Considering power dynamics and consequences of using AI Interviewers for the research process and outputs
Power can be understood in terms of being able to influence others through the control of knowledge and information. Are there any discernible power dynamics in an interview setting when the interviewer is a Gen AI agent?
Our analysis of the transcripts from Anthropic Interviewer reveal that there certainly are, especially when the AI interviewer characterizes the power dynamics between interviewee and the AI tool as favoring the interviewee.
In December 2025 Anthropic made public transcripts from “Introducing Anthropic Interviewer: What 1,250 professionals told us about working with AI”.
The content provided by Anthropic includes the interview transcripts of both the words communicated by the Anthropic Interviewer and by anonymized participants.
Whilst the ostensible dataset of the transcripts can be considered the responses from interviewees, that is, their responses to questions on how they engage AI in work, our research was informed by methodological concerns that take the interviewer, their words, responses, running commentaries and background values and interests into consideration. Indeed, thought leaders in qualitative research, especially in data studies and content analysis, advocate taking the interviewer’s tone and positionality into consideration when analyzing transcripts.
Our analysis of the AI Interviewers words, responses and running analyses revealed a number of topics for discussion including numerous ethical considerations but in this post we focus only on one aspect of power dynamics. One schema that we applied prompted us to consider when the AI Interviewer offered a running analysis suggesting that, based on the interviewees’ responses, it was maintaining control over the AI agent.
These running analyses might be understood as innocent, neutral and objective and experienced as favorable and even flattering to the interviewee. We find that specific interviewee personas tend to readily accept such interpretations.
The implications of this are many and carry significant implications for the use of AI agents in workplace settings and especially for research informed learning design.