If your brand pages and product listings are suddenly more visible — and listicles less so — in AI answers, there’s a technical reason: ChatGPT is running more parallel, scoped web searches (fan‑out queries) and increasingly using operators like site: and qualifiers such as “official” or “gov.” Those background queries shape which pages the model considers and which it will cite, and that changes the signals marketers must optimise for.
What changed: more searches, more deliberate scoping
Recent model updates show two clear shifts. First, ChatGPT is issuing more fan‑out queries per prompt. Second, those queries are more tightly scoped — often targeting specific domains or authoritative site types. Together, these moves reduce reliance on broad, SEO‑heavy pages (listicles, comparison posts) and increase retrieval of brand, product and official pages.
Industry measurements demonstrate the effect. Peec AI reported that when ChatGPT 5.6 became the default, single‑fan‑out prompts dropped from 94.0% to 43.5% while average retrieved sources rose from roughly 12 to about 24. In the same dataset, fan‑outs using the site: operator climbed from about 0.3% to roughly 23%.
Another study by Nectiv, sampling about 4,000 prompts, found average fan‑outs per prompt rising from 2.17 to 7.61, with the longest query chain growing from 4 to 29. In that analysis site:, “official” and “gov” ranked among the top tokens; site: appeared in roughly 64% of fan‑outs. Differences between these datasets underscore a key methodological caveat: results depend on whether researchers capture fan‑outs from OpenAI’s API or the ChatGPT consumer UI.
Retrieval vs. visible citation — a widening gap
Distinguish retrieved pages (those the model fetches during fan‑outs) from cited pages (those linked in the final answer). Several teams report retrieval counts increasing while the number of unique domains cited visibly in responses falls.
Resoneo measured unique domains cited per response dropping from 19 to 15 after a recent update. Ahrefs’ dataset of most‑cited domains shows consistent citation winners: Reddit, Wikipedia, Forbes, Merriam‑Webster, Consumer Reports, Healthline and Walmart. In practice, ChatGPT may consult many pages but privilege a narrower set of domains for the links it shows.
Who wins and who loses
Across datasets and hands‑on observation, product pages and brand domains are gaining retrieval share. Peec AI reported product pages made up about 16.39% of retrieved pages, ahead of listicles. When answers require specs or pricing, the model often runs site: searches against manufacturer or retailer domains. For subjective prompts, the model still leans on community discussion sites like Reddit and specific subreddits.
Formats that have been heavily optimised for broad organic traffic — listicles, how‑to guides and comparison posts — are losing retrieval share. Scoped queries (for example, site:brand.com or queries that include “official”) make it harder for spammy or SEO‑manipulated pages to influence synthesis.
Practical implications for marketers and publishers
Two consequences matter operationally. First, presence in ChatGPT’s retrieval set now favors clear authority signals: canonical brand pages, official resources and recognized publisher domains. Second, with fewer visible citations, much of the model’s synthesis may draw on sources without naming them — complicating measurement of AI‑driven referral or attribution.
Model variant and user tier influence what most users experience. OpenAI ships multiple variants (for example, Sol and the cheaper Luna). Analysts observed Luna rolled out as the default for Free and Go users around August, and that the free plan represents over 90% of weekly ChatGPT users — so the default free behavior has outsized impact on perceived answers.
What to do next
Instrument for two signals: retrieval presence and visible citation. Use tools that surface fan‑out queries (Resoneo plugin, FanoutFox, or API logs if accessible) to check whether your pages appear in the model’s searches for priority queries. Ensure canonical product and pricing pages are accurate, crawlable and explicitly attributable to your brand; the model increasingly prefers brand pages for factual details.
For YMYL (Your Money or Your Life) topics, prioritise authoritative, well‑sourced pages and structured metadata that reinforce trust signals. Maintain a visibility portfolio: continue standard SEO for organic discovery while also optimising the authoritative pages that scoped fan‑outs are likelier to retrieve and cite.
What to watch
Track two leading indicators: a continued rise in scoped fan‑outs (more site:, “official” and “gov” tokens) and a widening divergence between retrieval counts and visible citation counts. These trends show how models balance breadth against trustworthiness — and they will determine who gets seen and who gets credited in AI‑generated answers. Operationally, focus on crawlability, canonicalisation and clear brand attribution on core pages; those are the signals the model is increasingly tuned to respect.
