An independent test of 10,000 ChatGPT responses found that brands were recommended 44% more often when at least one page from their website was cited as a source. The finding is useful for AI visibility strategy, but it should be treated as evidence of an association, not proof that earning a citation directly causes ChatGPT to recommend a brand.
The researcher built 1,000 "what is the best" prompts across categories, ran each prompt ten times, then compared brand recommendations with citations. The methodology and reported results were shared publicly in the bigseo community on Reddit.
What did the 10,000 response test measure?
The test asked whether recommendation rates differed when a brand's own domain appeared among ChatGPT's cited sources. It did not simply count links. It compared whether the brand itself was recommended when one of its pages was cited versus when none of its pages was cited.
The reported pattern held across several page types. Product pages showed the largest lift in the shared breakdown, while articles, homepages, and category pages also showed higher recommendation rates when cited.
Why could citations and recommendations move together?
A citation can be a signal that the system found the brand's website relevant to the question. If a page contains clear product facts, category information, evidence, or useful comparisons, the same information that makes the page worth retrieving may also make the brand easier to include in the answer.
That is why AI visibility should look beyond simple mention counts. Marketers need to understand which pages are being retrieved, what those pages say, which prompts trigger them, and whether the resulting answer positions the brand accurately.
The important distinction is that the citation may be part of the same retrieval process as the recommendation. It is not necessarily a separate lever that marketers can manipulate independently.
What did the page type results show?
The researcher reported the following recommendation rates when a brand page was cited compared with when it was not cited.
- Product pages: 43.6% versus 29.8%, a reported 46% lift
- Category pages: 41.2% versus 29.7%, a reported 39% lift
- Homepages: 41.0% versus 29.9%, a reported 37% lift
- Articles: 40.4% versus 29.6%, a reported 36% lift
The relatively small spread between page types is interesting. It suggests that marketers should not assume only product pages matter for AI discovery, although the dataset alone cannot establish a universal ranking rule.
What does the study not prove?
The test was not an OpenAI study, a peer reviewed paper, or a controlled experiment showing causation. We do not know whether citations caused recommendations, recommendations increased the likelihood of citations, or both were driven by another factor such as brand relevance, source quality, or retrieval behavior.
The result also comes from one prompt format centered on "best" questions. Commercial comparisons, technical questions, local searches, branded queries, and informational prompts may behave differently.
For marketers, the responsible conclusion is that owned site citations deserve measurement. The irresponsible conclusion would be that getting any page cited guarantees a 44% recommendation lift.
How should marketers respond to the finding?
First, identify the commercial prompts where a recommendation would matter. Then record whether the brand is mentioned, whether it is recommended, which pages are cited, and which competitors receive the same treatment.
A useful learning center approach is to improve pages that already answer real buyer questions rather than creating thin content solely to chase AI citations. Clear product information, original evidence, useful comparisons, specific use cases, and consistent positioning can make a site more useful to both people and retrieval systems.
The goal is to build pages worth citing and a brand worth recommending. Those outcomes may reinforce each other, but they should still be measured separately inside a broader marketing strategy.