A citation to YouTube, Reddit or LinkedIn tells us where content appears—not necessarily who is speaking. Behind the link might be a creator, a company, an institutional publisher or a conversation involving several people.
Tocanan’s analysis of more than 1.3 million recorded response–URL pairs reveals a mixed picture. Seven social-platform groups account for a small share of the sample, yet their sources include both individual and organisational voices. The cited video channels are not uniformly large, either.
For brands, the useful question is not simply “Which platforms get cited?” It is “Whose knowledge reaches the answer—and in what form?”
Key findings
In Tocanan’s observational dataset recorded within 28 November 2025–30 September 2026, seven named social-platform groups accounted for 4.16% of 1,311,880 AI response–URL pairs. YouTube represented 44.51% of that seven-platform subtotal. Cited destinations included individual creators, institutions and commercial publishers—not just influencers. Among eleven selected YouTube channels checked on 1 October 2026, four had fewer than 10,000 subscribers.
What our citation records can tell us
We analysed an observational sample from Tocanan’s GEO monitoring, covering records within 28 November 2025–30 September 2026. The analysis focused on recorded citations in successful, nonempty AI responses.
That produced 1,311,880 distinct response–URL pairs. The same URL counts once within a response but can recur across responses. These are recorded citations, not unique websites, people, visits or endorsements. These findings describe the monitoring sample, not AI use as a whole.
Seven named social-platform destination groups—YouTube, Reddit, Facebook, Instagram, LinkedIn, TikTok and X—accounted for 54,526 pairs, or 4.16%.
The share remained similar when we restricted the analysis to the final 180 days (4.49%) and 90 days (4.32%), ending 30 September 2026. These are consistency checks within the sample, not universal benchmarks.
It would be tempting to read the small share as evidence that human voices barely enter AI answers. That would be the wrong conclusion.
Which social platforms appear in AI citations?
The combined 4.16% conceals a very uneven distribution. YouTube accounts for 44.51% of the seven-platform subtotal, followed by Reddit at 18.88% and Facebook at 15.46%. TikTok and X are much smaller in this captured corpus—not necessarily in consumer influence.
| Destination group | Response–URL pairs | Distinct exact URLs | Share of seven-platform subtotal |
|---|---|---|---|
| YouTube | 24,269 | 2,247 | 44.51% |
| 10,293 | 720 | 18.88% | |
| 8,432 | 962 | 15.46% | |
| 6,912 | 546 | 12.68% | |
| 4,010 | 191 | 7.35% | |
| TikTok | 587 | 51 | 1.08% |
| X | 23 | 6 | 0.04% |
| Total | 54,526 | 4,723 | 100% |
The percentage column uses only the 54,526 social-destination pairs as its denominator. The 4,723 exact URLs are not 4,723 accounts: one account can publish many posts, while URL variants, community threads, discovery pages and platform homepages also appear. Rounded platform percentages may not sum to exactly 100%.
What types of accounts do AI answers cite?
The selected examples reveal different kinds of participation: identifiable publishers and creators, multi-author discussions, discovery pages and accounts whose ownership remains unclear. They illustrate the mix, rather than measure each account type’s share.
YouTube: personal creators alongside institutional publishers
The identified channels behind the leading videos include personal creator channels, specialist advice channels, an information-publishing channel and an official communications channel. The associated video metadata points to explanatory, comparison and experience-led formats—not one uniform kind of speaker.
This makes YouTube a clear example of individual and institutional voices appearing on the same platform. Channel identity alone does not establish editorial independence, sponsorship or who contributed to a particular video. These selected examples cannot tell us what proportion of all YouTube citations belongs to creators.
Cited channels are not necessarily large channels. Eleven selected YouTube channels displayed roughly 5,000–279,000 subscribers, with four below 10,000. That range shows why citation visibility should not be equated with a large following; it does not establish how subscriber size affects citation.
Reddit: community questions and peer discussion
The reviewed destinations include question-led discussions, practical advice and recommendation threads. The relevant unit is often a conversation: an original poster raises a question, and multiple contributors add experiences, suggestions or competing views.
That differs from a single creator publishing a standalone recommendation. Community participation also does not guarantee independence; commercially affiliated contributors may take part. A cited thread should not be treated as one verified expert or one shared opinion.
Facebook: institutional publishing and community participation
The reviewed destinations include an identified broadcaster account and group-discussion pages. Other account paths suggest editorial and branded publishers, although those identities were not all independently verified.
The distinction matters: a page can carry an organisation’s institutional voice, while a group can contain contributions from many people. Neither format should automatically be classified as individual influencer content, and the page owner is not necessarily the author of every contribution.
Instagram: separate the author from the discovery surface
The leading destinations include post and reel URLs alongside discovery-style pages. Posts and reels can potentially be traced to an uploader; a discovery page does not identify a single speaker.
The reviewed post and reel authors remain unresolved, so there is not yet a defensible account-type breakdown for this subset. The supported finding is about destination format: an Instagram citation cannot automatically be counted as a creator’s authored recommendation.
LinkedIn: company publishers and founder-professionals
Verified examples include company-published explanatory content, product-led corporate posts and an individual founder sharing professional advice. These represent organisational publishers and named professionals with commercial affiliations.
Individual authorship and commercial interest can coexist. A founder’s post may contain useful expertise, but should not automatically be labelled independent consumer opinion. Equally, a company article published on a social platform remains company-authored content.
TikTok: creator profiles with commercial activity
The profile review found a personal creator presence with associated commercial offers. Other cited handles suggest publisher and institutional accounts, but the links between those public profiles and the exact cited TikTok videos remain incompletely verified.
The useful distinction is between personal presentation, publishing identity and commercial involvement. A creator can operate a business, and an organisation can use a creator-style format. The current evidence supports those distinctions, not a quantified split between independent creators and organisations.
X: distinguish account posts from platform-level links
The small set of destinations includes post URLs, account-shaped paths and the platform homepage. Account ownership remains unresolved, so we cannot responsibly characterise the dominant speaker type.
A platform-level link is not an authored opinion. Keeping it separate from an identifiable account post avoids overstating how much individual participation the citation record demonstrates.
The central finding is a mix of voices, not a contest between “people” and “brands”. The reviewed examples include personal creators, institutional publishers, commercially affiliated professionals and community contributors. These categories can overlap. The remaining 95.84% of citation pairs cannot simply be labelled brand-owned either: individuals also publish outside social platforms.
A person’s voice does not have to live on a social platform
Our publisher review identified ten personal or creator-led website scopes outside those seven social groups. Together, those domains accounted for 16,090 response–URL pairs in the filtered all-history sample.
That is not a complete creator count. Nor does it mean every cited page was independently checked for authorship. The evidence identifies creator-led websites; particular articles may involve collaborators, sponsors or other contributors.
But it establishes a practical problem with the original assumption: searching for people only inside social-platform domains misses people publishing on their own websites.
The reverse problem also matters. A social-platform URL can host a company’s article or a commercially interested recommendation. Its host alone does not establish an independent human perspective.
“Individual creator”, “commercial publisher” and “social platform” describe different things. A creator can run a company. A company can publish on LinkedIn. A community discussion can contain several voices with different affiliations. Treating these as mutually exclusive categories would make the story simpler—and less accurate.
Do cited accounts need large followings?
Not universally: four of the eleven selected YouTube channels had fewer than 10,000 subscribers. However, this was not a random sample and did not compare cited with uncited channels. It cannot show that smaller channels are more likely to be cited, or that follower count does not matter.
AI’s information sources are not one uniform list
The observed AI platforms did not present identical source lists. But their prompts, models and coverage were not fully equivalent, so differences cannot be attributed to engine preference alone. Nor do the available records support a controlled China-versus-non-China comparison.
For brands, the practical point is straightforward: visibility in one monitored AI platform should not be assumed to represent visibility in another.
Being cited is not the same as being trusted
A citation identifies a source in an answer. It does not establish that people trust the source, or that a model evaluates credibility in the same way a person does.
An uncited creator may still have influenced the wider information environment. A frequently cited company page may still need scrutiny. Citation frequency is useful evidence of visibility, not a credibility score.
The topics being monitored also shape which sources appear. More observations do not, by themselves, make a sample representative.
What does this mean for a brand’s GEO strategy?
Assess human credibility and AI citation visibility together—but do not mistake them for the same achievement.
A partnership with a knowledgeable creator may be valuable to people without producing a directly identifiable AI citation. Equally, a cited corporate article is not proof that consumers trust it. The useful work is to trace the expertise, evidence and attribution behind the content—not merely count platform logos or increase publishing volume.
For a GEO and AI social listening firm such as Tocanan, this changes the question from “Should brands choose people or content?” to “Where does credible expertise appear, who is speaking, and what can we actually observe?”
Before concluding that AI overlooks people, make sure your measurement can recognise where those people publish. Count the citations, but understand the voices behind them.
To distinguish citation visibility from social conversations, explore our AI social listening and consumer insights. For how citation links are presented, read our analysis of Google’s five AI Overview citation surfaces.
About this analysis
This observational, non-random sample covers 1,311,880 distinct response–URL pairs from successful, nonempty AI responses recorded within 28 November 2025–30 September 2026. Repeated exact URLs within a response count once; unresolved citation discrepancies were excluded. Coverage varies by topic and platform. Account profiles were reviewed selectively, with unresolved identities left unclassified. Findings describe this sample—not all AI usage—and citations do not establish trust, endorsement or causation. Subscriber counts were checked on 1 October 2026 for eleven purposively selected YouTube channels; they are not historical counts at citation time.