The best model didn't win. The default did.

A milestone slipped by quietly this month: ChatGPT's share of the AI assistant market fell below 50% for the first time. It sat at about 46% by May, down from 65% in December 2024. Gemini climbed to roughly 28%, and Claude to about 10%.

Here's the part worth sitting with: ChatGPT is still growing — over a billion monthly users. It didn't get worse. It just stopped being the only door people walk through. And the main reason Gemini is gaining isn't a benchmark. It's that Gemini is the default — already inside Android and Google Workspace, turning passive Google users into active Gemini users without anyone choosing it.

Distribution beats the model

This is the oldest lesson in software, and AI keeps re-teaching it. The best product doesn't win the market; the one that's already in front of people does. Google didn't out-think OpenAI. It put its assistant where a few billion people already were and let the default do the work.

For builders this cuts two ways:

• If you're betting your product on "we use the smartest model," remember that the smartest model is a six-week lead, while a default is a moat. Capability is rented; distribution is owned.
• If you have distribution — an app people already open, a workflow they already live in — adding AI to that beats building a better standalone AI that nobody has a reason to open.

People also leave over trust, not just features

The other driver in the data is quieter and more interesting: people switched on values. The report notes a measurable spike in ChatGPT uninstalls after OpenAI's deal with the U.S. Department of Defense, plus pushback when ads arrived in the free tier. Users didn't leave because a competitor shipped a feature. They left because of who the company is and how it makes money.

That's a real signal for anyone building on top of these providers: your users' trust in you is partly tied to the company whose model you embed. Pick your defaults — and your providers — with that in mind.

So what do you actually do

• Don't compete on raw model quality alone. It's the most perishable advantage there is.
• Find the default. Build AI into the place your users already are, or partner with one, instead of asking them to adopt a new front door.
• Stay model-agnostic. When a provider's politics or pricing turns your users off, you want to swap the engine without rebuilding the product.

The bottom line

ChatGPT losing its majority isn't a story about a worse model. It's a story about where people already are.

The best model rarely wins the market — the default does, and trust decides the rest. Build for distribution, keep the model swappable, and remember that "smartest" is a lead measured in weeks while "already there" is measured in years.