Interviewing Candidates Who Practiced With a Chatbot

What is a behavioral interview question actually for? When you ask someone to tell you about a time a project slipped, you aren't really after the story. You're after what happens while they tell it: how they organize it, which parts they hurry past, where they hesitate when you follow up. The question works because the candidate is answering in real time, a little underprepared and slightly under pressure. The imperfection is what you're measuring.

That assumption is getting weaker. Candidates can now rehearse your interview format with AI before they ever talk to you. Google's Interview Warmup and Yoodli do this, and so does any ChatGPT prompt that says "pretend you're interviewing me for this role." Someone who has done fifteen practice sessions like that doesn't just know the STAR method. They've written and polished answers to every common version of "tell me about a time you handled conflict" and "describe a project that missed its deadline." They're better prepared, obviously. What's less obvious is whether the answer you hear tells you anything that wasn't written in advance.

Preparation itself isn't new. For twenty years, applying for a professional job has meant researching the company, getting a few STAR stories ready, and maybe practicing out loud with a friend. That never broke structured interviews, because it was slow and the feedback was thin. Your friend could tell you an answer rambled, and not much more.

What changed is how fast a candidate can iterate. An AI coach critiques delivery, points out vague language, suggests better evidence, and then runs the question again, as many times as you like. Getting from a first-draft answer to a polished one used to take a week of mock interviews. Now it takes an evening. Meanwhile your question bank hasn't changed much, and it's close to a public document. Motivated candidates can drill it far more thoroughly than was possible before.

So when you ask the standard questions, you're less likely to hear someone think through a real situation, and more likely to hear the version they've rehearsed eleven times, tightened by feedback and stripped of the pauses and self-corrections that used to tell you something. Structured interviews still have value. But the format was designed for a world where the interviewer knew what was coming and the candidate didn't, and that gap has mostly closed.

We've watched this happen once already, one stage earlier in the funnel. AI-generated applications are flooding ATS systems and making text-based screening much less useful. The interview is now getting the same treatment.

The part you can't rehearse

So what can't be rehearsed? Mostly the follow-up. A fixed list of questions can be anticipated. A conversation that listens to the answer and asks about what was actually said can't be scripted, because the candidate doesn't know which part of their story will get picked up.

That's why Asendia screens candidates in a spoken conversation instead of a question list. We chose voice AI for this on purpose. A spoken answer happens in real time, so there's less room to reach for a prepared paragraph, and whatever the person says gives the next question something to grab onto. Asendia phones candidates soon after they apply, often the same evening, and follows what they say: it asks for specifics when an answer is vague, pushes for a concrete example when things stay abstract, and steers back when a response drifts away from what the role needs. Suppose a candidate has a polished answer ready for "tell me about your leadership style." The follow-up might be: "You mentioned you led a team through a reorganization. Who on that team pushed back, and what did you do about it?" That's a hard question to have prepared for, and the answer tells you a lot more.

What the recruiter gets back is a structured qualification summary with the candidate's own words quoted, so they can read how the person talked about their experience when they weren't reciting. I think that holds up better in a hiring debrief than a rubric score does, because it shows how someone thinks and not just how well they've been coached. The summary is saved to the candidate's record in the ATS. A shift like this also changes which pipeline numbers are worth watching, which I wrote about in a post on recruitment KPIs built for a pre-AI world.

None of this is an argument for penalizing candidates who practice with AI. They're doing what any sensible person would do with a free tool. But if a few evenings with a chatbot can substantially raise someone's interview score, the interview was measuring preparation more than ability, and that's a problem with the process rather than with the candidate. Refreshing the question bank won't fix it, since the new questions will get drilled just as quickly. What helps is screening that responds to what people actually say, and I suspect that will matter more every year as the coaching tools get better.

Ready to transform your hiring strategy? Schedule a Demo with our founders today!