Guide

Can a recruiter tell if a CV was written by AI?

Last updated 11 October 2026

Sometimes, but not reliably, and not from the writing alone. Research from 2024 to 2026 finds that AI-text detectors still make large errors. Readers who often use AI writing tools can spot AI text in short articles, but we found no study that tested CVs, and no sign is proof. Our advice is to check at the screening question or the interview, where the candidate has to explain the detail in their own words.

What are the signs a CV was written by AI?

No sign proves a CV was written by AI. Researchers have measured some tells in articles and abstracts, employers report others, and each can also appear in writing by people who never used AI.

Each sign below is labelled by where it comes from: research, a recruitment firm, or our own practical advice.

Research on articles and abstracts, not CVs: stock vocabulary. Kobak and others (Science Advances, 2025) found a sudden rise in certain style words in biomedical abstracts after AI writing tools appeared, with “delves” among the words showing strong excess usage. In a 2025 study by Russell, Karpinska and Iyyer, expert readers named “AI vocabulary” as the most common giveaway. The words on such lists are ordinary words that many people use in a CV.
Research on articles, not CVs: formulaic shape. The same expert readers pointed to “not only … but also …” phrasing, examples listed in groups of three, and optimistically vague conclusions.
One recruitment executive’s view: polish and sameness. In September 2026 the trade press reported a survey by the recruitment firm Robert Half. One of its executives said it is harder to identify the best fit when large volumes of applications “appear highly polished or follow near-identical formats”. The report gave no sample size.
A recruitment firm’s statement: inflated detail. Robert Half’s March 2026 release says, without giving a figure, that “in some cases, generative AI tools are fabricating or embellishing work history and skills”.
Our advice, not tested in a study: claims you cannot test. Be curious when the wording would fit any employer, or when the CV does not match the application form. These are reasons to ask a question, nothing more.

None of these is proof. A style guide, a template or a person who happens to write that way can produce every one of them. Kobak and others say they cannot separate direct AI use from “people adopting LLM-preferred words and borrowing them for their own writing”, and they describe their 13.5% estimate for 2024 abstracts as a lower bound for a whole collection of texts, not a verdict on one document.

Do AI detection tools work in 2026?

Not well enough to rely on for a decision about a person. A 2024 benchmark reported that many detectors claim accuracy of 99% or more, and found that adding a repetition penalty to the generating model, a setting that changes how the text is produced, cut accuracy by up to 32 points. A preprint from August 2026 tested 13 open research detectors, not commercial products, on older academic manuscripts by non-native English writers. False-positive rates ranged from 0.0% to 100.0% depending on the detector.

The 2024 benchmark tested general text in eight domains. Neither study tested CVs, and we found no detector study that did. In July 2023 OpenAI withdrew its own detector “due to its low rate of accuracy”. We have not tested any tool on sale today, and we do not recommend one.

Can a person spot AI writing?

Some people can, on some texts. In a 2025 study of English non-fiction articles, annotators who rarely or never use AI writing tools performed near random, while a majority vote among five who frequently use them misclassified only 1 of 300 articles.

Readers who often use AI writing tools did better than most detectors in that study, and text altered to sound more human “does not fool our experts in aggregate”. Individual experts did less well than the group vote: they caught 92.7% of AI articles on average, and each also wrongly flagged a few human-written articles. Readers who rarely use AI tools also overestimated themselves, giving high confidence scores to decisions that were close to guesses.

Limits matter. The texts were American English non-fiction articles under 1,000 words, not CVs. The 2025 study shows that practice with AI writing helps. It does not show that a recruiter can reliably spot a CV.

Is “it reads like AI” a safe reason to reject a candidate?

No. Detectors and readers both make mistakes, and the mistakes can land on people who wrote every word themselves. In 2023, researchers found that several detectors consistently misclassified non-native English writing as AI-generated, and an August 2026 preprint named editing style as a key confounding variable in detector scores.

A reader’s feeling that a CV sounds like AI is a guess about style, and in the 2025 study, readers who rarely use AI tools guessed near random. If you reject on that feeling alone, you may turn away a candidate who wrote every word.

Where can you check an AI-written CV?

Our advice, not a research finding: check at the screening question or the interview, where the candidate has to explain the detail in their own words. We found no study showing AI-written CVs being caught this way. Two sources point the same direction. A 2025 preprint on cover letters on an online labour platform reported that the link between a cover letter's textual alignment and callbacks fell by 51% and that employers shifted toward other signals, including prior work history. A March 2026 Robert Half release on a US survey said 38% of respondents had increased the number of interviews per candidate.

These sources suggest the written application tells employers less than it did. In our view, the step where the candidate speaks is the hardest for borrowed words to survive. A person who did the work can usually explain it, give the awkward details and answer a follow-up. A person who did not has to improvise. Either way, the mismatch is the evidence, whoever wrote the words.

Ask every candidate the same job-related questions. Put the same questions, in the same words, to each person. Our guide to screening questions for job applicants in Ireland shows how.
Ask for a specific example, then follow up. What was the task, what did you do, what went wrong, what would you change? Ask about the detail on the CV, such as a number, a tool or a team.
Use a short practical check where the job allows. For a driver, a route question; for an office role, a small task. Give it to every candidate for the role.
Verify the must-haves. If the role needs a licence or certificate, check it, whoever wrote the CV.
When answers do not match the CV, give the candidate a chance to explain. Note what was asked and what they said. Decide on the mismatch, not on a suspicion about the writing.

Can interviewers tell if a candidate is using AI in a live interview?

Not reliably. We found no study of how well interviewers spot AI help in a live conversation, and the sources we found warn mainly about impersonation. In January 2025 the FBI said it had seen North Korean IT workers using artificial intelligence and face-swapping technology during video job interviews, and it advised identity checks during interviewing, onboarding and employment of remote workers.

That warning is about fraud, which is a different thing from a real candidate using AI for help. On that, a 2025 report by Greenhouse, a hiring software company, said only 9% of UK and 12% of Irish candidates feel it is “totally acceptable” to use AI during live interviews. The page gave no sample or question wording.

What helps is the same approach as for a CV. Ask follow-up questions on the detail of an answer, ask the candidate to explain a choice they made, and ask the same questions of every candidate. Do not decide on a suspicion alone. For remote hires, check identity as the FBI advises: during interviewing, onboarding and employment.

About Fios AI

Fios AI does not offer an AI-writing detector and does not claim to tell whether a CV or an answer was written with AI. Fios AI is recruitment software built in Ireland for roles that draw many applicants. Aoibheann, an AI interviewer, runs a first interview by video, asking the questions the recruiter approved for that role, and a recruiter chooses whether to proceed, hold or decline for each candidate who completes the interview. A failed photo ID check can end a candidate’s progress before that point. Fios AI is in private build, and access opens through the waitlist. See the FAQ.

Sources

This page is general information, not legal advice.