Where do AI search engines actually pull their citations from?

Ranqo's same production dataset breaks down where AI search engines' citations actually point, and finds the citation surface is dominated by corporate web content rather than independent media. "When the AI search engines cite any sources, about 78% of those citations go to corporate websites, including the brand's own pages and third-party pages from companies in the same space." Among the roughly one-fifth of citations that aren't corporate, Ranqo reports YouTube is cited most often, ahead of Reddit, editorial media, and Wikipedia. At the page level, the single most-cited content format is the ranked "best-of" listicle, accounting for about 21% of all citations — Ranqo notes "a single 'best X tools' page can surface a brand across many AI answers at once."

Ranqo frames this as the practical, page-level expression of the same concentration dynamic behind its tier ladder: engines cite a narrow set of outlet types repeatedly rather than sampling broadly across the web, so a brand's odds of being cited depend heavily on whether it appears in the handful of formats — its own site, competitor or industry corporate pages, and third-party listicles — that the engines already lean on. The paper separates this from sentiment, which it finds comparatively unstable: "whether the AI frames a brand positively or negatively flips about 6.7 times more often than whether the brand is mentioned at all," so Ranqo treats mention, position, and sentiment as three separate metrics rather than one composite score, since source concentration governs the stable one (mention) and not the volatile one (sentiment).

This citation-source concentration parallels a bias documented from a different angle in Do language models favor sources regardless of item quality?, where shopping agents favor items from a preferred source independent of quality; Ranqo's finding shows a structurally similar favoritism at the search-engine level, where corporate and already-prominent pages crowd out independent media regardless of what an independent source might say. It also complements How consistent are AI brand recommendation lists across repeated prompts?: where SparkToro shows an engine's exact recommendation list is unstable across repeated runs, Ranqo shows the pool those lists are drawn from is narrow and corporate-skewed even as the specific picks shift. Both point to the same underlying structure — volatile surface, concentrated source pool — described from the brand-tier side in Do AI search engines favor well-known brands over others?, this run's companion note.

The excerpt gives source-type and format shares as descriptive statistics from Ranqo's own tracked-brand corpus; it does not test whether changing a brand's own page content shifts its share of corporate citations, and Ranqo attributes the content-quality mechanism (quotations, statistics, citations) to a separate study, Aggarwal et al. (2024), rather than demonstrating it directly in this dataset. The listicle finding is also a correlational share of citations, not evidence that producing a listicle causes citation — Ranqo proposes that causal test as a future v1.1 protocol rather than reporting it here.