Cross-Engine Citation Divergence: A Cold-Tested Benchmark of Entity Recognition Across Six AI Answer Engines

As AI answer engines increasingly replace ranked lists of links with synthesized responses, whether a given entity is named —cited within those responses becomes commercially decisive, yet its measurement remains immature. This paper introduces a cold-testing protocol for measuring entity recognition across six major AI answer engines (OpenAI ChatGPT, Perplexity, Google AI Overviews, Google Gemini, Anthropic Claude, and Microsoft Copilot) and applies it to a single real-world professional entity across five query classes. Key findings:1. A universal terminology-collision failure: a distinctive brand term resolved to an established generic domain (geospatial decision-support systems) on all six engines.2. A corroboration gradient: engines rank-order by the amount of independent, cross-source corroboration each requires before it will name an entity, from single-source-sufficient to multi-source-required.3. A priming effect: prior mention of an entity earlier in a conversational session materially inflates apparent recognition on later queries, contaminating naive (non-cold) measurement. The study introduces session-isolated, anonymous, repetition-controlled cold testing as a necessary method for measuring AI-mediated visibility, and discusses implications for measurement practice and for small and medium-sized enterprises. Measurement was operationalized using the GEO Decision System, a nine-layer framework for generative engine optimization.