Americans search in ChatGPT at about the same rate everywhere. The heavy users do something else.

Last week I read OpenAI’s own usage data for what it says about buyers: the people who hold the budget have turned ChatGPT into a search engine, and markets that adopt it deeper become search markets. That second point came from 150 countries. The same release carries the United States state by state, and I expected the same pattern inside the country. It isn’t there. The US story is different, and for a brand selling into the US it is simpler.

A reminder of what the data is: shares of messages, not people; consumer accounts only; product questions mixed in with facts and recipes under one label, seeking information, which I call search below. State figures are quarterly, and small states carry more noise than large ones, so the large states carry the argument.

The search habit rose in almost every state

Nationally, search went from 19.0% of messages in the first quarter of 2025 to 20.9% in the third quarter of 2026. It rose in 44 of the 51 states and districts, and in every one of the 20 largest.

The states that use ChatGPT most are not the ones that search with it most

Across countries, the heavier a country’s use per person, the larger its search share. Across US states there is no such link at all.

The ten states with the highest use per person put 21.3% of their messages into search; the ten with the lowest, 20.6%. The difference is inside the noise. What does separate them is writing: 21.6% of messages in the heavy states against 19.5% in the light ones. And self expression runs the other way: 9.4% in the heavy states, 10.8% in the light ones.

The District of Columbia is the clearest case. It has the highest use per person in the country, in every quarter since the data begins. It also has the highest writing share, 26% of messages, and one of the lowest search shares, 18%.

So the heavy use in the capital and on the coasts is work use: documents, emails, reports.

The search habit is not concentrated there. It is spread across the country at close to one rate.

Among the large states, the spread is narrow

Take the 20 most populous states, where the samples are large enough to trust.

Search runs from 19.1% in Virginia to 22.7% in Wisconsin, with a median of 21.0%. Wisconsin, Massachusetts, Washington, Minnesota and Colorado sit at the top; Virginia, Arizona, Michigan, Georgia and Pennsylvania at the bottom. The whole range is three and a half points. Every one of the 20 rose over the six quarters.

The small states look more dramatic: Delaware at 26%, Alaska at 24%, Montana at 23%, each up six or seven points. I would not build anything on those. OpenAI adds noise to small cells to protect privacy, and a state with few messages swings for that reason alone.

What this means if you sell in the US

There is no “AI search state”. Your buyers in Ohio search in ChatGPT at about the rate your buyers in California do, and the rate is rising in both.

A US brand does not need a state strategy for AI search. It needs a category answer.

When a buyer in any state asks the assistant about what you sell, what does it say, who does it name, and where in the conversation does it stop naming you? That answer is not in OpenAI’s data, for any state. It is in the assistants’ answers, for your category, and it has to be researched.

We do that research for brands. If you want to know what the assistants tell your buyers at each step, and where you lose them, start with the free assessment.

Data: OpenAI Signals v3.0, OpenAI, 2026, released 26 October 2026, licensed CC BY 4.0: the US topic by state by quarter file and the state ranking by messages per person (US Census 2023 population). Shares of classified consumer ChatGPT messages. Full download: openai.com/signals/data-download.