than as an established finding.
General information, current as of September 2026, and not personalised advice.
Frequently asked questions
Do ATS systems automatically reject resumes?
Mostly no. In typical systems, software ranks and filters while humans decide. A poorly parsed resume is not rejected; it is simply never found in searches, which has the same effect.
Should I use a resume template from a design site?
Only if it is a simple single-column text layout. Heavily designed templates are the most common cause of parsing failures.
How long should my resume be?
One page early in a career, up to two pages with substantial relevant experience. Length matters less to the software than structure; it matters to the human reading afterward.
