A short research paper from Cassette and the University of York
AI interview simulators are increasingly used as training tools, yet there is little evidence on whether people find them credible enough to learn from. Dr Daniel Johnston, from our team at Cassette, set out to test this in a short research paper written in partnership with the University of York. Cassette provided the technology: a voice-based interviewer that read each participant’s CV, asked tailored follow-up questions and gave feedback at the end.
Fourteen students took part, with their trust and perceptions measured before and after a session. It is a feasibility study, and the paper is clear that its findings are early signals rather than conclusions. Three stand out.
A single session markedly reduced participants’ distrust of the system but did not build active trust.
Credibility came from personalisation, from questions grounded in the participant’s own experience. It was undermined by small breaks in the conversation: awkward pauses, a voice that shifted mid-session, praise that repeated itself. Participants didn’t need the AI to feel human. They needed it to be consistent.
For anyone putting AI into coaching or development roles, the paper ends on the right question: not how realistic it can be made, but how reliably it helps someone improve.
Read the full paper, Useful, Not Human, by Dr Daniel Johnston (Cassette) and Sophia Hudelist (University of York).