Answers are grounded in this paper and cite their sources.
This article examines how evaluative dialogue changes when an evaluation professional's interlocutor is a large language model (LLM) rather than a human colleague. Drawing on interviews, focus groups, and structured reflections from 13 impact investing evaluation practitioners, the study identifies three properties that human interlocutors bring to evaluative dialogue—personal commitments and experience, shared organizational knowledge, and accountability to colleagues—and traces what happens when each is missing. Each absence proved both productive and limiting: LLMs broadened the options considered, exposed uncodified conventions, and reduced the social cost of candor. Yet they could be steered toward nearly any conclusion, could not read unwritten organizational norms, and, with nothing at stake, offered agreement and pushback that carried little weight. Prompting and document uploads compensated only partially. The findings position LLMs as preparatory interlocutors rather than substitutes for collegial exchange and clarify what evaluative dialogue requires of its participants.
No takes yet. Share an insight, caveat, or question.
Maoz Brown (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: