Verifiable Eco-Recommendations by AI Travel Assistants: Eye-Tracking and GSR Evidence on Verification, Trust Calibration, and Sustainable Hotel Booking

AI travel assistants are increasingly designating hotels as “eco”, yet when the evidence is not independently verifiable, these recommendations may serve as persuasive cues or credible decision support. We present a preregistered 2 × 2 between-subject laboratory experiment (n = 63) that manipulates autonomy framing (Recommend vs. Plan) and evidence verifiability (verifiable vs. non-verifiable) in a realistic hotel-booking workflow with a standardized “Verify eco-claim” drawer. Phasic arousal was recorded at recommendation onset (E1) and verification initiation (E3), employing eye-tracking indexed verification behavior (verify clicks, time-to-verify, verification depth) and event-locked galvanic skin response (GSR). Verifiability did not directly speed up or deepen verification (H1 not supported), but verification was common (74.6% clicked Verify). Rather, autonomy influenced checking: Plan slowed verification and altered verification depth. E1 SCR revealed an Evidence × Autonomy interaction, which is consistent with an autonomy-boundary account (H4), rather than credibility stress emerging as a simple evidence main effect at E1 (H2 not supported as stated). Verification served as a repair moment: depending on the availability of diagnostic cues, arousal dynamics from E1 to E3 supported differential “repair” (H3). SCR dynamics explained incremental variance in perceived manipulation/greenwashing concern beyond condition and eye-tracking indices (H5b supported), but verification depth did not mediate effects on trust or delegation (H5a not supported). Overall, users’ interpretation of AI sustainability advice is influenced by autonomy, and multimodal process measures offer useful signals for auditing eco-recommendation designs in travel platforms.