Brands ChatGPT rarely names without search stay rare even when cited pages mention them
Brands that ChatGPT seldom names when answering without search results tend to stay seldom named when it does search, even when those brands appear on many of the pages it cites. That is the main finding of an October 16 analysis by Graphite, a growth agency that sells AEO services, comparing logged-out ChatGPT answers to comparison prompts with and without search.
Among brands substantially mentioned on at least 15% of cited pages, median visibility with search was 11% when ChatGPT named them in fewer than a quarter of its no-search answers. Visibility with search rose alongside both visibility without search and the share of cited pages mentioning the brand, which Graphite reads as both factors mattering. The 57-question comparison drew on 3,115 entity-question pairs collected in August, and Graphite says two replications, one of 100 prompts through the API with search explicitly enabled and disabled and one of 200 prompts tracked over a different three-week window, showed the same relationship at comparable strength.
Graphite argues brands should work on what models absorb during training as well as on citations, calling that parametric knowledge optimisation. It also counted brand names in about 957,000 ChatGPT fan-out queries drawn from 22,953 non-branded comparison prompts collected between August and early October: 59.6% contained a brand, as labelled by Anthropic's Claude Haiku 4.5, and 22.7% used a site: operator. None of the prompts named a brand, and Graphite says a brand appearing in a fan-out query raises the chance that pages featuring it get retrieved.
The methods are exploratory. Graphite suggests checking brand visibility through the API with search turned off when a new model ships, collecting at least 50 responses per prompt, and letting training crawlers reach a site, including checking CDN and firewall restrictions. It puts the minimum time-to-impact for newly published content at roughly two to five months between a model's knowledge cutoff and its release, though that does not measure when a page is crawled into the training set or whether answers change.
Reported by Search Engine Journal ↗
Topics: ChatGPT, Citations, Measurement