Checkpoint: Where Is Competition Actually Happening?

For the consumer AI assistant market, the useful interim judgment is not “open” or “closed.” It is a map of where control is concentrated and where users still make meaningful choices. The evidence so far supports a strong general-purpose default, but it also shows task-level routing, distinct audiences, and different payment channels beneath that default. Connecting those observations tells us what kind of competition exists now—and what has not yet been shown to last.

The distinction matters because the word “competition” can refer to several different events. A rival may be available for anyone to try. It may win a recurring task. It may attract users who pay directly or arrive through an institution. Or it may force the incumbent to change because users can credibly divert valuable work. These are related, but they are not equivalent. The evidence is strongest for the first three and does not directly measure the fourth.

A strong default is not exclusive control

The consumer market has a clear center of gravity. ChatGPT is the primary assistant for 58.2% of the weighted AI-user sample, and Gemini is primary for 25.4%. Together, those reported primary shares amount to 83.6%. The ranking is also statistically clean through the middle: ChatGPT exceeds Gemini, Gemini exceeds Claude, Claude exceeds Copilot, and Copilot exceeds Grok in adjacent comparisons. This is strong evidence that general-purpose primary choice is unequal, not a collection of roughly equivalent options.

The right conclusion is about the default layer. When “primary” means the platform through which a respondent routes the majority of their tasks, a large primary share represents repeated reliance and a favorable starting position for future interactions. It can help a provider accumulate context, become familiar, and receive routine requests without a fresh comparison each time. The concentration is therefore strategically meaningful even if no user is formally prevented from trying another system.

But the same definition sets a boundary. A primary platform receives a majority, not necessarily every task. A user with an 80–20 distribution and a user with a 51–49 distribution receive the same categorical label, even though their exposure to alternatives is different. The measure does not reveal whether the minority platform receives the most sensitive, valuable, or technically demanding work. Nor does it measure revenue, time spent, retention, or the consequences of a price or policy change.

This is why the primary hierarchy cannot be converted into a claim that the market is already closed in every sense. It establishes that the default is concentrated. It does not establish that users are exclusive, that challengers cannot enter a valuable task, or that the observed ranking will persist as products and privacy policies change. A strong default is a condition that alternatives must overcome, not proof that they cannot do so.

The multihoming evidence makes that boundary concrete. About four in five users have tried a second assistant, while about 43% intentionally route different task categories to more than one platform. The gap is analytically useful. Trial means that an alternative has crossed the threshold of consideration or use. Intentional routing means that the user has assigned different jobs to different providers.