ChatGPT’s dominant share of generative AI chatbot web traffic slid from about 87% in January 2025 to roughly 64.5% in January 2026, according to data from Kavout. That’s a drop of more than 19 percentage points in just twelve months. But OpenAI isn’t losing users. The company’s service now draws over 900 million weekly active users, a massive base that continues to grow. The story isn’t a collapse—it’s the rapid expansion of a market where more people are using more AI tools for different jobs.

The Numbers Behind the Shift

Kavout’s 2026 analysis paints a detailed picture of an AI chatbot market in flux. While ChatGPT remains the single largest destination by a wide margin, competitors are catching up fast. Google Gemini’s share surged from 5.4% to somewhere between 18.2% and 21.5%, a 370% year-over-year increase. Elon Musk’s Grok went from near-zero to commanding 15.2% of U.S. mobile daily active users. Anthropic’s Claude, Perplexity, and a growing roster of specialized tools also carved out niches.

The percentage-point decline reflects a market that is ballooning in size. The global AI chatbot sector was valued at an estimated $10.32 billion to $11.45 billion in 2026, with projections to hit $32.45 billion by 2031. More users are trying multiple assistants, often picking the best tool for a specific task rather than sticking with one generalist. Crucially, these web traffic metrics don’t capture all AI use. Enterprise deployments, in-app assistants inside Microsoft 365 or Google Workspace, local models, and API calls never show up in such charts.

What This Means for You

If you open a browser tab every time you need AI help, you’re already living in a multi-model world. But the shift is even more relevant to anyone whose work depends on Windows, a Microsoft 365 subscription, or a fleet of managed devices.

Home and Power Users

The old habit of firing up ChatGPT for every query is giving way to a toolbox approach. You might still use ChatGPT for open-ended writing, image generation, or quick answers. But Google Gemini may be more convenient when you’re inside Gmail, Docs, or Search. Anthropic’s Claude often excels at digesting long reports or complex codebases. Perplexity surfaces sources you can verify. Grok can catch what’s happening right now on social platforms. The practical lesson: experiment with different tools and save the ones that fit recurring tasks. You don’t have to abandon ChatGPT; just stop treating it as the only option.

IT Administrators and Enterprise Teams

For organizations, the diversification is already happening at speed. Enterprise research cited in the Kavout report shows 81% of companies now test or deploy three or more model families. That’s up from 68% less than a year earlier. The reasons are straightforward: a model that’s brilliant at code generation may not be the right choice for sensitive HR documents. A cloud chatbot comfortable with public content might violate data-residency rules when fed customer records.

This means IT must plan for a legitimate, manageable multi-model strategy. Banning everything except one approved bot rarely works. Instead, provide sanctioned pathways—Microsoft Copilot for Microsoft 365 tasks, an approved chatbot for general use, a local model option for confidential data. Governance is the critical piece: classify data, define which tools can process which information, enforce identity and access controls, and audit use.

Developers

Your workflow already spans multiple models. GitHub Copilot is embedded in your IDE. You might call OpenAI’s API for prototyping, test Anthropic’s Claude on longer context, and run DeepSeek open-weight models locally when privacy or cost matter most. The key is to treat models as interchangeable components. Use the cheapest one that meets quality thresholds for the current job. Monitor API costs, latency, and output consistency. And never forget that any prompt sent to a public cloud service may be stored or used for retraining unless the provider explicitly contracts otherwise.

How We Got Here

The timeline from early 2025 to now tells a story of rapid commoditization and deliberate platform plays. When ChatGPT exploded, it had no serious competitor. Google’s Bard was an experiment. Bing Chat was nascent. Claude was powerful but less accessible. That changed fast.

By mid-2025, Google had woven Gemini into Search, Workspace, Android, and Chrome. Microsoft embedded Copilot across Windows 11, Edge, Microsoft 365, and GitHub. Anthropic opened up Claude to larger audiences. xAI positioned Grok as the bot that understands X. Perplexity made research-first AI its identity. Simultaneously, DeepSeek proved that frontier-quality models could be trained with dramatically less compute—challenging the assumption that only giants with unlimited budgets could compete.

Now, in early 2026, distribution and integration often matter more than benchmark scores. Gemini’s growth didn’t come from topping every leaderboard; it came from appearing where users already work. Copilot’s slower consumer pickup—hovering around 1.2% in U.S. mobile share—doesn’t reflect its enterprise footprint, which is deeply tied to Microsoft 365 licenses and Azure infrastructure. And Grok’s rise shows that a distinct identity attached to a live social platform can pull users away from generic assistants.

The market is maturing, and that means no single tool can be everything to everyone.

What to Do Now

  • Take inventory. List the AI tools you actually use, note what you use each one for, and flag any sensitive data you’re pasting into public chat windows.
  • Match the tool to the task. Start thinking in terms of strengths: ChatGPT for creative breadth, Claude for document reasoning, Gemini for Google ecosystem work, Copilot for Microsoft 365 integration, Perplexity for sourced research, Grok for real-time chatter, and local models for confidential work.
  • Set governance rules. For admins: publish a clear policy stating which data can enter which services. Ensure employees know that free consumer chatbots may retain prompts. Provide approved alternatives for protected data, such as enterprise-grade Copilot with proper data handling or local inference.
  • Evaluate local AI. Windows PCs with modern NPUs or capable GPUs can now run compact models like Phi, Llama, or DeepSeek variants. Use local AI when you need privacy, offline capability, or cost predictability. Just be realistic about performance limits and hardware demands.
  • Monitor costs. Multi-model can mean subscription sprawl. Track what your team is paying for—consumer subscriptions, API credits, enterprise seats—and consolidate where practical.
  • Stay flexible. The landscape will keep shifting. Avoid long-term exclusive commitments to any one model unless the integration, security, and pricing are demonstrably worth it.

Outlook

The AI chatbot market isn’t consolidating; it’s fracturing into specialized layers. The next phase will be defined less by raw model quality and more by how deeply a service is embedded into the tools you already use. Expect operating system vendors to push their own assistants harder. Local inference will improve on mainstream hardware. And regulators will likely scrutinize data flows, especially where foreign servers are involved. The wise move now is to build fluency across multiple platforms—because the era of the one true chatbot is over.