If you have applied for jobs recently and heard nothing back, you have probably run into the theory that a robot binned your application. AI CV screening has become the job seeker’s favourite villain — an invisible filter that supposedly throws away three-quarters of applications before any human opens them.
The reality is more boring, and more useful to know. AI does read your CV at most large employers. It rarely rejects it on its own. Understanding the difference tells you exactly where your application is actually getting stuck.
Does AI really reject your CV before a human sees it?
Usually, no. At most employers, AI CV screening ranks and summarises applications rather than deleting them. What genuinely filters candidates out automatically is not the AI at all — it is the set of yes/no “knockout questions” an employer configures, such as work authorisation or a minimum years-of-experience threshold.
That distinction matters because it changes what you should fix. If a scoring algorithm were the gatekeeper, the winning move would be gaming keywords. If a human recruiter is reading a ranked list, the winning move is landing near the top of that list and being readable once you get there.
Where the “75% of CVs get auto-rejected” claim came from
The single most repeated statistic in job-hunting advice is that 75% of CVs are rejected by applicant tracking systems before a human sees them. Career researchers who have tried to trace it keep arriving at the same dead end: a 2012 marketing pitch from a company called Preptel, which sold a product to beat those very systems and went out of business the following year. No study, sample size, or methodology was ever published.
The most-quoted number in job-search advice was a sales pitch, not a finding.
Recruiters who have audited real systems describe something different: applicant tracking software organises, sorts and ranks applications, and the great majority still get at least brief human review. Vendors have a commercial interest in the scarier version — the fear sells CV-optimisation subscriptions.
None of this means the process is fair or that you should ignore it. It means the failure point is usually competition and configuration, not a secret algorithmic bin.
How does AI CV screening actually work?
AI CV screening today runs in roughly three layers, and only one of them is new. Knowing which layer you failed at is the difference between a fixable formatting problem and a genuine mismatch with the role.
Layer 1: Parsing
Before anything scores you, software has to convert your document into structured fields — name, employers, dates, titles, skills. This is where creative CVs quietly break. Multi-column layouts, text inside tables or graphics, headings like “Where I’ve Been” instead of “Work Experience”, and text embedded in images can all parse badly. A CV that parses into empty fields looks like a weak candidate no matter how strong you are.
Layer 2: Knockout questions
These are the application-form questions employers mark as hard requirements: do you have the right to work here, do you have the licence, do you have five years in the field. Answers outside the accepted range can remove an application from the active pile automatically. This is the closest thing to true auto-rejection, and it is set by a human in the employer’s settings, not decided by a model.
Layer 3: Ranking and LLM summaries
The genuinely new layer is large language models reading a CV against a job description, scoring the fit, and writing the recruiter a short summary. It is far less literal than old keyword matching — it can recognise that “managed the P&L for a 40-person unit” satisfies a “budget ownership” requirement. It also means the recruiter’s first impression of you may be a machine-written paragraph you never see.
How do you get your CV past AI screening?
Write for a system that must parse you first and a human who must be persuaded second. Use a single-column layout, standard section headings, and the employer’s own vocabulary for the skills you genuinely have — placed inside achievement bullets with numbers, not stacked in a keyword block at the bottom.
The practical checklist:
- One column, standard headings. Summary, Work Experience, Skills, Education, Certifications. Save the design flourishes for a portfolio link.
- Mirror the job description’s language. If the ad says “demand forecasting” and you wrote “sales prediction”, use their phrase — where it is honestly true of your work.
- Put keywords inside evidence. “Cut stockouts 18% by rebuilding demand forecasting in Python” beats a bare “demand forecasting” in a skills list, for both the model and the recruiter.
- Lead with your target job title. A two-line summary at the top containing the exact role title anchors both the parser and the person.
- Answer the screening questions carefully. This is where real auto-filtering happens; a careless dropdown answer can end an application that your CV would have won.
- Quantify. Numbers survive summarisation. Adjectives do not.
- Never hide white text. Invisible keyword stuffing is detectable, and being caught reads as dishonesty rather than cleverness.
If you want to go further, our guide to whether you need to learn AI to keep your job covers which skills are worth adding to that CV in the first place, and our breakdown of AI roles and salaries in 2026 shows how employers are phrasing those requirements now.
What does the law say about AI in hiring?
Regulators have decided that hiring is one of the highest-stakes uses of AI, and job seekers now have real rights in several markets.
New York City moved first. Local Law 144 requires employers using an automated employment decision tool for hiring or promotion in the city to commission an annual independent bias audit, publish a summary of the results, and give candidates at least ten business days’ notice before the tool is used on them. Enforcement began in July 2023, with penalties starting at $500 and rising to $1,500 a day for continuing violations. The city’s Department of Consumer and Worker Protection FAQ sets out the details. Academic reviews have found compliance patchy — many covered employers simply never published an audit.
In the European Union, the EU AI Act classifies systems that filter job applications and evaluate candidates as high risk under Annex III, with CV-sorting software used as the textbook example. The full high-risk obligations for these stand-alone systems, originally due on 2 August 2026, have been pushed back to 2 December 2027 under the Digital Omnibus package. The transparency and AI-literacy duties arriving in August 2026 are unchanged, including the duty to tell people when they are dealing with an AI system.
The practical upshot for a candidate: in a growing number of places you are entitled to know that an automated tool is being used, and in New York City you can look up the published bias audit before you apply.
The honest summary
AI CV screening is real, it is nearly universal at large employers, and it is worth writing for. But it is a sorting and summarising machine sitting in front of a human decision, not an execution chamber. Most rejections are still the ordinary, deflating kind: a lot of qualified people applied, and someone else ranked higher.
The fixes that work are unglamorous — parseable formatting, honest keyword alignment, quantified achievements, and careful answers to the screening questions. None of them require beating a robot. They require making it easy for a machine to read you accurately, so a person can judge you fairly.
Frequently Asked Questions
Do companies really use AI to screen CVs?
Yes. Applicant tracking systems are near-universal among large employers, and an increasing share now add a large language model layer that scores CVs against the job description and writes summaries for recruiters. The AI usually ranks and summarises rather than rejecting outright.
Is the “75% of CVs are auto-rejected” statistic true?
There is no published research behind it. Attempts to trace the figure lead back to a 2012 sales pitch from a company that shut down in 2013 and never released a methodology. Treat it as marketing, not data.
What actually gets my application filtered out automatically?
Knockout questions on the application form — work authorisation, licences, minimum experience, location — are the main automatic filter, and employers configure them manually. Badly parsed formatting is the second common cause, because it makes a strong CV look empty.
Should I use ChatGPT to write my CV?
It is useful for tightening phrasing and mirroring a job description’s vocabulary, but everything it produces must be factually yours. Generic AI-written CVs read as interchangeable, and inflated claims collapse in an interview.
Can I ask an employer whether AI screened my application?
In several jurisdictions, yes. New York City requires advance notice to candidates and a published bias audit, and EU transparency rules from August 2026 require people to be told when they are interacting with an AI system.
