Ask the question a script can’t answer.

Candidates have always prepared answers, and they should. What’s newer is that some read answers generated in real time by an AI tool, or recite a polished story that isn’t really theirs. You don’t need to catch every case. You need an interview where a memorized or generated answer doesn’t do well on its own, and a review process that doesn’t punish people for being nervous, careful or speaking in a second language. This guide covers the signs worth noticing, the ones that aren’t evidence, follow-up techniques that separate memory from a script, and what to do when you suspect one.

Preparation isn’t the problem

Good interview advice tells candidates to prepare stories, often using the STAR method: situation, task, action, result. A candidate who arrives with three well-structured examples has done exactly what they were told. That’s a sign of effort, not dishonesty. Many interviews also allow notes.

What you’re really screening for is whether the experience behind the answer is real and the candidate’s own. Someone who did the work can go past the prepared version. They know what happened next, what went wrong, and what they’d change. Someone reading a script usually can’t.

Signs an answer may be read or rehearsed

None of these is proof on its own. Together, and especially when an answer falls apart under a follow-up, they’re worth a closer look.

• It answers a nearby question. The answer fits the question the candidate expected, not the one you asked.
• It’s complete but empty. It covers every angle and names nothing specific: no numbers, tools, dates, places or outcomes.
• It sounds written. “Firstly… additionally… in summary” is how people write essays, not how they talk.
• The pacing is unusual. A long silence, then a fluent paragraph with no pauses or restarts, repeated after every question, simple ones included.
• The voice changes. Casual small talk, then a sudden shift in vocabulary and tone as each answer starts.
• It’s all “we”. The story describes what the team did, and the candidate’s own part never becomes clear.
• It collapses under a follow-up. The candidate can’t add a single detail, or gives a second answer that contradicts the first.

Signs that aren’t evidence

Many things that look like scripting are just people being people. Don’t mark someone down for:

• Nerves. Pauses, filler words and restarted sentences are normal, especially early in an interview.
• Looking away. People look up or away when they think. On video, they often look at their own picture.
• Accent, grammar or mixing languages. Someone speaking a second language may prepare more carefully and sound more formal. That’s a language pattern, not a script.
• Short answers from someone new to interviews. They may not know how much detail you want. A follow-up usually fixes it.
• A flat or formal delivery. Some people speak that way all the time.

A candidate who is nervous but real will usually get more specific when you ask for detail. A candidate reading a script will usually get less specific.

Follow-up techniques that separate memory from a script

The best defense is a good follow-up. These work because the answers aren’t in any prepared script.

You can also design questions that are harder to script: situational questions with details specific to your company, and explain-it questions like “Explain our refund policy to a customer who’s annoyed”. There’s no generic answer to copy.

Reading conduct flags

LessRounds adds conduct flags when answers seem evasive, generic, very short, scripted or off topic. Treat a flag as a prompt to check, not a verdict. Open the transcript, read the flagged answer, and play the recording. Scoring works from the transcript and never looks at faces, eyes or backgrounds, so a flag reflects what was said.

In AI-led interviews, the interviewer asks follow-ups based on each answer, which does some of the probing above for you. Separately, conduct warnings cover things like abuse or trying to manipulate the interviewer, and two of them end the interview. Sounding rehearsed isn’t one of those things.

What to do when you suspect a script

• Don’t reject on suspicion alone. Look at the whole interview. One polished answer among several real ones is probably just a well-practiced story.
• Use the next round. If the scores are strong but you’re unsure, move the candidate forward and probe the same examples live. A real story holds up.
• Ask neutrally. “Walk me through that project again. What was the hardest week?” Don’t accuse anyone. Let the detail, or the lack of it, speak.
• Apply the same standard to everyone. If you probe one candidate’s polished answer, probe the others’ too. Suspicion that only lands on some candidates, such as those with an accent or an unusual background, is its own kind of bias.
• Keep a note of what you saw. Write down the specific answer and what made you unsure. It keeps the next interviewer focused on facts rather than a vague impression.
• Set expectations up front. Tell candidates notes are fine but you want to hear about their own experience in their own words. Most people follow the rules they’re given. See telling candidates about AI interviews.