Half of Rejected Candidates Blame a Robot.
Follow the Number Back and It Falls Apart.
Read before you rewrite your CV for a machine
Quick answer
The famous claim that around 75 percent of CVs are auto-rejected by software has no traceable source. Applicant tracking systems mostly sort, search and store applications rather than delete them. What really happens is a very fast human skim. Where AI genuinely does decide is later in the process, in one-way video interviews and automated screening calls, and that is where candidates are losing without ever being told. Write for the human skim, and prepare separately for the recorded interview.
You send 40 applications. You hear nothing. Somewhere in that silence a thought arrives, and it is a comforting one: a machine threw it away before a person ever looked. Half of job seekers who were rejected without a word believe exactly that. It is the most repeated idea in job hunting right now, and the evidence for it is remarkably thin.
The number everyone repeats
You have seen the figure. 75 percent of CVs never reach a human. Sometimes it is 70 percent. Sometimes 88. It appears in career advice, in LinkedIn posts, in the marketing of every product that promises to beat the robot.
Try to follow it to its source and it dissolves. There is no employer study behind it, no published method, no sample size. The trail leads back to the sales material of a CV-checking company that closed more than a decade ago. That is the first thing worth noticing. A real statistic has one value and one method. This one has three values and none.
Why this matters more than it sounds
If you believe a keyword filter is deleting your CV, you will do specific things to fix it. You will stuff the document with phrases lifted from the advert, strip out formatting, repeat the job title, and write sentences aimed at software rather than a person. Then a person reads it, finds it hollow and repetitive, and moves on. The myth does not just waste your time. It actively makes your application worse.
What an applicant tracking system actually does
The software most employers use is a filing system with a search box. It receives your application, parses it into fields, stores it, and lets a recruiter sort and search. Recruiters can configure knockout questions, and many do: right to work, required licence, location, minimum qualification. Those are real and they do filter. But they are questions you answered yourself on the form, not a secret judgement passed on your prose.
Sorting is not rejecting
A recruiter searching a database of 900 applicants for a specific certification is not deleting the rest. They are choosing where to start. You can be findable, or you can be buried on page 11. That is worth optimising for, and it is a completely different task from beating a filter.
The knockout questions are visible
Do you have the right to work. Do you hold this licence. Do you have this many years of experience. These are on the application form, and answering them wrongly or leaving them blank genuinely ends the application. Read them properly instead of rushing them to get to the CV upload.
Parsing failures are the real technical risk
This is where formatting matters, and it is much narrower than people think. Text inside images, tables and text boxes, headers and footers, and unusual fonts can come through scrambled. A clean single-column document saved as a standard file avoids nearly all of it. That is the whole fix. It does not require keyword stuffing.
Then a human skims, fast
The decision that actually rejected you was made by a person moving quickly through a long list. They were looking for a reason to shortlist, found nothing in the first few lines, and moved on. Almost every rejection people attribute to a robot is really this.
Where AI genuinely is deciding
None of this means AI has stayed out of hiring. It has moved somewhere later and more consequential, and far fewer people are talking about that part. Greenhouse surveyed 2,950 job seekers across the US, UK, Ireland, Germany and Australia and published the findings in April 2026. The picture is not about CV filters at all.
| What the survey found | Share of candidates |
|---|---|
| Have already been through an AI interview (US respondents) | 63 percent |
| Say the use of AI was not clearly disclosed to them beforehand | 70 percent |
| Have withdrawn from a hiring process because it involved an AI interview | 38 percent |
| Never received any outcome after completing an AI interview | 51 percent |
| Think disclosure of AI use should be a legal requirement | 57 percent |
Source: Greenhouse candidate AI interview report, published April 2026, based on 2,950 job seekers in the US, UK, Ireland, Germany and Australia. Figures marked as US respondents apply to that group.
Read that table again with your own experience in mind. The one-way recorded interview, where you talk to a countdown timer and nobody replies afterwards, is not a fringe practice. It is common, it is frequently undisclosed, and it very often ends in silence. That silence is the thing people are misdiagnosing as a CV robot.
“The machine is not reading your CV. It is watching you answer question three with 11 seconds left on the clock.”
What to change, in order
Write the first four lines for a person in a hurry
Not a summary of your character. A statement of what you do, at what level, in what sector, with one number attached. If a stranger cannot tell within four lines whether to keep reading, nothing further down the page will save you.
Keep the document machine-readable, then stop
One column, real text rather than images, no critical information trapped in headers, footers or text boxes, a normal font. That is the entire technical requirement. Spend the time you save on the writing.
Mirror the language of the role once, honestly
If the advert says stakeholder management and you have done it, use their phrase rather than your internal one. This helps a recruiter searching the database find you. Repeating it nine times helps nobody and reads as desperate.
Prepare for the recorded interview as a separate skill
It is not a conversation and it does not reward the same behaviour. There is no interviewer to read, no chance to ask a clarifying question, and a hard time limit. Practise answering out loud to a timer before it happens for real, because your first attempt should not be the one that counts.
Ask, politely, what the process involves
A short line in your reply to the recruiter asking what stages the process includes and whether any are automated is entirely reasonable. Most candidates never ask. Given how many report that AI use was not disclosed, asking puts you ahead of the majority.
The four-line test
The scenario
The opening of a CV, as read by someone with a stack of 200
✗ Avoid
A highly motivated and results-driven professional with a proven track record of delivering excellence in fast-paced environments, seeking a challenging opportunity to leverage my skills and contribute to a dynamic organisation.
✓ Use instead
Operations manager, 11 years in pharmaceutical distribution. Currently accountable for a four-site network moving 40,000 units a month and a team of 34. Reduced order-to-dispatch time from six days to two. Now relocating to the UK with full right to work.
The first version could belong to anyone in any industry at any level, which is precisely why it gets skimmed past. The second answers what, how senior, how big and any obvious objection, before the reader has finished their coffee.
Before you send the next application
- Read your first four lines out loud. If they could describe a different person in a different industry, rewrite them.
- Check every knockout question on the form. Blank or careless answers end applications far more reliably than any keyword filter.
- Open your own CV as plain text and see what survives. If sections vanish, the layout is the problem.
- Record yourself answering a common question in 90 seconds. Watch it back once. That is more preparation than most candidates do.
- Stop rewriting for the robot. Start rewriting for the six seconds of human attention you actually get.
There is something almost reassuring about believing a machine rejected you. It means the decision was arbitrary and nothing was lost. The harder and more useful truth is that a person looked, briefly, and was not persuaded. That is a problem you can actually do something about.
Go deeper
Did AI Write This?
Employers and universities now screen applications for AI-written text. Here is what they check, and how to stay on the right side of it.
Read it →Did this help?
About the author
Muskan
Co-founder, WizardTeacher · Content and curriculum lead
Muskan is co-founder of WizardTeacher and leads all content and curriculum development. She researches exam formats, marking criteria, and preparation strategies to ensure every resource on WizardTeacher is grounded in what the examiners actually reward.