In Graphite’s analysis of ChatGPT comparison prompts, brands the model named more often without search also tended to appear more often when it searched. The finding suggests that citations alone may not account for which brands make it into an answer, but it does not show that changing a brand’s web presence will change what ChatGPT says.
Watch Desk analysis
What happened
Search Engine Journal reports that Graphite analysed 3,115 brand-and-question pairs across 57 questions, using logged-out ChatGPT data collected in August. Brands mentioned in fewer than a quarter of the model’s no-search answers had a median visibility of 11% in search-enabled answers, even when they appeared substantially on at least 15% of the pages ChatGPT cited.
Graphite says the relationship also appeared in two replications: one using 100 prompts through the API with search explicitly enabled and disabled, and another tracking 200 prompts over a separate three-week period. The account does not explain how the main 57-question comparison separated answers that did not use search; Graphite specifically describes only the 100-prompt replication as explicitly switching search on and off.
Why it matters
A search-enabled answer is not simply a list of whatever appears on the pages it cites. In this dataset, brands that ChatGPT named more often without search also tended to fare better with search. That makes a citation count an incomplete picture of how often a brand appears in an answer.
The study shows an association, not that more citations or stronger training-data visibility cause more mentions. Graphite is a growth agency that sells answer-engine-optimisation services, and its recommendations about how brands should improve visibility are its interpretation of the data, not a result established by the comparison.
Our read
The useful takeaway is that cited pages may not tell the whole story of why a brand appears in an AI answer. The important qualification is that this analysis does not establish a reliable recipe for changing that outcome. Graphite has a commercial interest in the subject, and the main comparison’s method for identifying no-search answers is not fully explained in the account. Interesting signal, not a new marketing law carved into silicon.
What to watch
- Whether Graphite publishes fuller methods for separating search and no-search answers in the main comparison.
- Whether independent researchers reproduce the association with other prompts and models.
- Whether further testing can distinguish the contribution of model familiarity from the content of cited pages.
Discussion spark: If ChatGPT names a brand more often when it answers without search, what evidence would convince you that improving its web visibility can change its search-enabled answers?
Sources and evidence
- Source update (26 October 2026, 12:24 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.