The Skill Shift

How Recruiters Use AI to Screen Resumes — and How to Beat It

How Recruiters Use AI to Screen Resumes — and How to Beat It

Updated September 2026.

You apply at 9 p.m. The rejection lands at 9:04. Nobody read it — or so it feels. Here is what the numbers actually say. In a June 2026 survey of 1,500 U.S. hiring managers by Resume Genius, 58% said they now use AI to screen resumes and applications, up from 35% a year earlier. But in the same company’s separate January 2026 survey of 1,000 hiring managers, only 6% said AI can reject or advance a candidate with limited human review. The machine is almost always in the loop. It is rarely the one deciding. Understanding the difference is the whole game, because you write differently for a filter than for a person — and in 2026 you have to get past both.

A conceptual editorial illustration of a tall stack of paper resumes passing through a translucent glass funnel that…

AI-generated illustration

The Real Version

AI screening is not one thing. It is a pipeline, and each stage does something different to your application. The first stage is parsing: an applicant tracking system (ATS) reads your file and turns it into structured fields — employer, title, dates, skills. The Resume Genius January report found 71% of hiring managers’ companies use an ATS, and 53% say the format that works best is a text-based PDF with no images or complex layout; only 13% say design-heavy layouts parse reliably. The second stage is the knockout: rules a human wrote in advance (must have this certification, must be in this country) that reject automatically. The third stage is ranking: software scores how closely the parsed fields match the job posting and sorts the pile. The fourth stage is a person reading the top of that sorted pile. And for a growing minority, there is a fifth: 23% of hiring managers say their company runs AI pre-screen interviews — automated phone screens or AI video interviews — before a human conversation.

The stage most candidates fear — a machine issuing the final no — is the least common, according to hiring managers themselves:

Who makes the screening decision?Share of hiring managers
Humans review resumes personally; the ATS flags, ranks, or organizes49%
The ATS screens out applications using criteria the team set37%
AI recommends or ranks, but humans make every final decision32%
AI screens out some candidates using human-written rules19%
AI can advance or reject candidates with limited human review6%

Source: Resume Genius 2026 Hiring Insights Report (1,000 U.S. hiring managers; multiple answers allowed). The same report captures why the software exists at all: 60% of hiring managers use an ATS to review applications faster and 50% to manage volume — a volume LinkedIn put at 11,000 applications per minute on its platform in 2025, up 45% in a year, as covered by The New York Times. SHRM‘s 2025 Talent Trends survey of 2,040 HR professionals points the same way: AI use in HR rose from 26% in 2024 to 43% in 2025, with recruiting the most common use. And the managers using these tools are not uncritical of them: 33% say their ATS “overemphasizes keyword matching over overall fit,” and 21% report parsing or formatting problems that make review harder.

In Plain Language

Think of the AI as the bouncer, not the host. The bouncer has a checklist written by the host — the stated requirements — and a clipboard that ranks the line by how well each person matches it. The bouncer decides who gets inside the door. The host decides who gets the job. Most “beat the ATS” advice treats the bouncer as the whole night: cram in keywords, game the score. But by employers’ own account, the score mostly gets you seen; a person then reads what you wrote. So the failure modes differ at each door. At the first, you fail by being unreadable (a two-column layout the parser scrambles) or by not stating a requirement you actually meet. At the second, you fail by sounding like a template. The Resume Genius January report found 80% of hiring managers say they can often tell when a resume was written by AI, and 76% say AI-written resumes make it harder to understand what a candidate actually did. The July report ranks heavily AI-generated materials as the second-biggest red flag (49%), behind only job-hopping (65%) — a double standard the report itself names, given 58% of those same managers use AI to screen. Two doors, two ways to fail, and most candidates only prepare for one.

A hiring manager's hands holding a printed resume across a desk during an interview, coffee cup beside a laptop — the human review stage that follows the automated one

A Concrete Example

When the bouncer gets it wrong, there is no host to appeal to — and one lawsuit shows what that looks like at scale. In 2023, Derek Mobley sued Workday, one of the largest HR software vendors, alleging its applicant-screening technology discriminated on the basis of age, race, and disability. According to SHRM‘s reporting, Mobley says that since 2017 he was rejected from more than 100 jobs at companies using Workday, often immediately or in the middle of the night, which he believes means the rejections were automated. In 2024, Judge Rita F. Lin ruled that Workday could be treated as an employer’s agent under federal anti-discrimination law because its software performs screening the employer would otherwise do; a nationwide age-discrimination collective action was approved in May 2025; and on June 22, 2026 the court denied Workday’s motion to dismiss California claims while narrowing others. Nothing has been decided on the merits, and Workday denies the allegations, saying its tools “do not make hiring decisions” and look “only at job qualifications, not protected traits like race, age, or disability.” A separate January 2026 class action makes a similar allegation against Eightfold AI — that low-ranked candidates were discarded before a human saw them — which Eightfold also denies.

The research record explains why courts are taking this seriously. A Stanford-led study of the pymetrics platform, reported by eWeek, analyzed 4 million applications across 156 employers from 2018 to 2022 and found adverse impact for Black applicants in roughly one in ten positions — and because 42 models were shared across those employers, rejection by one company’s model raised the odds of rejection by another’s. Harvard Business School’s Hidden Workers research (2021) found the systems used by 98% of Fortune 500 companies routinely screen out an estimated 27 million capable U.S. workers with gaps, no degree, or no recent full-time job; according to the study, 88% of executives admitted qualified candidates are “vetted out” for not matching exact criteria. Regulation is only beginning to bite: New York City’s Local Law 144 requires bias audits and a notice to candidates, but a December 2025 state audit summarized by DLA Piper found enforcement “ineffective” — and only 35% of hiring managers in the July survey say their company always discloses AI evaluation.

A stylized editorial illustration of a single wooden chess pawn standing on a chessboard where every other square holds…

AI-generated illustration

Why This Matters to You

Because the “beat the ATS” playbook you’ve probably seen optimizes for the wrong door. Here is what the data supports instead, layer by layer.

For the machine layer, be parseable and explicit. Use a single-column, text-based PDF or .docx with standard section headings — the formats 53% and 43% of hiring managers respectively say work best. State every requirement you meet in the posting’s own words, because the top reported rejection reasons are missing required skills or poor alignment with the job description (42%) and failing basic requirements (36%), not keyword count (28%). Make your work history unambiguous — titles, employers, months and years — since unclear history is the third-biggest reason (33%). Include a short summary and a skills section: 90% and 85% of hiring managers respectively say they want them.

For the human layer, sound like a person who did the work. Replace generic phrasing — the top AI tell at 51% — with specifics: what you did, for whom, and what changed. Cut buzzwords (41% flag them) and inflated claims (41%). If you use AI to draft, edit it until it reads like you; 60% of hiring managers say they want to see AI skills demonstrated, not claimed. And remember the human is still persuadable by things software cannot score: 79% say enthusiasm and a positive attitude make a candidate more hirable, and 39% have hired someone who reached them through an unconventional channel like a direct message — a second route past the bouncer entirely.

If you are over 40, have a gap, or suspect you are being screened out unfairly, know that the rules are changing but slowly. If you’re applying in New York City, you’re entitled to notice that an automated tool is used and can request an alternative process. Elsewhere, your leverage is legibility: the clearer your application, the less any model has to guess. The AI decides whether you’re seen. Make sure that when you are, there’s a person on the page.

Sources & References