Cross-Engine Consensus in AI-Generated Brand Recommendations: An Empirical Study Across Seven Cities and Five Large Language Models

As consumers increasingly ask conversational AI systems for commercial recommendations ("which accounting software should I use," "which law firm handles employment disputes"), a brand's visibility in these systems has become a measurable and commercially consequential property, distinct from classical search engine ranking. This paper reports an empirical, multi-city study of cross-engine consensus: the degree to which independently queried, commercially deployed AI assistants converge on the same named brand or provider in response to the same buyer-intent question. We collected 278 production API responses across 56 buyer-intent prompts spanning seven cities (London, Berlin, Madrid, New York, Paris, Rome, and Dublin) and five widely used AI systems (OpenAI ChatGPT, Google Gemini, Anthropic Claude, Perplexity, and Meta AI), using each vendor's own production API rather than simulated or scraped output. One engine (ChatGPT) is excluded from analysis after an account-level API quota failure affected all 56 of its scheduled calls; the remaining 222 responses (four engines) form the analytic dataset. We find that categories with a small, well-documented, easily verifiable set of providers (project management software, business banking applications, large corporate law firms) converge strongly, frequently to unanimous agreement, while categories populated by many similarly qualified, less differentiated local or individual providers (employment law, independent financial advice) rarely converge at all. Two further findings extend this picture. First, query language can change not merely the ranking but the entire considered set of competitors: a wealth-management query in Paris returns boutique French firms when asked in French and major international private banks when asked in English, with almost no overlap between the two sets. Second, one city (Rome) shows zero categories reaching even three-engine agreement, the only such case in the dataset, while simultaneously showing the strongest single-engine, cross-lingual stability observed anywhere in the study (one engine reproduced an identical, identically ordered five-item ranking in both Italian and English). We report these findings with their accompanying data-quality caveats in full, and discuss their implications for brand visibility measurement and for the emerging practice of Generative Engine Optimization.