ChatGPT, Gemini, and Claude: How Generative AI Web Traffic Is Redistributing Across Three Distinct Moats

Similarweb panel data through June 2026 shows the consumer web front shifting from a near-monopoly to a three-player market — and the two challengers got there by entirely different means.

What Happened

Similarweb’s worldwide panel data, tracking generative-AI website visits through June 2026, shows the clearest picture yet of a de-concentrating market. ChatGPT’s share of consumer-facing gen-AI web traffic has fallen from roughly 76% in June 2025 to approximately 52–53% a year later — a near-monopoly position cut nearly in half in twelve months. Google’s Gemini is the single largest gainer, rising from about 9% to roughly 27% over the same window. Anthropic’s Claude is the other notable mover, reaching approximately 9.2% in June 2026, up from just ~2.2% six months earlier. The remaining field — DeepSeek at ~3.6%, xAI’s Grok at ~2.5%, Microsoft Copilot at ~2.0%, and Perplexity at ~1.1% — accounts for a combined slice well under 10%.

Before drawing strategy conclusions, the metric warrants precision. This is Similarweb panel data: proprietary, worldwide, all-device traffic estimates that are not audited by Comscore, Nielsen, or any third party. It measures visits to consumer-facing web properties — not API consumption, enterprise seat count, or revenue. Two implications follow immediately. First, ChatGPT’s falling share in a category that is itself growing does not mean ChatGPT’s absolute traffic is shrinking; a smaller slice of a rapidly expanding pie can still represent more visits. Second, this chart captures only the consumer web surface, leaving OpenAI’s API platform, its mobile apps, its newer agentic products, and its hardware initiatives largely unmeasured.

The honest, specific reading: the consumer web front for generative AI is de-concentrating, and two challengers are taking share — by routes that have almost nothing in common with each other.

The Structural Read

For roughly two years, “ChatGPT” functioned as a category name — the way “Google” stands in for search. A ~52% share with the next two players filling most of the remaining space marks the end of that reflex and the beginning of a genuinely competitive consumer market. The map of AI on the consumer web has been redrawn from a monopoly to a three-player structure, with a long tail of smaller properties accounting for less than one-eighth of all visits.

The more instructive question is how the share moved, because the two gainers illustrate the two cleanest moats available in this market right now.

Gemini’s climb from ~9% to ~27% is almost entirely a distribution moat. Google has surfaced Gemini across Search, Android, Chrome, and Workspace — surfaces where billions of users already spend time, requiring no new acquisition behavior. The user does not need to decide to try Gemini; Gemini is already where the user is. That is the one structural advantage OpenAI cannot easily replicate at consumer scale on the web. The recent Gemini 2.5 Pro launch — with a 2-million-token context window and Deep Think mode — shows Google layering capability on top of distribution, which compounds the advantage.

Claude’s climb from ~2.2% to ~9.2% in roughly six months is a capability-and-reputation moat. Anthropic did not distribute Claude through a pre-existing surface; it earned usage by being the model developers and power users actively prefer for coding, agentic tasks, and — increasingly — a distinct model character that resists sycophancy in ways users notice and talk about. The character and sycophancy measurement work Anthropic has published is not incidental to Claude’s growth; it is part of what makes Claude legible as a different kind of model in a crowded field.

Meanwhile, OpenAI’s visible strategic energy is going precisely where this chart cannot see: apps, agentic products like Codex, and hardware — a screen-free speaker being the most visible signal of a bet that the next major distribution endpoint is not the browser tab at all. That is both a rational hedge against exactly the de-concentration this data shows and a forward-looking claim that the chatbot webpage is not the final form of consumer AI. OpenAI is not defending the web surface it is ceding; it is building around it.

Three Implications