The honest answer to the ATS question, with sources. What applicant tracking systems really do, what they do not do, and the three things worth checking.
You have probably read that 75% of resumes are rejected by a robot before a human sees them. It is the most repeated statistic in this industry, it is on the pricing page of half the tools that will sell you a fix, and there is no study behind it.
The number traces to a 2012 sales pitch by a company called Preptel, which sold resume-optimisation software. Preptel published no study, no dataset and no method. It went out of business in August 2013. Every article citing the figure points back through an uncited mention to the same defunct vendor.
So here is the useful version of the question. Not "will a robot reject me", but "what does the software actually do with my file, and which parts of that can I control".
An applicant tracking system is a database with a workflow attached. It stores applications, moves them between stages, and lets recruiters search. Nearly every large employer uses one: a 2026 survey detected an ATS on 487 of 500 Fortune 500 career pages, with Workday alone above 40%.
Three of its behaviours matter to you.
It parses your file into fields. Name, employer, title, dates, skills. This is the step where a resume goes wrong, and it is the only step you fully control.
It can auto-reject, but not on your resume. Greenhouse's own documentation is specific: an auto-reject rule fires "based on an applicant's answer to a question", and works only with Yes/No, single-select and multi-select question types. Ashby documents the same design. Neither evaluates resume content, keywords or formatting. When you get rejected in four minutes, it is almost always because you answered no to "are you authorised to work in this country" or similar.
It shows recruiters a score. Workday's HiredScore presents parsed resumes alongside calculated signals like employment gaps and time in position, and recruiters act on candidates from that view. The score is advisory. A human still clicks.
There is a real, well-sourced version of the fear, and it is more useful than the myth.
Harvard Business School and Accenture surveyed more than 8,000 workers and more than 2,250 executives for Hidden Workers: Untapped Talent in 2021. They found that 88% of employers agree qualified, high-skilled candidates are vetted out because they do not match the criteria written into the job description.
Read that again, because the difference matters. The filter is not a robot misreading your layout. It is a human-configured rule about stated facts: a degree you do not have, a gap you did not explain, a job title that does not match. The software applies the criteria. A person wrote them.
That is a writing problem, not a formatting problem, and it is fixable.
This is the one nobody shows you. Every tool in this category will sell you a score; almost none will show you the extracted fields.
Upload your resume and look at what comes out. Is your most recent employer in the employer field? Are your dates intact? Did your job title survive? If a two-column layout collapsed your work history into a single run of text, you will see it immediately, and no score out of 100 tells you that.
Layout genuinely affects this, and the effect has been measured. A 2025 study found roughly 20% of resumes use non-linear, multi-column layouts that break the top-to-bottom, left-to-right reading order parsers expect. Adding layout-aware preprocessing raised extraction accuracy from 0.919 to 0.959 F1, which removes about half the remaining error. So columns do not break an ATS. They just make it more likely that a date lands in the wrong field.
Not keyword stuffing. Coverage.
If the posting says "incident response" and your resume says "handled outages", a keyword search will not find you and a recruiter skimming for the phrase will not either. Both readers are looking for the same words.
Go through the posting, list the requirements it states, and check which ones your resume answers, which it answers weakly, and which it never mentions. Then decide what to change. That is the whole exercise, and you can do it by hand.
What you should not do is trust a percentage. Every "a 75% match gets you 4x more callbacks" claim found in the research traces to a vendor blog with no method, no sample and no publication. Nobody outside a specific company's system can calculate your match to it.
Send a PDF or a Word file. Both work. Greenhouse accepts .doc, .docx, .pdf, .rtf and .txt; Workday's HiredScore documentation lists the same set. The advice to always send Word is out of date, and has been for years.
What still causes trouble: text stored as an image, because a scan of a printed resume has no text to extract at all. Fonts embedded oddly. Text inside a header or footer region, which some parsers drop.
One more, and this one is new. Do not put hidden white text in your resume to game an AI screener. Researchers analysing 200,000 real resumes submitted between 2019 and 2025 found about 1% contained exactly this kind of prompt injection, and the rate rose sevenfold between mid-2024 and late 2025. It is now something screening vendors actively look for, and one of the larger resume tools took public criticism for shipping a feature that did it. Getting caught is worse than any formatting problem on this page.
Stop optimising for a robot that mostly is not doing what you were told it does.
Make sure the parser can read your file, which takes one upload to check. Make sure your resume answers the criteria the posting actually states, which is writing, not formatting. Then spend the rest of your time on the words, because the person reading them is who decides.
Rzume shows you the parse first, before it says anything about your writing. It takes about twenty seconds, there is no account, and there is no score out of a hundred, because we cannot honestly calculate one and neither can anyone else.
Upload your resume and see the parse, or read what each plan includes first.
All accessed 29 August 2026.