Category-Conditioned Web Exposure Tracks Brand Presence in ChatGPT Fanout Queries

Search-enabled language models like ChatGPT can insert brand names into web-search queries before retrieval occurs, substantially influencing which brands are later surfaced. This study investigates why some brands enter ChatGPT's first fanout query while other well-known brands do not. Using gpt-5.6-luna, live Responses API web-search queries, a controlled fanout-query simulator, and Common Crawl measurements across multiple software and product categories, the study separates three concepts: brand accessibility, candidate ranking, and invitation into the fanout search query. The results show that ChatGPT can clearly know a brand while still leaving it out of its first fanout query. When the model is allowed to include more brand names, many of those omitted brands reappear, suggesting that the first query reflects a limited ranking of candidates rather than simply what the model knows. Brands with stronger category-specific presence across the web also tend to rank higher in this selection process, a pattern that was reproduced in a held-out e-signature category.