For most of the generative-AI era, the assistant market had a simple shape: there was ChatGPT, and then there was everyone else. That shorthand is now out of date. According to Sensor Tower’s 2026 State of AI report, ChatGPT has slipped below half the global AI-assistant market for the first time — a symbolic threshold that says less about decline than about the arrival of genuine competition. The interesting story isn’t that one product lost a few points. It’s why users are moving, and what a real multi-model world means for the teams deciding which tools to build their workflows around.
The numbers that moved
Sensor Tower’s 2026 figures, surfaced via Build Fast with AI, put ChatGPT at roughly 46.4% of the global AI-assistant market — the first time it has dipped under 50%. Google’s Gemini sits at around 27.7%, and Anthropic’s Claude at about 10.3%. (We’d flag these as worth verifying directly against Sensor Tower’s primary release, but the direction of travel is consistent with what operators have been reporting for months.)
The headline number deserves context, because share and scale are not the same thing. ChatGPT still commands roughly 1.1 billion monthly active users — a base no rival is close to matching in absolute terms. Falling below half the market while still being more than twice the size of your nearest competitor is not a collapse; it’s a maturing category. The pie is growing fast, and challengers are taking a larger slice of new growth even as the incumbent keeps adding users. For founders and marketers, that distinction matters: ChatGPT remains the default surface where most consumers live, but it is no longer the only place where serious work happens.
What’s genuinely new is that two credible alternatives have crossed from “niche favourite” into “meaningful market presence.” Gemini benefits from Google’s distribution muscle — Android, Workspace, Search — pulling it into hundreds of millions of daily touchpoints. Claude has grown more quietly, on a reputation for careful reasoning, long-context work, and writing quality that a specific, influential slice of professional users prefer.
Why users are switching
The more telling shift sits beneath the share numbers: people are actively comparing assistants rather than defaulting to whatever they opened first. Early on, ChatGPT enjoyed enormous default loyalty simply because it defined the category. That advantage is eroding as users develop opinions about which model is better for code, which drafts cleaner prose, and which they trust with sensitive material.
Trust and values are now part of the selection criteria — not a soft afterthought, but a driver of real behaviour. Build Fast with AI reports that OpenAI’s February deal with the US Defense Department reportedly triggered a measurable spike in uninstalls. Whatever one thinks of the politics, the signal is clear: a non-trivial number of users treat a vendor’s institutional choices as relevant to whether they keep the app. In a market where the underlying capabilities are converging, brand posture, data practices, and perceived alignment with a user’s own values become tie-breakers.
Then there’s the quality signal that should make every product team pay attention: paid conversion. Roughly 13% of Claude users are reported to pay for subscriptions — the highest paid-conversion rate in the field, per the same source. Free-tier scale tells you about reach; paid conversion tells you about revealed preference. When people open their wallets at an unusually high rate, it suggests the product is solving a problem they can’t easily solve elsewhere. A smaller assistant with a loyal, paying, professional base is a different — and arguably healthier — kind of business than a giant one monetising a thin slice of casual users.
What it means for teams
The practical lesson for teams is the one most have been slow to internalise: don’t standardise on a single assistant. The instinct to pick one vendor, sign an enterprise agreement, and roll it out company-wide is understandable — procurement loves simplicity. But in a multi-model world, that instinct quietly locks you into one company’s roadmap, pricing, and values at exactly the moment the competitive gaps are narrowing and the leader is changing.
A smarter posture is to match the model to the task. In practice, that looks like:
- Long-form writing and editing: teams frequently favour Claude for tone, structure, and handling large documents in a single context window.
- Research, multimodal, and Google-native workflows: Gemini’s integration with Workspace, Search, and Android makes it the path of least resistance for organisations already in that ecosystem.
- General-purpose breadth and ecosystem depth: ChatGPT’s plugin and tooling maturity, plus its sheer ubiquity, keep it the safe default for the widest range of tasks.
The exact allocation matters less than the principle: treat assistants as interchangeable components, not as a single platform decision. And watch switching costs closely. Lock-in in this market rarely arrives as an explicit contract clause. It creeps in through custom GPTs, saved prompts, fine-tuned integrations, and the muscle memory of an entire team. Every workflow you hard-wire to one vendor’s proprietary feature is a future migration cost you’re signing up for. The more model-specific your tooling, the harder it becomes to act on the very competition that’s now working in your favour.
The operator takeaway
If you run a team or a company, the strategic response to a multi-model market is to build for optionality rather than betting the org on a single winner.
First, build model-agnostic workflows. Where you can, route requests through an abstraction layer — an internal API, an orchestration tool, or a gateway — so swapping the underlying model is a configuration change, not a re-platforming project. Keep your prompts, evaluation sets, and institutional knowledge in your own systems, not trapped inside one vendor’s interface. The goal is to make the model a commodity input you control, not a dependency that controls you.
Second, evaluate on your own use cases, not on benchmarks or share charts. Public leaderboards and market-share reports — including the Sensor Tower figures discussed here — are useful for spotting trends, but they don’t tell you which model writes your customer emails best or debugs your stack most reliably. Assemble a small, representative test set of the tasks your team actually performs and run the leading assistants against it. The results will frequently surprise you, and they’ll be far more decisive than any headline.
Third, revisit your default quarterly. The pace of change in this category makes any “forever” decision a mistake. A model that trailed three months ago may now lead on the dimension you care about; a price change or a policy decision may shift the calculus overnight. Put a recurring review on the calendar — re-run your evaluations, re-check pricing, and reassess whether your chosen vendors still align with your needs and values. Treat your AI stack the way you’d treat any vendor relationship in a fast-moving market: with healthy, scheduled scepticism.
The drop below 50% isn’t really a story about ChatGPT losing. It’s a story about users gaining leverage. For the first time, the people choosing AI assistants have credible alternatives, real grounds to compare, and reasons beyond raw capability — trust, values, and fit — to make a switch. The teams that benefit most won’t be the ones who pick the right horse. They’ll be the ones who refuse to bet on a single horse at all.
