How AI Screens Your Resume Before a Human Sees It — and How to Write for It
Did a robot reject your resume before any human saw it?
Probably not — but three different machines read it first, and each one can quietly bury it. Before a recruiter opens your application, software parses it into data, scores it against the job description, and writes a short summary the recruiter reads first. This post maps all three steps with the research behind them, then gives you a checklist for writing a resume each machine reads correctly.
The number everyone quotes is wrong
The claim that "75% of resumes are rejected by an ATS" has no study behind it — it traces to a 2012 sales pitch. The figure comes from marketing material published by Preptel, a resume-optimization company that went out of business in 2013 without ever publishing a methodology, a sample size, or a survey. Researcher Christine Assaf traced the origin in a 2020 HRTact article, and no published study since has measured an ATS rejection rate — because that data sits inside employers, not in public research.
The one genuine 75% in this story is different: in Harvard Business School and Accenture's 2021 Hidden Workers report, 75% of U.S. employers said they use recruiting-management software. That is a share of employers using software, not a share of resumes it rejects. The distinction matters, because the fix for a machine that ranks you is different from the fix for a machine that supposedly deletes you.
Why the machine exists: 244 applications per job
In 2025, the average job posting drew 244 applications — more than double the 116 of 2022 — while recruiting teams shrank by more than half. Greenhouse's Hire Standard benchmark (March 2026), built from more than 6,000 companies and 640 million applications, puts the rise at 111% in three years. Over the same period, recruiters per organization fell from 10.43 to 4.62, and annual applications per recruiter jumped from 146 to 746 — a 412% increase. Time to fill rose 37% anyway.
Read those numbers together and the machine stops looking like a choice. Nobody can read 244 resumes per opening by hand, twice as many as three years ago, with half the team. Software does the first pass because the volume leaves no alternative. Recruiting is one of the tasks AI is changing first — our look at which tasks change first, according to researchers puts hiring-screening work near the top of that list. (One caveat: Greenhouse's numbers describe its own customers — mostly technology and growth companies — not the whole labor market. The direction, though, shows up in every other vendor benchmark too.)
How widespread is the screening layer itself? SHRM's 2025 Talent Trends survey of 2,040 HR professionals found 43% of organizations now use AI for HR tasks, up from 26% in 2024. A Resume.org survey of 1,399 U.S. workers in August 2025 found 57% of companies use AI somewhere in hiring, and 79% of those use it to review resumes. The software is no longer a big-company exception.
The three filters your resume actually passes through
"The ATS" is not one filter — it is three different steps, and each one fails in a different way. Most advice treats them as a single black box. Separating them is the whole trick, because what saves you at one step does nothing at another.
The first filter is the only one that truly auto-rejects, and it is the most honest. A knockout question is a yes/no gate the employer wrote: "Are you authorized to work in the U.S.?" or "Do you hold a valid CPA license?" Answer it accurately. No formatting trick passes a gate you do not meet, and no formatting mistake fails one you do.
The second filter never sees your resume the way you designed it. It sees plain text pulled out of your file, sorted into fields. Multi-column layouts get read top-to-bottom, left-to-right, scrambling the order. Text inside headers and footers is frequently dropped — Adobe's PDF Extract API, for example, leaves headers and footers out of its output. Text inside an image is not text at all. If the parser misplaces your dates or your job titles, the scorer downstream works with broken data, and your qualified candidacy becomes invisible.
The third filter is the 2026 change. Every major hiring platform now ships an AI layer on top of the old parser: Workday added Recruiter Agent, Greenhouse rolled out AI sourcing tools, iCIMS has Copilot, and Ashby is AI-native. The layer reads the parsed output, writes a structured summary — years of experience, current title, skills, education, location — and a fit score. Recruiters see that summary card before they see your resume, and most read the summary first. An ATS-friendly resume in 2026 is therefore not just a parsable resume; it is a resume whose summary comes out right.
A small 2025 survey of 25 U.S. recruiters by Enhancv fits that picture: 23 said their systems do not auto-reject resumes, 2 had set up content-based auto-rejection, and all 25 use knockout questions when a posting has them. Twenty-five people is not a representative sample — but it matches how the vendors describe their own products: the systems rank; the recruiters cut.
What the research says actually filters people out
The best evidence says the filter is real — but it is the employer's criteria, not the software's judgment, that does the filtering. Harvard Business School and Accenture's Hidden Workers: Untapped Talent (September 2021) surveyed 8,720 hidden workers and 2,275 executives in the United States, the United Kingdom, and Germany. Its findings: 88% of employers said qualified high-skills candidates are screened out because they do not match the exact criteria in the job description — 94% for middle-skills candidates. More than 90% of employers use their recruiting systems to filter or rank candidates at the first stage. And 48% filter out middle-skills candidates with a gap of more than six months in their work history. The report estimates more than 27 million Americans were "hidden workers" in 2020 — qualified, available, and invisible to these processes. Nearly every Fortune 500 company routes applications through an ATS: the report cites 99%, and Jobscan's own 2026 count puts it at 97.4%.
Then comes the human — and the human is fast. Ladders' 2018 eye-tracking update measured an average initial scan of 7.4 seconds (up from 6 seconds in 2012), with about 80% of that attention landing on six items: your name, current title, current company, previous title, dates, and education. A study of hiring at one large firm (Li, Raymond and Bergman) found it received roughly 200 applications per opening and rejected 95% of applicants at the initial resume review — humans doing the rejecting, deciding entirely from what was submitted through the portal. The machine narrows the pile; the human still makes the call, in seconds.
The stakes are now legal as well as practical: a certified nationwide collective action, Mobley v. Workday, alleges automated screening rejected applicants in violation of anti-discrimination law. The claims are unproven allegations, and Workday's own filings put roughly 1.1 billion applications rejected through its system in the relevant period — a measure of how much hiring now flows through these tools, whatever the outcome.
How to write a resume the machines read correctly
Write for the parser first, the scorer second, and the human third — in that order, because that is the order your resume is read.
1. Use one column and standard headings
Stick to "Experience," "Education," and "Skills." Parsers read columns top-to-bottom, left-to-right, and often scramble multi-column layouts. Headers and footers are frequently dropped entirely, so keep your name and contact details in the body of the page, not in a header.
2. Mirror the posting's language — without stuffing
Use the exact job title and skill terms from the posting where they are true for you. Modern scoring weighs meaning and context, not just matching words, so a skill named naturally inside a bullet about real work scores better than the same word repeated in a list at the bottom.
3. Put the six facts the human scan looks for at the top
Name, current title, current employer, dates, previous title, education — a recruiter's eyes land there in the first 7.4 seconds. If those six facts are clean and match the role, you earn the longer read. If they are buried, you may not get one.
4. Explain a gap in one line
Nearly half of employers filter out middle-skills candidates with a gap over six months. A single line — "Career break: full-time caregiving, 2024–2025" — gives the parser, and the human, something to file instead of a blank.
5. Test the parse yourself
Copy your resume, paste it into a plain-text file, and read what comes out. If the text is jumbled, the parser sees the same jumble. Fix the layout until the plain-text version reads in order.
6. Send the right file
A text-based PDF or .docx is fine for modern systems. Never send a scanned image or a design file where your text is a picture — to the parser, that resume is blank.
What not to do
- Keyword stuffing: repeating a skill twenty times in white text or a hidden block. Modern scoring weighs context, and any recruiter who spots it reads it as dishonesty.
- Tables, text boxes, and graphics for key facts: dates and titles inside tables or images are the first things a parser misplaces.
- Fancy fonts and symbols: stick to standard fonts. Unusual characters can fail to import, and a missing glyph is a missing fact.
- One resume for every job: tailor the top third — the summary and the skills lines — to each posting. The rest can stay the same.
Frequently asked questions
Should I submit a PDF or a Word document?
Either works for modern systems. Older parsers handle .docx slightly more reliably; if the posting doesn't specify, send a text-based PDF. The test that matters: can you select and copy the text in your file? If you can't, it's an image, and the parser can't read it either.
Do small companies use AI screening too?
Less often. ATS use is near-universal at Fortune 500 companies — 97.4% by Jobscan's 2026 count — but far less common at companies with a handful of employees. The same resume still wins there: a clean layout is also what a busy owner reads fastest.
Can AI write my resume for me?
It can draft it — our guide to AI tools for job seekers covers what to hand to AI and what to check yourself. Never let it invent experience you don't have, and check every number it writes: a fabricated job title or date is exactly the kind of mismatch the screening layer is built to catch.
Key Takeaway
The machine is not your enemy, and it is not a mystery. It is a parser, a scorer, and a summarizer working through 244 applications per opening because no human team can. The "75% rejected by a robot" figure is a myth; the real risks are a garbled parse, a wrong summary, and a knockout "no" you could have answered honestly. Write one clean column, mirror the posting's real language, explain gaps in a single line, and test the plain-text paste before you send it.
Your resume's first reader is software. Write it so the software gets you to the human.
Sources:
- Harvard Business School & Accenture, "Hidden Workers: Untapped Talent," September 2021 — full report (PDF)
- Greenhouse, "The Hire Standard" benchmark report, March 2026 — full report (PDF)
- Ladders, recruiter eye-tracking study update, November 2018
- SHRM, Talent Trends 2025 (survey of 2,040 HR professionals)
- Resume.org, August 2025 survey of 1,399 U.S. workers
- Enhancv, interviews with 25 U.S. recruiters, September–October 2025
- Christine Assaf, "Your Job Application Was Rejected by a Human, Not a Computer," HRTact, 2020

Comments
Post a Comment