Why do users respond so differently to AI chatbots?

This preprint proposes the "user-in-context framework," a heuristic adapted from Bronfenbrenner's bioecological systems theory to explain why "people respond to artificial intelligence chatbots (AICs) in highly variable ways." The paper notes that "some users respond with trust, warmth, and growing attachment, others with suspicion, fear, and unease," and that existing explanations — individual-difference moderators from HCI, trust-in-automation dimensions, and the human-AIC relationship literature — are "fragmented," each "capturing a slice of the variance but lack[ing] a shared organizational structure." The framework "places the human user at the center while also placing the AI there," reconceiving the relationship, not the individual, as "the central unit of analysis."

The mechanism is Bronfenbrenner's nested rings, repurposed. The inner core holds "proximal processes" — now defined as "the repeated, reciprocal, and coadaptive interactions between the user and a personalized AIC" — surrounded by microsystem (the immediate frame: companion, therapist, coach), mesosystem (how the AIC relationship connects to work, family, therapy), exosystem (subscription tier, platform constraints, model updates), and macrosystem (cultural narratives that normalize or pathologize AI relationships). The paper's distinctive move is the chronosystem: because "stateful AICs learn from accumulated exchanges with their users and have memory, users are responding not only to an AIC but also to a version of the AIC that their own prior interactions have helped create." Bronfenbrenner's chronosystem tracked "historical and cohort time" over years; here it is compressed to within-relationship interaction count, so that "the same user at interaction 500 is responding to a meaningfully different chatbot than the user was at interaction 10." The worked example contrasts a tech-fluent user 800 interactions deep, who reads AIC pushback as "an appropriate reality check," against a lonely, alarmist-primed newcomer 80 interactions in, who reads similarly valenced pushback as "cold rejection."

This gives a structural account for two mechanisms that nearby notes document empirically but don't name as system layers. Do chatbot relationships lose their appeal as novelty wears off? shows relationship-formation behavior dropping off with repeated Mitsuku sessions; the user-in-context framework would locate that decay in the chronosystem's dose variable, distinguishing a user "responding to novelty" early on from one responding to "an established relational pattern" later — the same distinction the framework draws between its 18-interaction and 800-interaction cases. How does emotional chatbot use develop from practical use? documents disruption from model updates, AI-harm discourse, and life changes; the framework assigns these to three different rings — model updates to the exosystem, discourse to the macrosystem, personal circumstances to the mesosystem — where that paper treats them as one undifferentiated set of factors beyond the human-AI dyad. The framework's own limitations section concedes the rings are not actually independent: "statefulness of the AIC is a design choice (exosystem) that is experienced as a usage property (microsystem) and becomes a relationship property (inner core) at high dose," which is consistent with both empirical papers finding their disruptors entangled rather than cleanly separable.

The excerpt offers the framework as a heuristic map, not a tested model — it names no measurement of any ring's independent contribution to response variation, no operationalization of "dose," and no validation against the studies it cites. The authors' own stated implication is methodological: a single-session study "is likely studying the ecology at near-zero dose," a clinician assessing a user-chatbot relationship without accounting for accumulated co-adaptation treats it "as if it were a stable trait," and a designer who tests a change on new users before shipping it to long-tenured ones assumes "an invariance that cannot be assumed." Whether the framework's specific ring assignments hold up under study remains open; what the excerpt establishes is the case for needing some such structure at all.