Gemini slipped for two months. Google is playing a bigger game.
ChatGPT pulled back ahead in chatbot web traffic. But the next contest Google is preparing for is not better answers. It is an agent platform that ties Search, Android, Workspace, Photos and Cloud together and actually finishes the job.
Is Gemini really losing?
Reporting in September, citing Similarweb data, put ChatGPT at 55.5% of generative AI chatbot web traffic in August 2026, Gemini at 25.6% and Claude at 9.3%. ChatGPT rebounded from 52.7% in May. Gemini went the other way, from 27.8% to 26.5% to 25.6%, down two months in a row.
In the short term, yes, Gemini's growth has stalled. Stretch the frame out and the story changes. A year ago ChatGPT held roughly 73% and Gemini roughly 13%. Gemini has nearly doubled its web share in twelve months. This looks less like a Google failure than a stretch where the company has to prove it can reaccelerate after a fast spring.
The real battlefield is off the website
Web traffic matters. It shows which AI a user deliberately goes to visit. But calling up Gemini on Android, summarising a thread in Gmail, fixing a document in Docs or using the AI features inside Search may never register as a visit to the Gemini website.
ChatGPT's advantage is still obvious. It has claimed the act of asking an AI as a brand. Google's position favours a different play: rather than turning Gemini into one more destination, it can push AI into the services where users already sit.
| ChatGPT | Gemini | |
|---|---|---|
| Main entry point | Standalone general-purpose AI interface | AI embedded across Google services |
| Strength | Brand, generality, habit | Search, Android, Workspace, Cloud, Photos |
| Next contest | A neutral agent that works across many tools | An agent that executes end to end inside Google's own services |
Every new Google card points at agents
Gemini 3.8 Flash, released on 2 September, leads on long-horizon software engineering, autonomous agents and complex enterprise workflows. The input limit is 1,048,576 tokens, maximum output is 65,536 tokens, and reasoning effort can be set to low, medium or high. Launch pricing is $0.75 per million input tokens and $3.75 per million output tokens.
Those numbers are not simply about accepting longer questions. They are the foundation for reading large codebases and document bundles, calling multiple tools, checking intermediate results and carrying a long task through to the end.
Gemini 3.8 Flash Cyber and the Fairwind Program, announced alongside it, make the direction plainer. Google tuned Flash Cyber for vulnerability detection and automated patching, and paired it with CodeMender to cover security work from discovery through verification to fix. Google says more than 650 partners are taking part in Fairwind.
The unit of competition here is not a single model score. What enterprises actually pay for is the model plus the operational apparatus around it: permissions, logs, security, verification, compliance and rollback procedures.
Why Spark inside Photos matters more
For ordinary users, the clearer signal is Gemini Spark's integration with Google Photos. Rolling out to eligible Google AI Pro and Ultra subscribers in the United States, it handles photo search, duplicate triage, editing, album creation, sharing, text extraction from images, and follow-on actions in other Google apps.
Edits save as copies rather than overwriting originals. New albums default to private. Anything that leaves the user's own space, such as sharing or sending by email, requires user approval. That design matters because trust in the agent era does not end at whether the answer is correct.
The moment an AI can reach your photos, mail, documents and calendar, the competitive ground shifts to product design: approval before execution, scope of permissions, change history, undo. The stronger the data connectivity, the bigger Google's edge, and the bigger its exposure when an agent acts wrongly.
What Alphabet investors should watch instead of web share
Gemini's 25.6% of web traffic is a real number. The larger question for Alphabet's valuation is whether Gemini can defend and extend Google's existing revenue engines.
In Search, AI has to protect queries and ad revenue. In Cloud, it has to generate model API, agent and security revenue. In Workspace, it has to lift revenue per seat. In Android and Chrome, Google has to own the default point where users summon an AI.
The trickiest variable is cost. An agent can run far more inference and tool calls than a chatbot that answers once and stops. So Google pitching 3.8 Flash as a workhorse is not only a performance boast. It is tied to unit economics. As usage grows, ad, subscription and cloud revenue has to grow faster than inference cost.
OfficeChai — Similarweb August traffic data
Google AI for Developers — Gemini 3.8 Flash specifications
Google — Gemini 3.8 Flash and Flash Cyber
TechCrunch — Gemini Spark for Google Photos
9to5Google — Spark Google Photos controls
Insight Times Editorial Desk
