No detectable change in mental-health information-seeking during Italy's 2023 ChatGPT ban: a natural experiment with a verified first stage

Regulatory bans on AI platforms are accelerating globally, yet their behavioural consequences for affected populations remain unmeasured. Italy's temporary ChatGPT ban (31 March – 28 April 2023) provides a rare natural experiment: an exogenous, country-specific shock to AI access. I verify the first stage using Google Trends data on VPN and ChatGPT searches — Italian VPN searches peaked at nearly five times their pre-ban level on the first day of the ban, averaged twice their baseline throughout, and returned to pre-ban levels within a week of reinstatement; no equivalent change occurred in France or Spain. Against this confirmed exposure, I find no detectable change in loneliness-related information-seeking. The 2023 ban-window deviation (+3.5 index points above the preceding eight weeks) was smaller than in 2019 and 2022 (+19.3 and +19.4 points respectively) but larger than 2024 (−12.0 points); year-permutation tests return p ≥ 0.25 under both calendar and Easter alignment. An apparent "rise then fall" pattern in an earlier, single-draw version of this analysis reflected seasonal variation and single-draw sampling noise: three independent draws correlate at r ≈ 0.74 and show no consistent post-reinstatement decline. No construct shows a ban effect once seasonality is accounted for: the only secondary signal, a rise in sadness-related search that survives multiplicity correction and exceeds all placebo windows, is matched or exceeded by the same spring weeks in other years. I identify three methodological lessons for Google Trends research: the consequences of retaining sparse terms, relying on a single data pull, and using grammatically gendered phrases whose baseline asymmetry may reflect lexical rather than demographic differences.