ChatGPT is now using the site: operator to search specific domains: what does that mean for SEO?
What it means for SEO when ChatGPT runs site: searches against specific domains, what to check on your own pages, and which tools expose the fan-out queries.
The reports that ChatGPT now runs site: searches against specific domains matter for one reason: they suggest the model sometimes decides which sites to look inside before it searches. That is a different behavior from sending a broad query and taking whatever comes back. Before drawing conclusions, a caveat. We have not independently verified how often ChatGPT does this, for which kinds of prompts, or what index sits behind it, and nobody outside OpenAI can state that with confidence. Treat it as a pattern to test on your own category, not a settled rule.
What follows is what the pattern would imply if it holds, and how to find out whether it applies to you.
What a site: query actually does
In classic search, site:example.com topic limits results to one domain. If a model writes that kind of query as one of its fan-out searches, two things follow.
First, the model has already chosen the domain. Something earlier in its reasoning, perhaps its prior knowledge, a source mentioned in another result, or the brand named in your prompt, pushed it to look there. Second, the page that comes back is decided by how well your own site answers that narrow query, because competitors are not in the result set at that moment.
So the contest shifts. It is less about outranking rivals on a broad query and more about being a domain the model thinks to look inside, then having a page that answers the question when it does.
What this means for SEO work
Domain recognition matters more. If a model scopes a query to a named domain, being a recognized source for your topic is the first gate. That is built through citations on third-party pages, consistent naming and being referenced in the places models already pull from.
Your site's internal findability matters. A scoped search only helps you if your own pages are indexed, titled and written so a narrow query surfaces the right one. Thin or vaguely titled pages lose here even when the brand is known.
Crawler access is still the baseline. If the relevant bots cannot fetch your pages, nothing downstream works. The real names to check in your logs and robots.txt are GPTBot, OAI-SearchBot and ChatGPT-User, along with ClaudeBot and PerplexityBot for other engines.
Comparison and review pages carry weight. When a prompt asks for a comparison, a model may look inside a review or comparison domain. Being present and accurate on those sites is part of the job.
None of this is exotic. It reinforces ordinary SEO fundamentals and adds one habit: look at the actual fan-out queries.
How to check whether it applies to you
You do not have to guess. The searches ChatGPT runs behind an answer are visible with the right tooling.
- Pick the ten to twenty prompts that matter to your category.
- Capture the fan-outs for each, noting any that use
site:and which domains they name. - Compare those domains against where your brand is cited.
- For domains that appear repeatedly, decide whether you need a presence there.
Promptwatch covers this with query fan-outs per model, plus a free GEO Inspector Chrome extension that shows ChatGPT fan-outs and the difference between sources fetched and sources cited, and runs locally. Its citation analytics then show which pages and domains are actually cited, with a Page Tracker for specific URLs, yours or a competitor's. Agent Analytics adds IP-verified crawler logs, so you can confirm that ChatGPT-User or OAI-SearchBot are fetching the pages you think they are, with a crawl-to-citation path and a citation rate per page. Crawler logs start on the Professional plan ($245 a month).
How the tools compare for fan-out visibility
| Tool | What it shows about fan-outs | Notes from the catalog | Price |
|---|---|---|---|
| Promptwatch | Fan-outs per model, free GEO Inspector extension | Also citation analytics, crawler logs, visitor analytics | Free tier; paid from $95/mo |
| Ceyo | Fan-out capture and version drift | Models per plan capped at 2 to 4 | From $39/mo billed yearly |
| Mentionable | Captures fan-out queries and drafts the missing page | Seven engines | From 79 EUR/mo |
| Aiso | Fan-outs from a consent panel of real conversations | Tracker, not a full platform | From $19/mo |
| Encited | Fan-out query insights | Business plan and above | Business $134/mo annual |
Ceyo captures fan-outs from real interfaces across eight engines, though it caps models per plan. Mentionable is a French-built tracker that captures fan-out queries and drafts the page you are missing. Aiso builds its data from a consent panel of real human-AI conversations, which is a different method. Encited adds fan-out insight on higher tiers but is an agentic infrastructure product first. AthenaHQ and Scrunch AI focus on tracking and optimization rather than fan-out inspection, according to their catalog entries.
Who should pick which
Promptwatch is the best fit if you want to see the fan-outs, then act on them: it connects the query, the citation, the crawl and the visit in one place, and the free extension lets you start without paying anything.
Ceyo or Mentionable suit teams that want fan-out capture in a lighter tracker and can accept narrower coverage on lower plans.
Aiso suits teams who care about where the prompts come from.
Encited suits teams already routing SEO work through an MCP-connected agent.
What to do this month
Run the fan-out check on your top prompts. If site: queries name your domain, make sure the page that should answer is indexed, clearly titled and fetchable by the ChatGPT bots. If they name other domains, treat those as targets for earned presence. If they never appear, you have learned that this is not yet a factor in your category, and you saved yourself a reorganization.
No tool can guarantee that ChatGPT recommends you. What good measurement does is replace speculation about operators with a list of the actual queries and sources involved.