Research
What we measured, with the data
Original measurements on how AI assistants recommend brands. Each study states its method, its limits and publishes the raw answers, so the numbers can be recounted and cited.
Study, 2026-08-23
How repeatable are AI recommendations? 4,500 answers from three model APIs
The same buyer question asked 5 times to the same model returned the same set of brands in 15.0 percent of cases (mean Jaccard 0.54); the first-named brand held in 75 percent of runs; half of all brands named (49.8 percent) appeared only once. Dataset and raw answers downloadable, CC BY 4.0.
Read itStudy, 2026-08-23
Which brands do ChatGPT, Gemini and Claude agree on? Cross-model overlap across 4,500 answers
Of 1,511 brand and category pairs named by at least one of the three models (in 2 or more of its 50 answers), 148 (9.8 percent) were named by all three, 181 (12.0 percent) by two and 1,182 (78.2 percent) by one model only. The three agreed on the most frequent first-named brand in 37 of 300 buyer questions (12.3 percent). 41 brands in 15 of 30 categories were core to all three. Computed from the same raw answers; per-brand table downloadable, CC BY 4.0.
Read itEarlier data note, 2026-08-20
FAQ page or H2 sections? What AI models actually cite
438 URLs cited by ChatGPT and Gemini in 60 days of live buyer answers, classified by page type. None was a FAQ page; the models pointed at home and product pages.
Read it