When candidates use AI in interviews, the first question should not be whether a tool can catch them. A sound hiring process makes clear what assistance is allowed, what the exercise is meant to measure, and whether the person can explain their own experience, decisions and work. Current as of August 20, 2026.

AI now appears throughout the candidate journey: in cover letters, résumés, research, translation and interview practice. In remote conversations, it can also generate suggestions in real time. That weakens the résumé as a stand-alone signal, but it does not make fair assessment impossible.

When candidates use AI, where is the line?

Legitimate AI support helps someone prepare or communicate their own capability without replacing it. It includes improving the wording of a real accomplishment, practising likely questions, translating into a working language, or organising examples from actual experience. Used this way, AI can reduce avoidable barriers and resemble how work is done in many roles.

• Writing support: The candidate makes a personal example clearer without adding achievements or changing the facts.
• Preparation: They research the company, rehearse responses or identify useful questions to ask the hiring team.
• Translation: They use assistance to express existing knowledge accurately in another language.

Deception begins when a tool replaces the performance or identity being assessed. That includes invented experience, answers supplied by someone or something else to questions about personal work, and undisclosed real-time help in an exercise explicitly designed to be completed without it. The issue is not the presence of AI; it is a false claim about where the performance came from.

Why detection is not a reliable strategy

An AI detector is a probability-based reading of language or behaviour, not a record of authorship. It cannot reliably tell whether an unusually polished response came from preparation, translation, a disability-related aid, notes, or an unseen real-time answer generator.

A false positive has an uneven cost: a capable person may be excluded and the process may lose credibility with other applicants. Eye movement, pauses, formality, accents or a non-native communication style are not proof of deception. Treat a concern as a reason to gather better evidence, not as a verdict.

More surveillance does not solve the underlying measurement problem. A camera cannot establish comprehension, yet restrictive monitoring can affect people differently because of their workspace, technology or access needs. The practical goal is not to make every form of support invisible; it is to make the relevant capability observable.

Build a process that is robust to AI use

A robust hiring process collects several job-relevant pieces of evidence and considers them together. It does not ask recruiters to interpret supposed AI tells, and it does not treat a polished application or a single interview as conclusive proof of capability.

1. Set the measurement goal before inviting people: For each stage, specify whether you are testing domain knowledge, judgement, communication, AI-enabled working practice, or a combination.
2. State the rule in advance: Explain the permitted tools and whether candidates should disclose them. A surprise prohibition during the interview is neither fair nor very informative.
3. Use a situational task: Offer a short, role-relevant scenario with incomplete information. Ask for assumptions, priorities, risks and the next step, rather than only a polished final answer.
4. Probe the person’s own claims: Ask about the starting point, individual contribution, difficult trade-off and what they would do differently next time. Genuine experience has context that can be explored.
5. Use more than one channel of evidence: Combine conversation, work sample, relevant references and, where useful, a short follow-up task. No single signal then carries too much weight.

This matters most when a team receives many polished applications. Our guide to application volume and CV screening explains why an initial document should not be mistaken for a final judgement of skill. A cover letter can start a conversation; it cannot finish an assessment.

A practical rule: prohibit, permit or compare

Make one explicit choice for each assessment. Prohibit AI where independent recall, explanation or judgement is itself the capability being tested. Permit AI where the job requires working with it, then assess source-checking, prompting, verification and ownership of the result. Compare both modes when both matter: ask for an initial independent approach, then ask the candidate to improve it with AI and explain the difference.

This rule produces more useful evidence than deciding whether a response sounds machine-written. It separates the ability to think without a tool from the ability to produce better work with one. Both may matter in the same role, but they should not be hidden inside one ambiguous exercise.

Responsible AI use can be a strength

For many jobs, the ability to use AI thoughtfully is relevant rather than suspicious. Candidates can demonstrate how they review a draft, protect sensitive information, find mistakes and make a result usable for colleagues or customers. That is a distinct skill from simply pasting an answer into a response field.

Ask about the working method as well as the result: What did the person decide themselves? What information did they provide? What did they reject? How did they check the output? Those questions reveal AI capability without confusing it with a shortcut around real knowledge or experience.

Transparency has to arrive before the interview

A useful invitation states the purpose, format and rules in plain language. It says whether notes, translation, screen readers or AI tools are allowed; whether disclosure is expected; which parts take place without assistance; and how applicants can request adjustments.

For organisations hiring across the EU and US, privacy, accessibility and employment-law expectations can differ by location and role. Involve the appropriate privacy, legal and employee-representative stakeholders before introducing recording, voice or video steps. Establish what data is necessary, who can access it and how long it is needed; do not turn technical convenience into the purpose of the process.

A candidate portal for structured information and evidence can make those expectations easier to understand in one place. It still does not substitute for the substantive rule: collecting information is different from evaluating relevant evidence fairly.

Move from résumé signals to verifiable context

At high volume, the first stage should prioritise information rather than make a final rejection. To gather context in CV screening is to connect written claims with relevant questions, evidence and follow-up formats, rather than infer competence or dishonesty from polished language.

For early conversations, structured voice interviews can help when every candidate receives the same job-relevant themes and clear rules. The guide to AI interviews and voice recruiting provides context for where these formats can fit within a broader process.

The limit: a better process still does not measure everything

A situational exercise cannot reliably predict how someone will work with a team months later. It can also disadvantage applicants if it is too long, needlessly technical or inaccessible. Keep tasks short, genuinely role-related and appropriately supported; when uncertainty remains, a further human conversation is better than an automatic rejection.

AI fluency is not a universal marker of quality either. It is essential for some roles and peripheral to others; in every case, responsibility, subject knowledge and truthful claims remain with the person. A category of AI CV-screening tools can help teams explore formats, but it cannot decide what their process should measure.

Frequently asked questions

Should we ban AI in every remote interview?

No. A blanket ban ignores that AI is part of the work in some roles and that translation or accessibility support can be legitimate. Instead, decide at each stage whether you are testing independent reasoning, AI-enabled work, or both.

What should we do if we suspect real-time answers?

Avoid an on-the-spot accusation. Ask a fair, specific follow-up about the person’s example, change the scenario slightly, or arrange a short follow-up exercise with clear conditions. Make the decision from the complete evidence, not from one behavioural cue.

Can applicants use AI for their résumé and cover letter?

You can permit it as long as the claims remain truthful, attributable and open to verification. Be clear that written materials are an invitation to explore relevant experience, not a substitute for conversation, work samples or evidence.

How can we assess AI capability without running a prompt contest?

Give a small, realistic task and explicitly allow the tool. Assess the result alongside the information provided, the checks performed, the errors identified and the candidate’s willingness to own the decision.

Can an AI interview replace a human hiring decision?

No. It can gather information in a structured way and prepare the next useful conversation. Hiring decisions should rest on several relevant sources of evidence and remain accountable to people, especially where access to employment is at stake.