Contemporary digital information environments increasingly expose citizens to conspiratorial content and AI-generated political communication, yet little is known about how these cues shape citizens' subjective sense of political competence. We examine whether exposure to conspiratorial narratives and AI source cues affects internal political efficacy using a randomized survey experiment conducted among 1,500 adults in Spain. Respondents were randomly assigned to one of four conditions: a neutral placebo, a conspiratorial narrative, the same narrative presented as AI-generated, or a conspiracy portraying AI as a hidden conspiratorial actor. Contrary to our preregistered expectations, exposure to conspiratorial narratives produced small but positive effects on internal political efficacy, with the strongest and most consistent effects observed when the narrative was attributed to AI authorship. Exploratory mediation analysis shows that perceived credibility does not explain this effect: the AI-authored narrative was rated less credible than the placebo, yet still increased efficacy — a pattern consistent with suppression, though the indirect effect itself does not reach significance. These findings challenge the assumption that conspiracy narratives necessarily undermine internal political efficacy and point to AI source cues as an underexplored driver of citizens' subjective political competence.

Building on this insight, we propose and experimentally test an alternative design strategy based on balanced, mid-prevalence (“moderate”) filler items in a context of false beliefs and a conspiracy theory (the sensitive item). Rather than maximizing variance through heterogeneous lists, this approach aims to stabilize response distributions and reduce the cognitive and strategic burden associated with polarized item sets. We argue that such lists may lower respondents’ awareness of the sensitive item and mitigate strategic responding by reducing the salience of any single item.

We field two list survey experiments that vary both list length (short vs. long, with the sensitive item included only in the long list) and filler composition (extreme vs. moderate). Across experiments, the sensitive item captures the prevalence of the same conspiratorial belief. This design allows us to assess which filler strategy—balanced (moderate) or polarized (extreme)—more effectively improves preference revelation relative to direct questioning. Our design contrasts two competing logics of ICT optimization: variance maximization through heterogeneous fillers versus distributional balance through moderate fillers. We evaluate these strategies in terms of prevalence estimates, variance, and statistical efficiency, the incidence of ceiling and floor effects, and potential design effects. More specifically, we expect that moderate filler lists will increase the estimated prevalence of sensitive attitudes, reduce ceiling and floor effects by avoiding extreme response patterns, and impr

By directly comparing these approaches, this study contributes to the methodological literature on indirect questioning by providing experimental evidence on how subtle features of list construction shape the validity and reliability of ICT-based estimates.