You have probably read that a robot throws out 75% of resumes before a person sees them. It is a frightening number, and it makes a hard job search feel pointless. It also has no published evidence behind it. Software does shape who gets read, mainly through form questions, searches and rankings. Most of that is easier to work with than a mystery robot.
Where the 75% figure came from#
The trail leads to Preptel, a US company that sold resume optimization to job seekers. Its ResumeterPro service cost $24.95 a month in 2012, and its homepage that year advertised that 82% of users with its top rating got interviews and 61% got offers. A CIO.com article, excerpted on Preptel's own news page, said these 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."
Look at two things there. The source was a company selling the cure. And the wording is about chances of an interview. Only later did it harden into "rejected by software" and "never seen by a human."
Preptel told customers in August 2013 that it was closing after three years of losses. The number kept traveling. Fortune's advice column said in 2014 that it "probably came from a study" by Preptel. CNBC repeated it in 2019, crediting Preptel and describing no method.
In 2020, HR writer Christine Assaf followed the citations back through resume-service sites and a Forbes article and could not find the study. A 2026 article from The Conversation reached the same point: no methodology was ever published. The archived Preptel pages retrieved for this article show the claim as a press quote, with no study, sample size or method. That does not prove the number wrong. It means nobody can check it, so it should not be repeated as fact.
What the evidence does show#
Nobody has published a measured rate of how many resumes software rejects across employers. What exists is surveys of what employers say they do, and they disagree.
| Source | Who answered | What it found | Read it with care because |
|---|---|---|---|
| Harvard Business School and Accenture, 2021 | 2,275 executives in the US, UK and Germany | 63% used a recruiting system, and over 90% of those used it to first filter or rank candidates. Of those users, 88% believed qualified high-skills candidates were vetted out for not matching exact criteria, and 94% said so for middle-skills candidates | It is employer opinion, and "filter or rank" joins two different actions |
| Enhancv, 2025 | 25 US recruiters, interviewed | 23 reported no automatic rejection for format, content or design. 2 used content-based rejection | Small and not representative. The publisher sells resume tools |
| Resume Genius, 2026 | 1,000 US hiring managers | 71% screen with an ATS. 19% say AI screens out some candidates under human-set rules. 6% say AI can move or reject candidates with limited human review | The publisher sells resume tools |
| MyPerfectResume, 2026 | 1,000 US hiring managers, as reported | 65% say AI automatically rejects applicants before human review | The publisher sells resume tools. Same country and year as the row above, with more than three times the 19% figure |
Notice what these numbers count: employers using a practice, not resumes rejected. An employer that auto-rejects a few applicants and one that auto-rejects most of them look the same in these polls.
Two conclusions survive that noise. The claim that software kills three in four applications is unsupported. The comforting reply is also too strong. Assaf's article ends by saying these systems "do not reject" applicants, yet vendor documentation describes automatic rejection plainly, as you will see below. The Harvard report also says almost half of surveyed companies screened out resumes with a work gap of more than six months.
None of those surveys covers South Asia. One HireRight figure reported by People Matters put ATS use among companies in India at 54% in 2022, with no sample size given. No checkable figure turned up for Pakistan. Be wary of anyone quoting a precise number for either market.
What the software actually does#
Parsing#
Parsing means pulling your text into fields: name, phone, employers, titles, dates, education. It saves data entry, and it makes mistakes. Greenhouse, an ATS vendor with detailed public help pages, tells candidates so directly and asks them to check parsed details. When a parse fails there, the candidate is not rejected. The resume stays attached and staff enter or correct the details by hand. The practical risk is quieter than rejection. A record with scrambled or missing fields is harder to find and harder to read.
Search#
Recruiters search their database much as you search a shopping site. Greenhouse offers full text search across resumes, with a snippet showing how each term is used. In India, the keyword search in Naukri's Resdex checks fields such as profile, experience summary, current organization and designation. So update your portal profile as well as the file you upload. If a recruiter types "GST reconciliation" and you wrote "tax matching," you may never appear. That is the plain meaning of keywords.
Knockout questions#
This is where true auto-rejection lives. Employers add questions to the application form, and certain answers end the application. Greenhouse's auto-reject rules run on answers to custom yes/no or select questions, and people set up for new-application alerts are not notified of those applicants. Workable lets employers disqualify on a "No", and the candidate still gets a confirmation email with no hint of it. Oracle's Taleo documentation describes disqualification questions such as "Are you entitled to work in the United States?"
If you are applying abroad from India or Pakistan, read these questions before you spend an hour on the resume. One recruiter in the Enhancv interviews estimated that a sponsorship question removes about 30% of applicants for technical roles. That is one person's estimate, but no resume edit can change it. Answer honestly. A false answer can surface at interview or when documents are checked.
Ranking, then a human#
Some systems score how closely an application matches the job. Greenhouse says its Talent Matching score "does not automatically advance or reject candidates." In Enhancv's interviews, 44% relied on an AI or fit score: 36% as a guide and 8% as decisive. Volume is what makes this matter. Harvard's report cites an average of 250 applications per corporate job posting, so a low score can leave you at the bottom of a long pile. Timing may help too. In the same interviews, 52% said they review in arrival order and that applying early helps, while 36% said timing makes little difference. When a job fits well, apply in the first few days if you can.
Formatting problems that really break parsing#
Greenhouse publishes its own list of causes for failed parses. Other systems differ, but this list comes from a vendor rather than folklore, and every fix is cheap.
| What breaks | Do this instead |
|---|---|
| Resume uploaded as an image or scan | Export from Word or Google Docs so the text can be selected |
| Columned layout | One column, top to bottom |
| Name and contact details in a header, footer or text box | Put them in the first lines of the body |
| Complex tables, graphics, photos, word art | Plain text, simple bullets, bold for emphasis |
| Spaces between letters, like "S K I L L S" | Normal text |
| No clear sections, or a different format in each | Standard headings and one date format throughout |
| Incomplete job titles, or company names with no Inc., Ltd or similar | Full title and full company name |
| File over 2.5 MB | Remove images, since Greenhouse will not parse above this size even though it accepts uploads up to 100 MB |
CV formats common in South Asia often carry a photo and personal details such as date of birth. Since photos are on that list, leave yours off unless the employer asks for one.
On PDF versus Word, the advice has changed. A 2014 Fortune column relayed Preptel's advice to use Word. Greenhouse now tells candidates to "upload a PDF for best results." Follow the employer's instruction first. Otherwise, a text-based PDF is a sensible default, with a .docx ready in case an upload fails.
Two free tests will catch most problems. Select all the text in your PDF and paste it into a plain text editor. If sections appear out of order or text goes missing, a parser may struggle too. Then, when an application form auto-fills from your resume, read what it filled in. That is the closest you will get to seeing your own parse. If the same error keeps appearing, fix the resume.
How to tailor keywords without stuffing#
The Harvard report gives the reason to bother. It says workers who describe skills "using language that differs from that utilized in the job description" are dropped from consideration. So your task is translation, and it has one firm limit: every word must be true of you.
Start with the posting. Mark the hard requirements: tools, licenses, certifications, job titles and any method named twice. Compare them with your resume. Where you have done the work under another name, use the employer's term. If they say "accounts payable" and you wrote "vendor payments," change it. If you have never done it, leave it out, because you would have to defend it on a call.
Write both forms of important terms once, such as "Search Engine Optimization (SEO)" or "Chartered Accountant (CA)." Jobscan reports that search in the Lever ATS matches word stems but not abbreviations. That is a vendor's observation, not Lever's documentation, but writing both costs you nothing.
Then put each term where evidence surrounds it. Since a recruiter may see a snippet of the surrounding text, a bare list looks thin. Compare these invented lines:
Stuffed: Skills: Excel, advanced Excel, Excel reporting, MIS, MIS reports, data analysis, data analytics, dashboards.
Better: Built monthly MIS reports and sales dashboards in Excel (pivot tables, Power Query) for 14 branch managers, cutting preparation time from two days to three hours.
Keep a skills section too. In the Resume Genius survey, 85% of hiring managers said they expect one. Limit it to terms that also appear in your experience.
Skip the tricks. A 2025 New York Times report said Greenhouse found tricks in about 1% of resumes it reviewed, ManpowerGroup detects hidden text in roughly 10% of those it scans with AI, and some recruiters reject applicants who try it. One recruiter found a hidden prompt simply by changing the font to black.
Give each application about 20 minutes. Keep one master resume, then adjust the summary, reorder bullets and swap in the employer's terms. If that takes much longer, the job may not be a close match.
Run your current file through the ATS resume checker. If the layout is the problem, rebuild it in the resume builder, then repeat the two tests above.
