Open any careers blog and you'll find the stat. "75% of CVs are auto-rejected by the ATS before a human sees them." It's been repeated so often it has the texture of fact.
It traces back to a 2012 sales pitch from a company called Preptel. Preptel sold resume-optimisation software. No methodology was ever published. Preptel went out of business in 2013. Forbes cited it in 2014, CIO cited Forbes in 2018, CNBC cited CIO in 2019, and the number has laundered into conventional wisdom every single year since (The Interview Guys).
A 2025 HR.com survey of recruiters found the opposite. 92% said their ATS does not auto-reject CVs for formatting (HR.com).
The real 2026 screening stack is weirder than the myth. Kinder, actually. And considerably more beatable, if you know what it's actually doing.
Who runs the systems
97.8% of the Fortune 500 use an ATS. Workday and SAP SuccessFactors alone cover 52% of F500 recruiting (Jobscan). Add iCIMS, Greenhouse, Oracle Taleo, Lever, and you have the pipeline almost everyone goes through.
In 2026, these systems are triage. A ranked list of 1,200 people for a tired human to work down from the top.
The real five-step stack
Here's what an application actually passes through.
1. Knockout questions. Are you authorised to work in this country? Do you have five years of X? Do you have a required certification? Conversational AI platforms like Paradox (Olivia) and HireVue Hiring Assistant do this via SMS, chat, or a form. McDonald's cut its hiring time in half by routing every applicant through Paradox first (Index.dev). Knockout questions are where most applicants get filtered out, and they're completely explicit. If you answer "no" to the work-auth question, you're out. That's by design.
2. Keyword and skill ranking. Workday, iCIMS, Greenhouse and their peers score your CV against the job description. Workday does this via HiredScore (acquired Feb 2024). iCIMS does it via Copilot. SuccessFactors via the Talent Intelligence Hub. The critical thing: they produce a ranking. You end up at some position in a ranked list that the recruiter works down from the top.
3. Semantic / LLM screening. Eightfold's "Talent Intelligence" uses vector embeddings, so a candidate who wrote "built machine learning models" matches a JD that says "ML engineering." Lever's Talent Fit does the same thing. Keyword stuffing is less useful than it used to be. Mirroring the language of the JD still helps, but the system also understands synonyms.
4. The human 7–11 second scan. A 2018 Ladders eye-tracking study found recruiters looked at a CV for 7.4 seconds before making a stay-or-go call. That number hasn't improved in 2026. It's arguably worse, because the ranked list in front of them is 1,200 people long. The recruiter's eyes go: current title → current company → dates → previous title → previous company → dates → education. Then the call.
5. AI interviews. HireVue has run 70 million+ AI-scored video interviews, serving 60%+ of the Fortune 100 (HireVue). Paradox has handled 200+ million candidate chat engagements. HireVue killed facial-expression analysis in March 2020 after an independent algorithm audit (SHRM). What's left is language-content analysis. Criterion validity is around r = 0.24 across five organisational samples (HireVue / Liff et al. 2024). Respectable. Not mystical.
So when someone asks "did the algorithm reject me?", the honest answer in 2026 is usually no.. a knockout question rejected you, or a tired human skipped past you at position #400 in a ranked list. The algorithm ordered the list.
Things that are false but widely repeated
Since the Preptel number is wrong, it's worth going through what else is.
- "The ATS can't read PDFs." It can. EDLIGO's 2025 analysis of 1,000 rejected CVs found PDFs had an 18% parse-failure rate versus 4% for DOCX, but that 18% was almost all image-based scanned PDFs. A text PDF exported from Word or Google Docs parses fine on every major system.
- "The ATS rejects tables and columns." Modern parsers handle both. Image-heavy CV templates with embedded graphics are a different story.
- "Stuff the CV with keywords to hit 80% match and you'll get an interview." Semantic-match tools already handle synonyms. Past a low threshold, keyword density stops helping.
- "Cover letters are dead." Resume Genius's 2025 hiring-manager survey found 83% still read them, 45% before the CV, and 81% have rejected applicants based solely on the cover letter (Resume Genius). What's dead is the generic AI-generated cover letter. Every one now passes some threshold and none stand out (Wharton / Fortune, March 2026).
- "HireVue judges your facial expressions." No. That was killed in 2020.
What actually works
With the stack laid out, the tactics that help are unglamorous and specific.
Tailor, but not the way you think. ResumeGo's 2024 field experiment on 7,000 applications found tailored cover letters produced 16.4% callbacks, against 12.5% for generic letters and 10.7% for none. A 53% lift over generic. The tailoring doesn't need to be heavy. One paragraph that references the actual product, team, or business problem does more than a reshuffled bullet list.
Quantify. LinkedIn's own guidance finds CVs with quantified achievements produce roughly 2.5× the interview rate. "Cut deploy time from 45 minutes to 4" beats "improved deployment process."
Match the recruiter's eye path. Current title, company, dates, previous title, company, dates, education. That's the 7-second scan. Those are the eight data points a human is grabbing. If any are ambiguous (employment gap, weird title), handle them in the first line.
Get referred. Referred candidates are hired at about 30% versus 7% for cold applicants (Pinpoint; Ashby). One referral is worth roughly 40 cold applications in expected-interview value. Single biggest lever in the whole stack.
Use AI, don't let it write the final thing. 62% of hiring managers say they can spot and reject AI-generated CVs that haven't been personalised. 90% report a noticeable increase in low-effort AI-assisted submissions (Resume Now, March 2025). Draft with AI. Rewrite in your voice. Put specific things only you would know.
The legal pressure changing the stack
Two things are reshaping what employers are allowed to do with AI screening.
In August 2023, the EEOC settled the first federal AI-hiring discrimination case against iTutorGroup for $365,000. The company's software was set to reject female applicants aged 55+ and male applicants aged 60+. The plaintiff figured it out by submitting two identical applications that differed only in birth date. The older DOB got rejected. The younger got an interview (EEOC).
In May 2025, a federal judge granted conditional class certification in Mobley v. Workday, an ADEA class action covering every US applicant aged 40+ who applied through a Workday-screened employer since September 2020. Workday argued it was a platform, not an employer. The court said it could be liable as an "agent" of the employer (Proskauer). The first AI-vendor class action in US employment history.
Add the EU AI Act's August 2, 2026 compliance deadline (hiring is classified as high-risk; fines up to 7% of global turnover) and Illinois HB 3773 (in force January 2026), and the 2026 trajectory is: more scrutiny, more bias audits, less willingness to automate rejection end-to-end.
Read the system, use the system
The honest picture is this: you're not fighting a machine that rejects you in milliseconds. You're sitting at position #400 in a ranked list being worked by a tired person with 14 other reqs open. That's the whole thing, right?
Every tactic that helps you move up that list is real work. The specific company reference, the quantified achievement, the one-line explanation of a gap, the referral, the portfolio link. Every tactic that doesn't (keyword stuffing, five different font sizes, a gorgeous two-column Canva template) is work that feels productive while moving nothing.
Flint scores jobs against you across six dimensions, using broadly the same techniques modern screeners use to score you against a JD. Having read what the screeners actually do, it's worth building your own.
Drop your CV into Flint free, in about 60 seconds. You'll see your top 10 scored matches right away, and where you sit for each one.