Making Interviews More Effective in an AI World

For decades, behavioural interviews have been treated as one of the most reliable ways to predict job performance. Evidence has support the use of retrospective questions such as “tell me about a time when…” rather than hypothetical questions such as “what would you do if…?”. The logic is intuitive and research backed. Past behaviour is the best predictor of future performance. Candidates are asked to describe something they have actually done rather than something they believe they might do.

The research base for structured interviewing remains strong, particularly when interviews are combined with another job relevant assessment such as a work sample, simulation or personality assessment.

Behavioural interviews are also relatively easy to design, train and embed. That has contributed to their popularity but it may also have created a degree of complacency. Hiring managers and candidates can go through the process on the assumption that asking a structured question and applying a scoring guide will automatically produce a reliable outcome. The method is not necessarily the problem, however the assumptions surrounding its use may be changing.

Historically, candidates were less likely to have rehearsed every probable question. They had more limited external support when identifying examples, structuring answers and filling gaps in their evidence. Interviewers were therefore more likely to hear relatively spontaneous responses that revealed something about how the candidate thought as well as what they had done.

Today, candidates can predict likely interview questions, generate examples for almost any competency, identify missing detail and rehears responses repeatedly. Generative AI has made this process easier and more sophisticated. While preparation is also not inherently negative, candidates should be expected to understand the role and think carefully about the evidence they want to present. The concern is that interviews may increasingly distinguish between levels of preparation rather than levels of capability.

Furthermore, AI software can now listen to questions in virtual interviews in real time and provide answers in the moment during an interview. Answers may now demonstrate the quality of the candidate’s prompting more than their judgement or experience.

There is a wider problem too. Behavioural interviewing can become highly procedural, particularly within hiring manager populations that have not been trained in effective interviewing. The interviewer asks a standard competency question and expects an answer in the STAR format. The Situation, Task, Action and Result has been a useful framework for helping candidates present evidence clearly. It encourages them to explain the context, describe what they needed to achieve, outline what they personally did and clarify the outcome.

This has led to complacency where the candidate provides a well-rehearsed answer following the expected format. The interviewer asks a series of predetermined prompts, following the same predictable format, and scores the response against a checklist. This moves interviewing closer to a process than an assessment.

However, following STAR is not evidence of capability in itself. Many calibration discussions will include observations such as “the candidate followed the STAR model“, “the answer was easy to follow” or “they covered all the actions”. These comments tell us something about the structure and presentation of the answer. They do not necessarily tell us whether the candidate demonstrated sound judgement, understood the complexity of the situation or would make similarly effective decisions in a different context.

The solution is not to abandon behavioural interviewing. Skilled interviewers have always challenged assumptions, explored alternative actions and tested the depth of a candidate’s understanding. The issue is that many organisations have reduced behavioural interviewing to the collection and scoring examples. A more active approach would combine hypothetical and experiential questioning.

Instead of always beginning with “Tell me about a time…”, the interviewer might start with a realistic problem:

“Imagine this happened tomorrow. What would you do?”. They could then explore the candidate’s options, priorities, assumptions, risks and likely consequences. What information would they need? What might change their decision? Who would they involve? What could go wrong?

This could then be connected to experience:

“When have you faced something similar?”

“What did you do differently?”

“What did you learn?”

“How would that experience affect your approach now?”

This combination has the potential to reveal critical thinking, judgement and relevant experience. It is harder to rehearse fully, gives the interviewer more opportunities to test whether the candidate genuinely understands the answer they are giving and will be harder for AI software to respond to in the moment.

The question is no longer simply whether retrospective or hypothetical interviews are better. It is which interview approach is most resistant to AI responses while still providing fair, consistent and job relevant evidence of capability. This requires more than rewriting a question bank. It requires hiring managers to become more active interviewers. They need to present realistic problems, challenge assumptions, follow interesting threads and adapt their questions in response to what the candidate says.

Ironically, AI may push interviewing back towards what the best interviewers have always done, having genuine conversations that uncover how candidates think rather than simply collecting polished examples.

Omni can help organisations review and redesign their interview approach, combining behavioural evidence with realistic scenarios and more effective probing. We can develop structured interview frameworks, scoring guidance and interviewer training that retain consistency and fairness while giving hiring managers a much deeper understanding of genuine candidate capability.