Why this matters
An interview can only inform a decision as well as the evidence it collects. Understanding AI’s interviewing weaknesses can help candidates make their experience clear and help organizations assess whether a conversation produced useful answers.
Core finding
AI can sound like a capable interviewer while struggling to decide what to ask next.
Researchers Shangzhe Dai, Feiyu Duan, and Zhongyu Wei developed InterviewBench, a tool that uses real interview transcripts. They evaluated 23 language models on individual interviewing decisions and tested four in simulated text conversations across 184 scenarios.
The central gap was between asking questions and obtaining clear answers. For example, in the structured setting without added disruptions, Gemini 3.1 Pro covered 100% of planned questions, but explicit target information appeared for only 65.4% of items.
The harder task was often drawing out the answer. Once relevant information was expressed, stronger models generally converted it into a structured record reliably.
The models benefited from being told which interviewing action to take, such as clarifying an ambiguous answer. They struggled though when choosing the next action step themselves. In other words, knowing how to phrase a follow-up was easier than recognizing which follow-up the conversation needed.
Giving the interviewer greater freedom did not consistently improve performance and often reduced efficiency. Misunderstandings or refusals could also lower AI-rated interview satisfaction even when the system continued through its question list.
Takeaway
These were text simulations based mainly on English media interviews. The study did not directly test commercial hiring platforms or whether interview scores predict job performance.
For job candidates, a reasonable practical implication is to make key evidence explicit: explain the situation, your actions, and the outcome. Relying on an AI interviewer to uncover every missing detail may leave your story incomplete. Besides detail, length and coherence are important. Having a long enough response to be adequately evaluated and communicating well enough to properly articulate the event described in the story.
Better stories are possible — and they start with preparation.
AI Interview Training Course