I run a small studio of websites, mostly built and maintained by AI agents. One of them, Monitoring Room, has a page that tracks what prediction markets price for the 2026 US midterms: monitoringroom.com/us-midterms-2026.
Our analytics feed splits arrivals by source every night. Here is what it read this morning (3 November, election day), for the whole site:
| source | last 7 days | 7 days before |
|---|---|---|
| chatgpt.com | 116 | 7 |
| direct | 85 | 47 |
| Google organic | 8 | 4 |
| all our X posts together | 5 | 1 |
Over 28 days, 119 human sessions landed on that one midterms page straight from ChatGPT, and 75 of them were engaged in Google Analytics' sense (10+ seconds on the page, or a second page). Perplexity showed up for the first time this week with 2. The readers came from the US (23 %), South Korea (15 %) and Germany (7 %) first.
We did not pitch anything to anyone. Nobody here knows exactly why the assistant started citing this page. Here is what I can tell you, and it is all checkable on the page itself.
/llms.txt
None of that was built for ChatGPT. It was built so a reader could check us. It may be the same thing.
A page that an assistant sends 100+ people a week to had no way to keep any of them. As of today it has one small box under the board: leave an email and get one message when these markets resolve, with the result next to what was priced on the eve. One field, one consent line, one-click unsubscribe. We'll see whether a reader who came from an AI answer wants the follow-up.
The page is information only: market prices are observations, not forecasts, and nothing here is betting or political advice.
If you run content sites, the thing I'd check this week is your own analytics for chatgpt.com as a source. Ours went from a rounding error to our biggest single source in seven days, on a page with a hard date in it.