Query fan-out is the process of turning one prompt into several background searches before an AI writes its answer. ChatGPT's fan-out changed on August 8, 2026: domain-scoped site: searches jumped from 0.37 percent to 16.8 percent of all fanout queries in a single day, and searches per response nearly doubled.
That change points AI retrieval at your own domain. When ChatGPT runs a scoped search against yourdomain.com, thin pages and unindexed content cost you answers on top of rankings. The same fortnight delivered a second lesson about how fast this ground moves, and both point at the same fix.
Query fan-out, explained
Query fan-out is the step between a person's prompt and an AI's answer where the system writes its own searches. One question becomes several background queries: reformulations, comparisons, entity lookups, and, since August, searches scoped to a single domain. The engine reads what comes back, then composes. Your page competes for those internal queries rather than for the prompt a user typed.
The practical consequence is that a page built to rank for one head term can sit outside the retrieval set entirely. Coverage across the sub-questions is what earns the citation, which is the same principle behind our answer engine optimization work and the wider argument in the 2026 AEO playbook.
The August 8 shift toward domain-scoped searches
ChatGPT started searching individual domains at scale on August 8, 2026. Fanout queries containing the site: operator moved from 0.37 percent to 16.8 percent of all fanout queries in one day, roughly a 46x increase in share (Promptwatch). Scoped searches went from about one in every 280 fanout queries to closer to one in six.
A site: query restricts results to a specified domain, URL, or URL prefix, and the operator works across all Google Search properties (Google Search Central). Read that alongside the behavior change and the picture is blunt: the engine is going to your site to pull facts, which raises the stakes on everything covered in our guide to optimizing a website for AI agents.
Fan-out depth doubled, and the answer surface widened
ChatGPT now runs roughly twice as many searches per answer. The average number of fanout queries per response moved from about 1.08 to about 1.83 on August 8 and has held there (Promptwatch). Because the jump coincided with the site: surge rather than replacing generic searches, the domain-scoped lookups sit on top of the retrieval that was already happening.
More searches per answer means more distinct chances to be found and more angles to cover. Pricing, alternatives, comparisons, and objections each become their own retrieval, which is why depth on a topic now beats a single well-optimized page. The tradeoff between ranking work and answer work is the thread we pulled in GEO vs SEO.
Reddit's citation collapse and the volatility it exposes
One platform change can erase a domain's AI footprint inside a week. Reddit held an average 3.83 percent share of ChatGPT Search citations from July 18 through August 7, 2026; across August 14 to 17 that average was 0.52 percent, a relative drop of 86.4 percent (Promptwatch). Search Engine Land reported the finding with the caveat that the data shows when the shift happened rather than why, and that a collection issue cannot be ruled out (Search Engine Land).
Google's engines moved differently over the same period, which is the useful part. Reddit's share of AI Overviews citations slid from about 2.37 percent to 2.10 percent, and AI Mode from 2.22 percent to 1.54 percent, both gradual rather than a cliff (Promptwatch). Each engine deserves its own baseline and its own metric, a case we made when introducing share of LLM, and a reason to keep building the third-party presence described in our off-page SEO for AI search guide.
Indexing coverage is the gate on AI answers
A domain-scoped search can only return pages a search engine has already indexed. Google states that an indexed URL can appear for a related site: query without any guarantee, and points to the URL Inspection tool when a page fails to show (Google Search Central). Coverage gaps now cost you answers, not only positions.
So the unglamorous work matters more than it did in July. Crawlable internal links, category pages with real content, no orphaned URLs, clean canonicals, and a sitemap that matches reality. Most of that sits in the technical pass of our 2026 website audit checklist, and it is the foundation layer our digital marketing and technology team ships before any content plan starts.
A query fan-out readiness plan for the next quarter
Three jobs, in order: index coverage, topic coverage, monitoring. Confirm every commercially important URL is indexed and reachable. Expand thin topics into the sub-questions a buyer actually asks, since roughly 1.8 retrievals per answer means several separate entry points. Then track citation share by engine, weekly, with dates you can align against platform changes.
Promptwatch flagged one detail worth acting on: the queries ChatGPT generates from a prompt changed materially on August 8, so keyword mappings built before that date are stale (Promptwatch). Re-run your fanout analysis before you rewrite anything. If you want that run as a standing program rather than a one-off audit, our Next Best Action team owns it end to end, and you can start a conversation to scope it.
