site: Operator Queries

A site: operator query is a search restricted to a single domain, subdomain or URL prefix, written as site: followed by the address and usually a topic, for example site:example.com pricing. Search engines have supported the operator for decades. What is new is that AI search engines generate these queries themselves while researching an answer, which turns a brand's own domain into a place they look things up directly.

In classic search, site: is a research and debugging tool. Google's documentation explains that it returns results only from the specified domain or prefix, that it does not necessarily return every indexed URL, and that a site: query with no search term is not ranked in any meaningful order. SEOs use it to spot-check indexing, find duplicate or spam pages, and see which pages on a site match a term. Bing supports the same operator.

In AI search, site: appears inside query fan-out, the set of background searches an engine runs before writing an answer. On August 8, 2026, ChatGPT's use of the operator changed overnight. In Promptwatch's tracking, the share of ChatGPT fan-out queries containing site: rose from about 0.4% to about 17%, and the average number of searches per response rose from about 1.08 to 1.83 on the same day, which suggests the scoped searches were added on top of general ones rather than replacing them (ChatGPT site:operator fanouts). The change came two days after OpenAI announced GPT-5.6 Luna as the new default model for free users, though OpenAI has not described any change to its search behavior. Other trackers reported similar jumps, and one, Searchable, found that about 12% of site: queries pointed at the brand's own domain, with review and reputation sites such as Reddit, Trustpilot, G2 and BBB among the other common targets. Not every dataset agrees: DEJAN, analyzing about 197,000 OpenAI fan-out queries, found site: use stayed under 0.2% in August and attributed the gap to model version and prompt style. Its data also showed long, research-style prompts triggering scoped lookups up to 90 times more often than short, repetitive tracking prompts.

Why it matters for GEO: a scoped search can only return pages the underlying search index has from your domain, and it ranks them against each other, not against the web. That shifts attention to your own site's coverage. If an engine runs site:yourbrand.com pricing and your best pricing page is not indexed, or an outdated page outranks it, the answer will use whatever the search returns.

Practical steps:

• Run the scoped searches yourself on Google and Bing for the topics buyers ask about (pricing, integrations, security, alternatives) and check which page comes back first.
• Make sure key pages are crawlable by both search engines and AI crawlers, listed in XML sitemaps and linked internally, and redirect or update stale pages that still rank.
• Write pages that answer the specific question in their title and opening lines, so they win the scoped result and are easy to quote (answer-ready content).
• Watch your presence on the third-party sites engines scope into, since a site:reddit.com or site:g2.com lookup will surface what others say about you.

Track fan-out queries for your important prompts over time. Search behavior is controlled by the platform and can change in a day, as August 2026 showed. Related: ChatGPT search and grounding queries.