Last year, a hiring manager at a mid-size fintech company told a reporter at The Verge that she’d started asking interviewees directly whether they’d used an AI assistant during the interview. Not to disqualify them, she said, but because the way candidates answered told her more about their self-awareness than most of the behavioral questions did. That detail stuck with me.
The conversation about AI tools in interviews has mostly been framed around cheating. That framing misses something. For a meaningful portion of candidates, these tools aren’t about getting answers they don’t know. They’re about staying calm enough to give answers they do know.
What interview anxiety actually does to performance
I wrote about this in more detail in a related post on interview anxiety, but the short version: high-stakes social evaluation triggers cortisol responses that reduce working memory capacity and verbal fluency. The candidate who blanked on a question they knew cold isn’t lying when they say they knew it. Their retrieval system was partially offline.
A 2023 paper in Frontiers in Psychology found that evaluative anxiety reduces spoken fluency and increases disfluency markers (filler words, restarts, pauses) in socially anxious individuals by nearly 34% compared to baseline. The effect was smaller but still present in non-anxious participants. So this isn’t just about people who have diagnosable anxiety disorders. It affects most people in high-stakes verbal evaluations.
What an AI assistant actually does in this context
Tools like Craqly run as overlays during live interviews, offering suggested talking points or answer frameworks as questions come in. The honest description of what this does to anxiety: it functions as a retrieval backup. If your working memory drops under pressure, the overlay gives you something to anchor to. Not a script to read, a structure to remember that you already knew.
This is different from the tool writing your answer for you. The candidate still has to articulate it, expand on it, field follow-ups. The tool can’t do that part.
There’s a version of this I think is undersold in the marketing: these tools are probably most useful for the 30 seconds after you get an unexpected question, when you know you have relevant experience but can’t immediately sequence it. That specific moment is where anxiety is highest and where the safety-net function is most useful.
The practice loop matters more than the live assist
Here’s an opinion I hold that could easily be wrong: for most candidates, the bigger benefit of AI interview tools isn’t in the live interview. It’s in the mock interview practice loop they enable.
Doing 15 practice sessions with a tool that gives you immediate feedback on pacing, relevance, and structure is stress inoculation. You’re building familiarity with evaluative discomfort in a low-stakes context. By interview day, the format is less novel, which reduces cortisol activation. This is the mechanism behind exposure therapy for performance anxiety, and it works at a milder scale even without a clinical protocol.
The APA’s summary of exposure-based approaches is worth reading if you want the theoretical grounding for why this works. The application to interview prep isn’t a huge leap.
What these tools don’t fix
An AI overlay doesn’t solve the underlying anxiety response. It manages one downstream effect of it (the retrieval gap). If you’re experiencing physiological symptoms before interviews, insomnia in the week leading up, panic responses, intrusive thoughts about failure, those require different interventions. Probably ones that involve a therapist, not a software tool.
The tools also don’t help with the part of the interview that’s about reading the room. Is the interviewer engaged? Are they skeptical? Did that answer land? Those are social perception tasks that anxiety disrupts, and no overlay helps with them directly.
I’m genuinely uncertain whether the confidence people report after using AI tools in interviews reflects a real reduction in anxiety or just a shift in what they’re anxious about (the content of their answer becomes less uncertain, but the social evaluation is still there). I’d want to see controlled studies on this before claiming the confidence benefit transfers to non-assisted interviews. I don’t think that evidence exists yet, at least not in published form.
The honesty question
More companies are explicitly asking about tool use. Some ban AI assistants during interviews. Others, like the fintech manager mentioned at the top, treat disclosure as a signal of self-awareness. Before using any tool in a live interview, check whether it’s permitted. This isn’t a gray area worth gambling on at the final-round stage.
For practice and prep, there’s no ethical issue. Use every tool available. The interview anxiety problem is real enough that you shouldn’t handicap your preparation on principle.
The harder question is whether the confidence you build with assistance translates to unassisted performance. That’s still genuinely open.