ChatGPT regains ground, but Gemini and Claude close the gap

ChatGPT held 73.3 percent of AI chatbot web traffic a year ago. Today, per The Decoder citing Similarweb data, it holds 55.5 percent. That 17.8-point decline in twelve months is not a rounding error; it is a structural shift in where people go to ask questions, and by extension, where brands need to be cited to be found.

The headline number flatters OpenAI slightly. ChatGPT clawed back share it had lost to Gemini earlier in the year, and 55.5 percent of a rapidly growing market is still an enormous pool of queries. But the direction of travel is the story. Gemini doubled its traffic share over the same period. Claude grew nearly fivefold. Both figures come with an important caveat: Similarweb measures website visits only, excluding mobile apps and desktop clients. Given that ChatGPT's own app has tens of millions of active users, and that Gemini is baked into Android at the operating-system level, the web-traffic numbers almost certainly overstate OpenAI's relative dominance and understate Google's.

What a fragmenting market means for citation strategy

For B2B brands whose authority depends on being surfaced in LLM answers, a world with one dominant model was actually simpler. A single retrieval architecture, a single set of training data preferences, one set of behaviours to understand. The market that Similarweb's numbers describe is messier. Claude, Gemini, and ChatGPT each have meaningfully different approaches to source selection, citation formatting, and the types of content they privilege in responses.

Claude, developed by Anthropic, leans heavily on long-form, well-structured content from sources with strong domain authority. Gemini pulls from Google's index and tends to favour content that already ranks well in traditional search. ChatGPT's browsing and search integrations have their own retrieval logic. Optimising for one is not the same as optimising for all three, and as Claude's fivefold traffic growth suggests, the cost of ignoring any one of them is rising.

For institutions in financial services, multilateral bodies, and major industrial groups, the stakes are specific. These organisations often operate in high-stakes query categories: regulatory guidance, procurement decisions, policy interpretation, technical standards. A senior official at a UN agency or a procurement officer at a large industrial group is now as likely to start with an AI query as with a Google search. If an institution's content is structured for ChatGPT's retrieval preferences but invisible to Gemini's index or Claude's training signal, it loses ground in exactly the moments that matter.

The fivefold growth in Claude's traffic share deserves particular attention from communications teams in policy and financial services. Anthropic has positioned Claude explicitly as a tool for professional and institutional use, and its enterprise adoption is accelerating. A policy brief, a technical standard, or a regulatory explainer that wins a citation in Claude is reaching a different, arguably more decision-proximate, audience than the same citation in a general-purpose consumer product.