How to Screen AI-Written Resumes: What Recruiters Can Actually Check

You cannot reliably tell whether a CV was written by AI, and you do not need to. Candidates who paste a job description into an AI tool get a CV that echoes its words, so keyword matching loses its signal. What still separates people is substance: specific work, real numbers, a believable path, and answers that hold up when you ask about them. Screen for evidence and test it in the conversation, not for style.

Why so many CVs now sound the same

A job description lists requirements, and an AI tool can rewrite a CV to mirror them. The result is a stack of CVs that all contain the same phrases. Recruiters in public discussions describe opening hundreds of them with the same language. The match between the CV and the advert is therefore a weak sign. It tells you the person can use a tool, not that they can do the job.

Do not rely on AI detectors

Tools that claim to detect AI writing are unreliable, especially on short, formal text like a CV, and they can wrongly accuse people who simply write plainly or write in a second language. Rejecting on a detector score is unfair and hard to defend. Judge what the CV shows, not how it was written.

What to look for instead: evidence behind each claim

For each requirement, look for the line that shows it: what they did, with what, for whom, and with what result. A phrase such as "strong communication skills" proves nothing. A line such as "ran a weekly review with three engineering teams and cut the release backlog by half" is checkable. If a CV repeats the requirement with nothing behind it, treat the requirement as unproven, not met.

Look at the path, not just the words

Check that dates, titles and responsibilities fit together. Does the seniority grow in a believable way? Do the tools match the years in which they were used? Inconsistencies are worth a question, not a rejection, because people also make honest mistakes.

Move the real test into the conversation

Ask the candidate to walk you through one project from the CV: what the problem was, what they decided, what went wrong. A person who did the work can answer follow-up questions easily. A short work sample or a structured set of questions, the same for every candidate, tells you far more than polished wording.

Change the job description to attract signal

Specific, honest job descriptions with a few real examples of the work give candidates less boilerplate to echo and give you more to compare. Ask one or two screening questions that need a real answer, such as describing a situation, so that a pasted paragraph stands out.

In Rezumz

Rezumz shows, next to each requirement, the line in the CV that supports it, or says plainly that nothing was found. Where a phrase appears with little behind it, the evidence view notes what is only partly shown. It does not claim to detect AI-written CVs, and the score is advice: a person decides who moves on.