Is it enough to be cited by ChatGPT?
Published on · updated on
No, for two reasons. First, the audience: according to Similarweb's monthly report, ChatGPT accounted for 55.5% of global web traffic for generative AI sites in August 2026, compared to 25.6% for Gemini and 9.3% for Claude—meaning one in three visits takes place somewhere other than ChatGPT, whereas it captured nearly three-quarters a year earlier. Second, the sources: in our measurements, the three models do not read the same web: across our five combined corpora, only 1.8% of the cited domains were cited by all three at the same time. Being cited by one says almost nothing about what the other two will do.
For what type of question? It all depends on your clients. If you know that your market only uses a single assistant, the question is framed differently. In the absence of this information, targeting a single AI is like betting on an audience share that has changed by half in a year.
What each AI represents, and how fast it is moving
Similarweb measures visits to generative AI sites worldwide. In its report published on July 29, 2026, ChatGPT's share went from around 76% in June 2025 to around 53% in May 2026, while the category as a whole grew by 70%: ChatGPT is not losing visits, it is growing slower than the market. Its monthly report from September 7, 2026, gives August figures of 55.5% for ChatGPT, 25.6% for Gemini, and 9.3% for Claude.
These figures measure website visits, not usage: they ignore mobile applications, assistants integrated into an operating system, and API calls. They provide an order of magnitude and a trend, not the exact distribution of questions asked by your clients.
The three do not read the same web
Audience is only half the problem. In our measurements, each model searches and chooses in its own way: ChatGPT mainly cites pages placed at the top of its own results and only retains about one in ten; Claude reads pages in full and cites nearly half of them, favoring those that multiple searches bring back; Gemini distributes its response across many sources. And, across our five combined corpora, only 1.8% of the cited domains were cited by all three at the same time.
In other words, a presence built for a single model does not carry over. What does carry over are the common traits: a precise question, an answer right from the start, a detailed text, and a name that several independent sources use.
What we do not know
Our measurements focus on the API of GPT-5, Claude Sonnet 4.6, and Gemini 3.5 Flash, not on the consumer interfaces, which may behave differently. They describe what models search for and cite at a given moment; they do not measure the effect of a publication on citations, and an AI changes its response from one time to another. The traffic shares are those of a web measurement panel; we have not reproduced them ourselves.
Sources
- Similarweb — "AI Search Stats", global web traffic shares of generative AI sites, .
- Similarweb — monthly report of AI traffic shares, August 2026 (ChatGPT 55.5%, Gemini 25.6%, Claude 9.3%), .
- FaireDuBruit measurements, from September 4 to 16, 2026 — see "How we measured".