User Heterogeneity in AI-Mediated Communication: Extending Cognitive Theories via Latent Class Analysis

Status: Accepted at CogSci 2026 (Rio de Janeiro, Brazil).

This paper challenges the dominant “one-size-fits-all” assumption in AI-mediated communication research by quantifying user heterogeneity—the systematic individual differences in how people experience and adapt to real-time AI scaffolding during conversation. We conducted a Wizard-of-Oz study (N = 40) of context-adaptive scaffolds and applied latent class analysis (LCA) to qualitative interview data (n = 29 in-depth coded transcripts).

LCA identified eight two-class groupings across thematic domains (e.g., AI usage and trust, communication strategies, feature-specific evaluations). Critically, user profiles in one domain did not predict profiles in another, indicating that adaptation patterns are domain-specific rather than reflecting a global user typology. This finding extends classical cognitive theories of language processing—which often assume between-individual variation aggregates into stable types—and motivates personalization frameworks that adapt across multiple, partially independent dimensions of user experience.

Recommended citation: He, W. P. & Fussell, S. R. (2026). "User Heterogeneity in AI-Mediated Communication: Extending Cognitive Theories via Latent Class Analysis." Proceedings of the 48th Annual Meeting of the Cognitive Science Society (CogSci 2026). Rio de Janeiro, Brazil. (Accepted)