On this page: Key takeaways · How many candidates are using AI to apply and interview? · Why are AI detectors the wrong answer? · What does AI change in a frontline interview, and what does it leave alone? · How do you design a frontline interview around job evidence? · What does this mean for staffing agencies and high volume employers? · What are the limits of this approach? · How Radius helps · FAQ
Key takeaways
- Generative AI is now ordinary: about a third of people aged 16 to 74 in the EU used it in 2025, and 44.7% in the Netherlands.
- AI detectors flag honest people, especially those writing in a second language, and a simple prompt gets around them.
- A polished CV or a generic answer was never strong evidence. Concrete detail about past work still is.
- Check identity, right to work and certificates from the documents before the offer, and never reject anyone on a suspicion of AI use.
How many candidates are using AI to apply and interview?
Many, and the share depends on how you ask.
In 2025, 32.7% of people aged 16 to 74 in the EU had used generative AI tools in the previous three months, and 44.7% in the Netherlands (Eurostat, 16 December 2025; Netherlands from dataset isoc_ai_iaiu). In the US, the CNBC and SurveyMonkey Quarterly AI and Jobs Survey (1,686 students and workers, 22 to 27 July 2026) found that 47% of workers said they had used AI to apply for jobs, with help writing a CV among the most common uses (CNBC, 23 September 2026; sample). It is a US media poll, not an independent study.
Neither figure measures AI use during live interviews: most published figures on that come from vendors that sell hiring tools, and we have not used them. Assume any written application may have had help, and design the screen so that it does not matter.
Why are AI detectors the wrong answer?
They wrongly flag honest people and are easy to get around.
A study in the journal Patterns tested seven widely used GPT detectors on 91 TOEFL essays by second language writers of English and 88 essays by US school pupils. The detectors read the pupils' essays correctly but labelled more than half of the TOEFL essays as AI generated: an average false positive rate of 61.3%. When the researchers used ChatGPT to make the TOEFL essays sound like a native speaker, that rate fell to 11.6%. And when ChatGPT rewrote its own essays in more literary language, detection fell to near zero (Liang et al., 2023).
Essays by second language writers wrongly flagged as AI generated
Average false positive rate of seven GPT detectors on 91 TOEFL essays
The detectors punished plain English and rewarded AI polish. The study tested written English essays, not interview answers, but many frontline applicants answer in a language that is not their first, and a screen that treats simple wording as suspicious turns away people it should hear.
What does AI change in a frontline interview, and what does it leave alone?
It makes fluent wording cheap. It does not supply lived detail or valid documents.
| What you see | Can AI help a candidate produce it? | What to do with it |
|---|---|---|
| A fluent CV or motivation text | Easily | Read it as a list of claims to check |
| A generic answer to "tell me about teamwork" | Easily | Replace the question with a situation from the job |
| Specific detail about past work: the truck, the site, the problem | Hard to keep consistent under a second question | Ask for it, then ask about one detail |
| Identity, right to work, a valid certificate | Wording cannot produce them | Check the document before the offer |
As we argue in We Hate CVs, a CV shows what someone wrote about themselves; AI only makes the writing better. A forged document is a separate problem, covered in which identity, right to work and certificate checks can happen before the offer.
How do you design a frontline interview around job evidence?
Ask about real situations, follow up on details and judge everyone the same way.
A large review of selection research found structured interviews to be the top ranked selection procedure for predicting job performance (Sackett et al., 2022). Structure also ties every question to the job, which makes scripted answers less useful.
A six step checklist
- Start from the requirements. List the four to six things the role needs, such as a reach truck certificate, night shifts or the safety check before a lift.
- Ask about situations, not qualities. "The last time a load did not fit the trailer, what did you do?" tells you more than "Are you a problem solver?"
- Plan one follow up per answer. Ask for a detail that someone who did the work would know: which machine, which site, what went wrong.
- Use the same questions and written criteria for everyone in the role, so answers can be compared (how to run structured interviews at high volume).
- Say what help is fine. Tell candidates up front that preparing is fine and that answers during the interview should be their own.
- Never reject on suspicion. If an answer sounds scripted, ask another question or invite the candidate to a call. A person decides (why the tool should never be the one saying no).
Illustrative example. Two applicants for a reach truck role send CVs that both say "experienced reach truck driver, safety focused". Both are asked: "What is the highest racking you have worked at, and how did you check the load before lifting?" One describes the racking height and an overhanging pallet she stopped to rewrap. The other gives a general answer about safety and, asked which truck he used, says mostly a counterbalance truck. Neither answer proves a certificate: that comes from the document. More in three years on reach trucks.
See a short, structured voice interview in the AI Interviewer.
What does this mean for staffing agencies and high volume employers?
At volume, detection does not scale, and design does.
No recruiter can listen for a scripted tone across hundreds of interviews a week. Consistent questions make answers comparable across branches and clients, and evidence per requirement lets an agency show a client why a candidate was put forward. Candidates who used a writing or translation tool on their application should not lose out for it.
What are the limits of this approach?
Better questions reduce the value of AI help without removing it.
- A follow up question shows consistency, not truth: some honest candidates are nervous and vague.
- We expect detectors to misjudge second language writers in other languages too, but that is our view, not a finding.
How Radius helps
Radius Hire is the AI hiring platform for frontline roles. It starts from a structured interview and the evidence for each requirement, not from a CV.
- The AI Interviewer is a short, structured voice interview that candidates can take on their phone, whenever it suits them, in the language set for the role. Candidates are told at the start that it is automated and that a person reviews the result. Each interview can be kept as a recording, a transcript and a summary.
- Match shows, for each requirement, whether it is backed by a document or a check, rests only on what the candidate said, or is still missing.
- The AI Interviewer gathers answers and does not check documents. Where the employer uses Radius Passport, it checks identity and right to work with a trusted verification partner and records certificates with their expiry dates, so a recruiter can review the evidence before an offer.
- Radius never rejects a candidate on its own. A recruiter reviews the evidence, can overrule anything Radius presents and makes every hiring decision.