Graphite says a ChatGPT citation does not guarantee a brand mention: Reveneau

Why it mattersA brand can win the citation and still lose the answer. Getting on more pages matters less than being a brand the model already knows.

Getting your page cited by ChatGPT is not the same as getting your brand named by it. A new study from Graphite, the growth agency, looks at how often ChatGPT names the brands whose pages it cites. The paper went live on 16 October.

What the study looked at

Graphite worked from 22,953 non-branded entity-comparison prompts. Those prompts produced about 957,000 fan-out queries. The data runs from August to early October 2026. The ChatGPT used was the logged-out one.

Graphite says citations are more common now. In the week of 3 August, 56.6 percent of answers carried a citation. By the week of 28 September, that number was 88.5 percent. The jump is in Graphite's own tracking only.

Across all prompt types, Graphite cites a Similarweb figure. Only 7 percent of United States ChatGPT answers carry any citation at all. Graphite reads this as a sign that ChatGPT searches when a question invites a comparison. On other questions it answers from memory.

The main finding

The headline number sits in the middle of the paper. Graphite looked at brands that showed up on 15 percent or more of the pages ChatGPT cites. For those brands, the median mention rate when ChatGPT searched was still only 11 percent. The condition: the brand was named in less than 25 percent of answers when the model did not search.

In plainer words. Brands the model has already learned about get named. Brands that only sit on cited pages often do not.

What Graphite calls it

Graphite calls the knowledge that lives in model weights "parametric knowledge". The term is not new in research. The paper is one of the first commercial write-ups to say it should drive brand work for AI search. The three named authors are Gregory Druck, chief AI officer, Ethan Smith, the CEO, and growth advisor Bevin Benson.

Graphite's advice at the end of the paper is short. A brand should be mentioned on many different pages and many different sites. The description of what the brand does should be the same across all of them. Winning one strong citation is not enough.

What a search team can take from this

A reader should keep the limits in mind. Logged-out ChatGPT is one surface. Entity-comparison prompts are one kind of question. Graphite sells services in this space. Matt G. Southern's write-up at Search Engine Journal reaches the paper the same way.

The practical read is simple. A plan built only around getting cited is a plan to be a source. The brand still has to be named to be chosen. Teams tracking AI visibility should log two numbers side by side. How often does the model cite the brand's page. How often does the model name the brand in the sentence. The gap between the two is where the next quarter of work goes.

Source

The Graphite research page, Parametric Knowledge Optimization, by Gregory Druck, Ethan Smith and Bevin Benson.
Reporting by Matt G. Southern at Search Engine Journal, Some Brands On ChatGPT's Cited Pages Still Got Few Mentions.

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.