YouTube appeared more often than any other website in a week-long study of Google AI Overviews for vitamin and supplement searches. The video platform was cited in 186 of 350 generated answers, or 53.1%, and was the only site present in more than half of the collected responses.
The result describes citation frequency, not the accuracy or medical quality of the cited material. The research, conducted by Searcherries, tracked a defined set of searches under one configuration. It shows how Google assembled answers during that period and offers no evaluation of whether a video, clinic page or government source gave the best advice.
A fixed panel makes citation patterns measurable
Searcherries selected 50 questions from Google Autocomplete and ran each one daily from August 30 through September 5, 2026. The topics included hair growth, energy, sleep, anxiety and weight loss. Every search produced an AI Overview, giving the study 350 generated answers to compare.
The collection used US English desktop results configured for New York. Across the week, the answers contained 5,036 citation occurrences pointing to 572 distinct pages on 218 websites. When the same website appeared repeatedly within one answer, the researchers counted it only once for the site ranking.
That method provides consistency across days, but its limits are equally important. The panel covers 50 autocomplete-derived queries, one location and one desktop configuration for seven days. It does not establish how AI Overviews behave for all health searches, in other countries or on mobile devices. The study also records what Google cited without testing the claims inside the cited pages.
Video led a mixed group of health sources
YouTube appeared in 53.1% of the answers and across 44 of the 50 queries. Its presence was especially pronounced among questions beginning with “vitamins for,” where it was cited in 78.6% of responses. Across the entire panel, it appeared almost twice as often as Healthline.
GoodRx ranked second, appearing in 48.3% of answers. Healthline followed at 28.6%, Cleveland Clinic at 28.0%, and the National Institutes of Health domain at 25.1%. These figures count the share of answers that cited each website, so they should not be read as shares of all individual citation links.
The ranking shows that the generated summaries drew from several kinds of source: a large video platform, commercial health information sites, a clinical institution and a US government domain. Searcherries did not classify individual videos or pages by creator, evidence level or editorial process. YouTube’s lead therefore measures reach within Google’s citation selections, not uniform authority across the material hosted there.
AI citations often diverged from organic results
Only 32.4% of pages cited in AI Overviews also appeared in the organic results recorded for the same search. The study found that 67.6% of cited pages were absent from those accompanying lists. When researchers relaxed the comparison from an exact page to any page on the same website, overlap rose to 52.5%.
That gap indicates that an AI Overview may select a different page from a domain already visible in organic search, or draw from a website not present in the recorded seven to ten results. The research does not explain why Google chose those pages, and it does not show whether the generated answer relied more heavily on citations that appeared first or were repeated.
Citation stability also varied by query. For “supplements for weight loss,” the National Institutes of Health domain appeared every day and YouTube appeared on six of seven days, while other sites changed. By contrast, the query about supplements for GLP-1 side effects returned the same cited pages throughout the week.
The study’s clearest finding is therefore narrower than a judgment about health information: for this panel, time and configuration, YouTube was the most consistently widespread citation source. Understanding what that means for answer quality would require a separate assessment of the videos, pages and claims Google selected.
