Does AI Reject Your CV Before a Human Reads It?
The 75% ATS rejection claim is a myth. What actually screens your CV, and how to survive it.
The short answer
No. Not the way most career advice sites tell you.
The claim that "75% of resumes are rejected by robots before a human sees them" is the most repeated falsehood in modern job-search advice. The statistic traces back to a 2012 sales pitch from a defunct startup, with no published methodology or sample size. Jobscan, one of the largest ATS-optimization vendors, states the opposite plainly: applicant tracking systems do not reject resumes. They store them and let recruiters search using keywords. An Enhancv survey of US recruiters found 92% confirm their ATS does not auto-reject on formatting or content.
What is true is more interesting: a growing share of hiring teams now use AI somewhere in screening, and candidates are flooding applications with look-alike AI-written documents. Both trends change what you should do before you click submit. This guide separates what actually happens from what the internet insists, then gives you a workflow that survives both.
Where the myth comes from
The "75%" figure circulated for over a decade because it sells CV-optimization tools. Career sites repeat it, tool vendors quote it, and every job seeker remembers the headline even after the source falls apart.
What the research actually supports:
- Most ATS software stores, parses, and ranks applications for human review. It does not reject files in bulk. Formatting and layout influence how well the parser extracts your information, which is a real but different problem from "the robot discarded you."
- A minority of companies configure threshold filters that auto-discard applications below a match score or missing required skills. Survey data puts this at roughly 8% of organizations using an ATS, and such filters are usually set by recruiters, not by the software itself.
- Human reviewers still make the final call, but they make it fast and often keyword-driven.
The distinction matters: a poorly parsed CV is fixable. A weak, generic, or dishonest application is not.
What actually happens to your CV
A typical application in a company with automated screening passes through three stages:
Parsing. The system extracts your name, roles, dates, skills, and education. Unusual layouts, images containing text, and PDFs that store text as graphics break this step. If the parser loses your data, the human cannot search for it later.
Ranking. The system scores how well your profile matches the job's required and preferred skills. This is where keyword alignment matters. The score influences ordering, not a binary accept or reject in most setups.
Human review. A recruiter or hiring manager looks at the shortlist. This stage is fast. When someone scans a CV in seconds, they are looking for role title, measurable outcomes, and relevant tools, in that order.
The real filter is the human one, and it operates on first impressions. That is why application volume with the same generic CV fails: not because an algorithm hates you, but because repetition gives a reviewer nothing to anchor on.
The AI adoption numbers
Companies are using AI in hiring, and the trend is real even if the myth is not. SHRM reported that 39% of organizations had adopted AI in HR by December 2025, up from 26% in 2024, with recruiting the single most common use case. Adoption splits by company size: 60% of organizations with 5,000+ employees have implemented AI in HR, versus 33% of organizations with under 100 employees. In the US, 99% of hiring managers recently surveyed said they use AI somewhere in the hiring process, most often for scheduling interviews and screening resumes, alongside tasks like skills assessments and drafting candidate emails.
Across the EU, the AI Act classifies AI hiring tools as high-risk from August 2026, and GDPR Article 22 restricts purely automated decisions that produce legal effects. Portugal applies both directly. The direction across Europe is the same: AI assists screening, but a human remains accountable for the decision.
The new risk: look-alike AI resumes
The candidate side of AI created a different problem. Recruiters report an uptick in look-alike, AI-generated resumes: 64% noticed more of them in 2024-25, and this increased their screening workload. When many applicants use the same models with the same prompts, the results converge into the same structure, vocabulary, and bullet phrasing. Applications become homogeneous, and homogeneous applications are hard to rank.
The reaction is predictable. In a 2025 survey, 54% of hiring managers said they care whether a candidate applied with an AI-written resume or cover letter. The reasons split four ways: 36% read it as expertise with the technology, 24% as a lack of effort, 23% as ATS keyword optimization, and 17% as simply impersonal. No matter which interpretation wins in a given room, the safer position is that AI helps you write, and you own the final document.
AI-assisted drafting is not the problem. Using AI to fabricate experience is, and it is the fastest way to fail a reference check or a follow-up question.
What that means for your workflow
The practical job-search playbook looks the same whether you are in Lisbon, London, or a remote-first team. Four steps, in order:
1. Verify the listing before you invest. Stale listings and ghost jobs waste the same amount of effort whether a human or an algorithm processes your application. Check posting date, required skills against your profile, and signs the role is still live. A role that is not worth applying to does not need a tailored CV.
2. Tailor honestly, not by keyword stuffing. One version of your CV, tuned for one role, with claims you can defend in an interview: that is the repeatable pattern. Generic "AI-optimized" buzzwords read as filler to both parsers and reviewers. Translate your experience into the language of the role without inventing it.
3. Keep control of what gets sent. Automated application blasting produces volume, not signal. Before any document leaves your hands, review it once, as a whole, the way a recruiter would read it cold. Everything else is noise.
4. Treat AI as an editor, not an author. Structured CV drafts, tone adjustments, and interview question prep are where AI earns its keep. Fabricated dates, inflated titles, and invented projects are where it ends your candidacy.
How Lemu Work fits
Lemu Work is a career agent built around those four steps, not a blast tool:
- Verify: confirm the job is real and worth your time before you spend an hour on it
- Tailor: adapt CV and cover letter to the listing from your profile, with your explicit approval before any send
- Track: one pipeline for applications, interviews, and follow-ups, so nothing disappears between tools
- Prepare: interview prep linked to the actual application record, not a generic starting point from memory
The goal is fewer, better applications. When you land the role, export your records and leave: the product is designed to be deleted, not to become a permanent CRM.
It will not find you referrals or replace networking. No tool does that honestly. What it removes is the repetitive drafting and context-switching that burns most of your week.
Try it
Open the Lemu Work app and run one role you actually care about through the verify → tailor → apply flow. Then decide with evidence, not with a myth.
Sources
- Enhancv recruiter survey on ATS auto-rejection behavior, 2025
- Jobscan statement on ATS storing vs rejecting resumes
- JobCannon / Resumeworded investigations tracing the "75% ATS rejection" claim to a 2012 startup sales pitch
- SHRM, "The State of AI in HR," December 2025
- Insight Global, "2025 AI in Hiring Survey Report"
- ResumeBuilder, "GenAI Resume Effects" survey, 2024–2025
- EU AI Act (Regulation 2024/1689) and GDPR Article 22 on automated hiring decisions