AHCI Comparative Longitudinal Dataset — AI Hallucination Crisis Recovery Trajectory across Four Brand-Incident Pairs in Five Anglophone Markets (2022-2026)

The AHCI comparative longitudinal dataset operationalises a four-brand naturalistic empirical design testing the AI Hallucination Crisis Recovery Asymmetry Hypothesis through search-behavioural data. The corpus comprises 4,014 weekly observations across four brand-incident pairs spanning the full Coombs (2007) Situational Crisis Communication Theory (SCCT) cluster space: Air Canada chatbot bereavement fares case (preventable, organisational responsibility, February 2024), Google Gemini image generation diversity scandal (preventable, algorithmic bias, February 2024), Microsoft Tay chatbot inappropriate responses (preventable, historical reference, March 2016), and OpenAI ChatGPT defamation hallucination cases (accidental, technical limitation framing, 2023-2024). Five anglophone markets are sampled: United States, Great Britain, Australia, Canada, and Ireland, across the period from January 2022 to April 2026. Each brand cell is operationalised through a paired trust-query (functional consumer searches indicating recovery) and failure-query (crisis-related searches indicating distrust signal) using the pytrends API. The AI Hallucination Crisis Index (AHCI) operationalises the ratio of failure searches to combined searches: AHCI = Failure / (Trust + Failure), with values ranging from 0.0 (full trust-recovery) through 0.5 (contested) to 1.0 (crisis-dominant). Theoretical anchors include Coombs (2007) Situational Crisis Communication Theory, Coombs and Tachkova (2023) triadic appraisal model integrating moral outrage into SCCT, Spearing, Gile and Fogwill (2025) source-discreditation interventions for AI-generated misinformation, Hwang and Jeong (2025) AI hallucination forewarning effects, and Yaprak (2024) conceptual framework on AI hallucination marketing risks. The dataset enables empirical testing of whether crisis recovery trajectories vary systematically across SCCT clusters in the AI failure domain. Files: ahci_weekly.csv (long-format weekly observations); ahci_summary.csv (brand-country aggregates).