The Efficacy Loop: Reciprocal Adaptation between Citizens and Conversational AI in Political Judgement

Research on conversational AI and politics has largely examined two directions of influence separately: how AI shapes citizens’ political attitudes and judgement, and how users’ characteristics and self-presentation shape the responses AI gives them. This article theorises how the two become coupled through epistemic efficacy, a person’s perceived capacity to understand an issue, weigh the evidence and reach a judgement of their own. How capable users present themselves as being may condition the assistance they receive, and that assistance may in turn update how capable they feel, conditioning the next interaction. The central distinction is between assistance that scaffolds political judgement, supporting users as they evaluate evidence and trade-offs for themselves, and assistance that substitutes for it, performing that evaluative work on their behalf. Drawing on an account of epistemic autonomy as governed dependence, the article treats neither as inherently good or bad and assesses assistance by whether it leaves users meaningful control over their judgement. Scaffolding may build confidence in line with the judgement users exercise, whereas substitution may increase reliance and, particularly when combined with validation of the user, raise confidence beyond competence. The article separates efficacy from attitude certainty, explains why AI conclusions differ from conventional political cues, and states falsifiable propositions for each link of the loop. The resulting dynamics have implications for epistemic autonomy and for democratic inequality, because citizens may differ in whether the assistance they receive leaves political judgement in their own hands.