Lily Ray (2026) — ChatGPT fan-out queries & site: search patterns

Practitioner

Publisher: Lily Ray
Vendor interest:: Low
Sample:: Mixed and not standardized. The article combines the author's own observations with third-party studies, including about 4,000 prompts in one cited dataset, 27 conversations and 3,554 retrieved pages in another, and roughly 20 prompts in one of the author's examples.
Methodology: Not disclosed
Geography: Global
Data period: 01.01.2025 – 31.08.2026
Published: 17 Aug 2026

Summary:

The article argues ChatGPT is changing how it searches the web: it appears to run more background “fan-out” queries and increasingly narrows results to “trusted” sources like official brand sites, .gov pages, Wikipedia, Reddit, and major review platforms.

It’s a practitioner take by SEO professional Lily Ray, not a formal study. She synthesizes recent industry findings with her own observations to suggest ChatGPT is using site: searches to filter spammy or low-quality sources.

Rather than one clear experiment, she draws on mixed evidence from multiple tools and third-party datasets (e.g., Peec AI, Profound, Resoneo, FanoutFox, Search Console, Bing Webmaster Tools), plus small internal tests. Sample sizes vary widely (from ~20 prompts to thousands).

Across the piece, the pattern is consistent: more fan-out searches, a big rise in site: usage, and query routing by intent — facts to official sites, legal/health to .gov, and opinions to Reddit/review platforms.

At the same time, retrieval appears to be rising while the number of domains cited in answers falls. Brand product pages may be replacing listicles/comparisons, and early “guessed” brands can shape later citations — sometimes incorrectly if the model picks the wrong official domain.

Main caveat: it’s a fast-moving, multi-source synthesis with uneven methods and some small or skewed samples, so the conclusions are suggestive rather than definitive.

16.39%

of ChatGPT retrieved pages are product pages, now more than listicles

Reddit, Wikipedia, Forbes, Merriam-Webster, Consumer Reports, Healthline, and Walmart are among the most cited domains in ChatGPT answers.

unique domains cited per ChatGPT response after 5.3 update, down from 19 previously

For every batch of roughly 20 legal prompts tested, ChatGPT returned only .gov domain citations every time.

23%

of ChatGPT fan-out queries now include the site: operator, up from 0.3% before version 5.6

For facts, ChatGPT fan-out queries go to official brand sites, while for opinions and product reviews, they favor Reddit and review platforms.

ChatGPT now runs more fan-out searches per prompt, with free and cheaper tiers catching up to higher-tier models.

7.5

average fan-out queries per prompt for ChatGPT 5.6 Sol, up from 2.17 in the 2025 baseline

of 27 ChatGPT initial queries contained brand names never mentioned by the user

33x

greater chance a brand named in ChatGPT's first fan-out query is cited versus a page only fetched

Using 'Official Site' or similar in your homepage title or meta description may help ChatGPT recognize your real domain, especially if your brand shares a name with others.

ChatGPT sometimes guesses the wrong brand domain, directing site: queries at parked or unowned domains – especially for lesser-known brands.

33.5%

of ChatGPT prompts needed a second fan-out iteration after version 5.6, up from 5% before

Lily Ray (2026) — ChatGPT fan-out queries & site: search patterns

Summary:

Cards from this research