ChatGPT claws back web traffic to 55.5 % — but the yearly trend still favors Gemini and Claude

The latest Similarweb figures, reported by The Decoder, show ChatGPT recovering from 52.7 percent to 55.5 percent of AI-chatbot website traffic in three months, while Google Gemini fell from 27.8 percent to 25.6 percent. On the surface, that reads as a clean comeback: OpenAI's chatbot clawed back its lead, and Gemini's growth streak is over.

The more interesting story is underneath — and it is about what this metric does not capture.

Six competitors, one table

We compiled the available figures into a single comparison. Note that "3 months ago" and "1 year ago" values are only listed where the source data provides them; the rest were not part of the published round-up.

All figures come from Similarweb and cover desktop and mobile website traffic only. Native mobile apps, desktop applications like ChatGPT Work, Android system integrations and API calls are excluded.

The comeback is real — but narrow

Within a single quarter, ChatGPT gained 2.8 percentage points and Gemini lost 2.2 percentage points. That is a genuine reversal of momentum, especially after Gemini's share roughly doubled over the past year. The question is whether it reflects product superiority or simply the news cycle: OpenAI shipped a new flagship model this week, Anthropic refreshed its Claude line at the start of September, and Google's flash-tier model landed in between — wave after wave of releases makes it hard to isolate any single cause.

One calculation worth making: Anthropic's Claude, at 9.3 percent, now has more web traffic than DeepSeek, Grok, Microsoft Copilot and Perplexity combined (3.4 + 2.4 + 1.6 + 0.9 = 8.3 percent). A year ago, Claude was at 1.9 percent — that is a nearly fivefold increase, the strongest relative growth in the entire ranking.

The yearly view is less comfortable for OpenAI

Zoom out twelve months and the headline changes. ChatGPT is down from 73.3 percent to 55.5 percent — a loss of almost 18 points. Gemini roughly doubled its share. Claude quintupled its traffic base. In other words: ChatGPT is winning the quarter but still losing the year, and the competitive structure of the market has permanently shifted from "one dominant default" to "a leader plus two serious alternatives."

What website traffic does not tell you

This is where the technical caveats matter. The ChatGPT mobile app and the deep Gemini integration into Android are arguably the two largest AI surfaces in the world, and neither appears in this data. A user who asks Gemini something from the Android home screen, or chats with ChatGPT in the standalone app, is invisible to Similarweb. So when Gemini's share "fades," it may simply mean that more of its usage moved to channels this metric cannot see.

For the same reason, traffic share is a poor proxy for the thing most of our readers care about: which model actually performs in production. That is a question we measure directly in our AI Arena benchmark rig — tokens per second, time to first token, VRAM usage and real cost per token. Web clicks measure brand heat, not capability.

The European angle: the AI Act is now in force

For European users, this traffic snapshot has a particular context. On August 2, 2026, the EU AI Act moved into its general application phase, and the voluntary era ended. Chatbot providers are now legally required to make machine-readable watermarks on synthetic content and to explicitly disclose when a user is interacting with an AI system. The European Commission and the EU AI Office simultaneously gained active enforcement powers, including formal evaluations and model recall orders for general-purpose AI providers.

In practice, every major player in this ranking — ChatGPT, Gemini, Claude, DeepSeek, Grok — now operates under the same binding transparency obligations when serving EU users. The three-month window covered by this Similarweb data therefore coincides almost exactly with the first weeks of full AI Act enforcement. That matters for European businesses: if you integrate any of these models into a customer-facing product, watermarking and AI-disclosure obligations are now your responsibility, not a voluntary badge.

There is also an uncomfortable regional observation in the table: no European provider appears in the top tier. Mistral, the EU's most visible AI champion, focuses on API and enterprise products rather than a mass-market consumer chatbot, so it does not register in consumer web-traffic rankings at all. European users vote with their clicks, and right now they click American.

Price per token matters more than share points

If you are choosing a model for a European product, the spread in announced API pricing dwarfs any traffic-share movement. At an approximate rate of 0.90 USD to the euro:

• Gemini 3.8 Flash (released September 2): $0.75 input / $3.75 output per 1M tokens — roughly €0.68 / €3.40, an introductory rate valid through December 31, 2026.
• Claude Fable 5.1 (September 1): $10 input / $50 output per 1M tokens — roughly €9 / €45.
• GPT-6 Astra (September 3): $4–$10 input / $20–$50 output per 1M tokens — roughly €3.60–€9 / €18–€45.
• DeepSeek-V4-Pro (GA since August 13): from $0.435 input and $0.87 output per 1M tokens — under €0.40 / €0.80 at the entry tier.

The difference between the cheapest and most expensive flagship output token is more than a factor of 50 — a gap no traffic-share chart will ever show you. For an EU developer running real workloads, that number, not a Similarweb percentage, decides the monthly invoice.

The September release wave — Claude Fable 5.1, Gemini 3.8 Flash and GPT-6 Astra all landed within days of each other — will only show up in future traffic data. For now, the honest summary is this: ChatGPT remains the default consumer brand, Gemini is the strongest long-term challenger, Claude is the fastest riser, and website traffic is the least reliable way to measure all three.