Signal The Language of the Question Selects the Market: Query Language and Exit IP as Separable Factors in Commercial Recommendations from a Generative Search Interface

Summary

When a generative search interface answers a commercial question, which market's products it names is decided before the model reasons about the products. Researchers ran a controlled probe of 234 executions against the logged-out ChatGPT web interface and the OpenAI API, collected on 29-30 August 2026 across four exit countries and six query languages, with six identical runs per cell. First, the top recommendation was unstable: it changed across six identical runs on four of six prompts, at an identical rate in the browser interface and in the API with web search both enabled and disabled, confirming instability as a property of the system itself. Second, query language rather than location decided whether local suppliers appeared at all: when the query language matched the country, a global brand won only 1 of 24 runs, while asking in English on the same connections meant local brands took 0 of 6 runs in Estonia and Turkiye. Third, language and location act separately: holding the query language fixed while moving only the exit IP shifted the named market while the answer stayed in the query language, as shown for Turkish asked from Berlin and Russian asked from Tallinn. A minority language occupied a middle tier -- Russian asked from Estonia named an Estonian supplier in 4 of 6 runs and a global one in all six, while Estonian named a local supplier every time and English named none. A negative control in a second category showed no language effect at all, and the authors interpret this as pointing to whether a category is nationally regulated rather than nationally supplied.