Even as Google is struggling with its AI models amidst the delay of Gemini 3.5 Pro, it’s simultaneously seems to be losing momentum in the consumer-facing AI space.
New data from Similarweb tracking generative AI website traffic shows Gemini’s share slipping for the first time in over a year, a reversal that comes right as Anthropic’s Claude continues adding share at a pace few would have predicted twelve months ago.
The numbers tell the story cleanly. A year ago, in July 2025, ChatGPT commanded 77.5% of generative AI traffic, with Gemini a distant second at 9.4%. Six months ago, that gap had already narrowed sharply — ChatGPT down to 63.8%, Gemini up to 22.8%. Three months ago, Gemini had climbed further still, to 27.0%, while ChatGPT continued bleeding share down to 54.2%. But in the most recent month tracked, Gemini actually dipped slightly to 26.8%, while ChatGPT held roughly flat at 54.8%. It is a small move in isolation, a couple tenths of a percentage point, but it breaks a run of essentially uninterrupted monthly gains that Gemini had put together since the middle of last year.
Gemini’s ascent over that stretch had been genuinely remarkable, and largely earned. The Nano Banana image model, released quietly last August before Google claimed it, turned into one of the biggest viral moments any AI product has had, pulling in 10 million new users within days and briefly pushing the Gemini app to the top of the App Store, ahead of ChatGPT. Google followed that up with Gemini 3 Pro, which closed much of the reasoning and coding gap with OpenAI’s models, and then Nano Banana Pro in November, which kept the image-generation momentum going with 4K output and Gemini 3 Pro’s reasoning baked in. Add Google’s ability to bundle Gemini into Android, Search, and Workspace by default, and the growth trajectory made sense — Similarweb had previously noted Gemini crossing 20% of traffic share for the first time in January, then 25% by March, a period during which its year-over-year traffic was growing multiple times faster than ChatGPT’s.
That momentum started cooling by June. Gemini’s month-on-month gain had shrunk to its smallest move of the entire run, with ChatGPT staying flat around 52-53% rather than continuing its decline. The latest data confirms that wasn’t a one-off blip — Gemini has now gone from gaining every single month to actually losing ground.
Part of the explanation sits squarely with Google’s product roadmap, or rather the absence of one lately. Sundar Pichai told a room full of developers at Google I/O in May that Gemini 3.5 Pro was on its way, promising it would land the following month. June came and went with nothing. Google instead spent July retraining the model on updated data after it reportedly underperformed on coding benchmarks internally, and when that retraining also fell short, the company shipped Gemini 3.6 Flash as a stopgap rather than the flagship everyone had been told to expect. Reviews of that release were unflattering — Gemini 3.6 Flash landed at the exact same score on the Artificial Analysis Intelligence Index as its predecessor, and a subsequent Semi Analysis report went as far as claiming that Google has quietly shelved 3.5 Pro altogether to redirect resources toward Gemini 4. Executives at OpenAI and Meta didn’t hold back either, publicly ribbing Google over the delay once Gemini 4’s pretraining run was announced without a corresponding ship date for 3.5 Pro.
Without a flagship release to point to, Gemini has essentially been running on the fumes of Nano Banana and Gemini 3 Pro for months, and consumer traffic numbers appear to reflect that. Meanwhile, the platform that has actually been converting new users at pace is Claude. Anthropic’s assistant sat at just 1.9% of traffic a year ago, barely a rounding error next to ChatGPT and Gemini. It crept to 2.2% by January, then nearly quadrupled to 8.0% by April, and has since climbed further to 9.7% — closing in on double digits from a standing start that most of the industry had written off as a distant third option. Some of that surge has come from genuinely unusual circumstances, including the publicity around Anthropic’s standoff with the US government over military use of its models, which briefly sent Claude to the top of the App Store ahead of both ChatGPT and Gemini. But the underlying trend predates that controversy — Claude’s app DAUs had already grown roughly 5x between October and February, and Anthropic’s broader business has been compounding just as fast, with revenue run rate reportedly crossing $44 billion this year.
The rest of the field remains largely static by comparison. DeepSeek has drifted between roughly 3-4% for over a year without breaking out. Grok has actually gone backwards, from 3.1% a year ago to 2.5% now, despite periodic bursts of attention around new releases. Copilot and Perplexity round out the pack in the low single digits, neither showing much sign of movement in either direction.
What this leaves is a market that, twelve months ago, looked like it was steadily consolidating into a two-horse race between ChatGPT and Gemini, and now looks considerably messier. ChatGPT has stopped shrinking, at least for now. Gemini has stopped growing, at least for now. And Claude is the only one of the three still moving with real velocity, adding close to two full percentage points in the last month alone. Whether that’s a temporary reshuffling while Google waits on Gemini 3.5 Pro or 4, or the start of Claude establishing itself as a durable third player in consumer AI the way it already has in coding and enterprise, is the question the next few months of data will have to answer.