Method. 150 shopping questions per category, each asked three times to two grounded AI assistants — ChatGPT (GPT-5.6, web search) and Gemini (google_search grounding) — for 960 answers per category. Every brand named and source cited was extracted and classified. This audits how a brand appears when a shopper asks an AI what to buy.
Notes. Visibility = share of the 960 answers naming the brand. Gemini exposes far fewer resolved citation URLs than ChatGPT, so its source-mix figures rest on a smaller base. All percentages are within-category. Data is cached; nothing is re-run on load.