Absurd Prompt Compliance and Alignment Drift in a Politicized Search Interface

This dataset documents a series of image-based interaction artifacts capturing the behavior of a Perplexity-powered search system (“Truth Search AI”) deployed within the Truth Social platform. The artifacts record system responses to deliberately absurd, category-confused, poetic, and reverent prompts that are non-violent and non-adversarial. Observed behaviors include over-explanation, selective moral correction, citation inflation, persona leakage, and participatory narrative adoption under conditions of low epistemic demand. Rather than refusing, minimizing, or reframing nonsensical or non-interrogative inputs, the system frequently responds with authoritative or mythic framing, indicating that conversational politeness, social resonance, and interface ideology can override epistemic discipline. The dataset is intended for qualitative analysis of alignment behavior in real-world, consumer-facing deployments and includes a companion methods-only negative results note documenting the absence of expected refusal or grounding mechanisms.