By Saul Albert
At the start of summer 2025, in the grand council chambers of Cardiff University’s Glamorgan Building, William Housley, Patrik Dahl, Elizabeth Stokoe and Saul Albert, joined by Conversational AI industry expert Cathy Pearl hosted the “First Pair Prompt”. This hybrid workshop gathered 40 EM/CA scholars from around the world to discuss, as Liz Stokoe puts it in her LinkedIn summary, “which conversation analytic and ethnomethodological signal(s) should cut through AI LLM ‘conversational’ noise?”
Participants work together in The Glamorgan Building Chamber and online. Photo: Liz Stokoe
The workshop took the form of a ‘promptathon’, inspired by the format of hands-on technical ‘hackathons’ where participants spent most of their time in hybrid breakout working groups, focused on a range of analytic puzzles thrown up by the current wave of new consumer technologies “that travel under the sign of AI” (Suchman, 2023). These included finding analytic approaches to model-generated videos of vox-pop interactions, talking to an LLM-based patient avatar in a doctor-patient interaction training simulator, and building our own voice agents by writing LLM system prompts.
The aim was to draw together as wide a range of critical, empirical responses to these speculatively situated technologies as possible. As generative tools move into every corner of social and institutional life, we need new analytic perspectives and strategies.
Most importantly, we had a lot of fun, spending most of the time engaged in rich, longer-form conversations in smaller groups, occasionally sharing in hybrid plenary sessions. But the workshop also clearly achieved its central aim: participants discovered many novel perspicuous settings and artifacts being (re)configured by these technologies. Participants also noted key limitations in methodologically ‘standardised’ approaches to transcription, collection, and analysis within EM/CA when straightforwardly applied to these new materials and settings. For example, for many ‘AI’-generated artifacts and data, there is no interactionally grounded endogenous rationale and structure, so some basic tenets of ethnomethodological inquiry do not necessarily hold. As one participant reminded us in their notes on analysing AI-simulated interactions, generated representations of interactional settings and data are not always linked to indexically accessible evidence of “the objective reality of social facts [ … i.e.,] that and just how every society’s locally, endogenously produced, naturally organized, naturally accountable, ongoing, practical achievement, [is] everywhere, always, only, exactly and entirely members’ work, with no time out, and with no possibility of evasion, hiding out, passing, postponement, or buyouts.” (Garfinkel, 1996 p. 11).
So while the First Pair Prompt was only a possible first, we felt there was certainly an enthusiastic enough response to warrant a Second Pair Prompt on Friday the 12th June 2026 in Loughborough UK and online.