TwoSetAI
Founder Insight

How AI Cites YouTube Videos: What Each Engine Reads

Angelina Yang, Founder & Host at TwoSetAI

Listen on TL;Listen Prefer to listen? Hear this article read aloud.

Last updated: 2026-10-13

TL;DR

  • The split: In OtterlyAI’s study of 100 million+ AI citations, 38.7% of YouTube citations came from Perplexity, 36.6% from Google AI Overviews, 19.6% from Google AI Mode, and 4.4% from ChatGPT [1].
  • Google favors video in aggregate: Surfer SEO’s analysis of 36 million AI Overviews and 46 million citations, reported in Stockebrand’s O’Reilly book on AEO, found YouTube led Google AI Overviews citations at roughly 23%, with Wikipedia second at 18% [2].
  • ChatGPT leans text: Analyze AI’s 83,670-citation study put ChatGPT’s Wikipedia citation rate at 12.1%, Perplexity’s near zero, and Claude’s at 0.1% [2].
  • Our audit: Among our most recent guests, Perplexity cited 90 percent and Google AI Mode cited 90 percent, but When ChatGPT did not cite a YouTube video, it reached guests through third-party authoritative posts [3].
  • Caveat: these are averages across millions of queries; your category may cite YouTube far more or less, and answer sets shift from run to run [1][13].

In OtterlyAI’s YouTube Citation Study 2026, 38.7% of all YouTube citations came from Perplexity and 4.4% came from ChatGPT [1]. The spread is not a size gap. It is a retrieval gap. Each engine reaches a video through a different text layer, and some engines barely reach video at all.

This piece sits under the video pillar, which showed that 94% of YouTube citations go to long-form video and that engines read video through its description, chapters, and transcript [1]. Here I narrow the question: which engines actually cite YouTube, how do they get to the video, and what does a founder do with each one? I call the specific text layer an engine traverses to cite a video its Reach Path. Four engines matter, and each takes a different one.

Why does ChatGPT barely cite YouTube videos?

ChatGPT supplied 4.4% of all YouTube citations in OtterlyAI’s data, far below Perplexity and Google [1]. The reason sits in how ChatGPT answers.

Stockebrand’s O’Reilly book on AEO lays out two modes [2]. In parametric mode, ChatGPT answers from its training data without retrieval, and there are no real citations. In live-search mode, it issues a search through a combination of Bing’s API and OpenAI’s own retrieval, fetches pages with the ChatGPT-User crawler, and returns numbered footnotes alongside a sources sidebar [2]. Live search triggers on brand, recommendation, and fact queries, which is where video would help [2]. But the sources ChatGPT reaches for are text-heavy. Wikipedia accounted for 12.1% of ChatGPT citations in Analyze AI’s 83,670-citation study, with Amazon and Reddit picking up share on commerce queries [2]. OpenAI’s own description of ChatGPT search frames it as answers with links to web sources [8]. A YouTube video has a title, a description, and a transcript, but ChatGPT appears to prefer passages it can extract directly from text pages.

Our audit saw the same pattern. In August 2026, When ChatGPT did not cite a YouTube video, it reached our guests through third-party authoritative posts (founder profiles, twosetai.com pages, Medium posts) [3]. That is why the Founder Visibility Program ships a written editorial layer alongside the recording. The clearest single case was Alexander Berger, COO of Bolt.new, where ChatGPT named our episode “Best overall” for his story but reached it through Medium posts, not through YouTube [11]. Berger said something in that same conversation that frames why ChatGPT’s gap matters: “the importance of distribution and go to market is becoming a lot more important as it becomes easier to take an idea and turn it into software” [11]. Building the video is cheap; being findable in the engine where your buyer looks is the work. If you want ChatGPT to cite your video, pair the video with a written companion. The video on its own may not be enough. The sibling spoke on how AI engines decide what to cite goes deeper on the mechanism.

Why does Perplexity cite more YouTube videos than anyone?

Perplexity supplied 38.7% of all YouTube citations in OtterlyAI’s data, the largest share of any engine [1]. The design of the product explains the result.

Perplexity was built as a RAG-native answer engine. Stockebrand reports that every query triggers a real-time search against a proprietary index of more than 200 billion URLs, with retrieval combining roughly 30 vector results and 20 keyword results, and 3 to 5 inline citations per answer [2]. Perplexity’s own help documentation says every answer carries numbered citations [7]. Peec AI’s research found that AI answers on Perplexity draw on both a video’s description and its auto-generated transcript, and separately documented Google AI Mode building an answer directly from a description [4]. The transcript turns a video into a searchable text corpus, and Perplexity ranks it alongside any other authoritative page.

Perplexity also leans into community content. Reddit was about 6.6% of total citations in Profound’s 680-million-citation dataset [2]. Wikipedia was near zero [2]. Combined with YouTube, that is Perplexity’s source mix: long-form video, community discussion, and canonical third-party research. Freshness matters here too: a Seer Interactive study from mid-2025 found that 50% of Perplexity’s citations came from content published in the same year [2].

Our audit saw it the same way. Perplexity cited the overwhelming majority of our most recent guests [3]. In the Alexander Berger case, Perplexity called the episode “the best fit” and independently recommended the 21-minute highlight version [11]. The video shows up because the description, chapters, and transcript give Perplexity three places to retrieve from. For a founder, Perplexity is the engine where video pays back fastest.

What does Google’s AI Mode and AI Overviews read when they cite a video?

Taken together, Google’s two AI surfaces supplied the majority of YouTube citations in OtterlyAI’s data: 36.6% from AI Overviews and 19.6% from AI Mode, 56.2% between them [1]. On aggregate, YouTube is also the single most-cited source in Google AI Overviews. Surfer SEO’s analysis of 36 million AI Overviews and 46 million citations, cited in Stockebrand’s book, found YouTube led at roughly 23% of citations, with Wikipedia second at 18% [2]. That figure is an average, so some categories cite YouTube far more and some far less [2].

The mechanism inside Google is specific. Google documents native video input: the Gemini API accepts public YouTube URLs, processes audio and visual streams, and takes questions about moments by MM:SS timestamp [5]. On top of that, Google uses query fan-out, breaking a query into parallel subqueries and stitching the results together [2]. AI Overviews behaves more like a summary layer over organic rankings, while AI Mode generates a draft answer and fans out to validate it, which Stockebrand reports pulls from substantially different source pools: Ahrefs found only 13.7% citation overlap between the two [2].

Chapter-level citation showed up only on Google’s AI surfaces in OtterlyAI’s data: 73% of chapter-level citations landed on AI Overviews and 27% on AI Mode [1]. Google’s own guidance says it prioritizes the key moments you set in the description [9]. The entry points are the description, the chapter list, and the transcript, and Google surfaces them with precision. The sibling spoke on how AI engines parse video covers the inner mechanics in more depth.

Our audit saw Google AI Mode cite the overwhelming majority of our most recent guests, often across multiple sections of the person’s bio [3]. The Berger case is the clearest: Google AI Mode cited the interview across his biography, title claims, and a specific topic section [11]. One caveat worth holding: Google AI Overviews is unstable. Ahrefs research cited in Stockebrand’s book found about 70% of AI Overviews answers for the same query change between runs, and 45.5% of citations get swapped [2]. Appearing in Google AI once is not the same as holding that spot.

Does Claude cite YouTube videos at all?

Claude is the hardest engine to measure on video, and OtterlyAI’s YouTube study does not report a separate Claude share [1]. The adjacent evidence suggests the share is small.

Stockebrand’s book is specific about Claude’s citation behavior [2]. In Analyze AI’s study, Claude cited LinkedIn zero times across the period and cited Wikipedia at 0.1%, versus ChatGPT’s 12.1% [2]. Claude’s web search is powered by Brave Search rather than Bing or Google, and Claude’s default conversational mode does not cite sources unless asked and given source material [2]. Anthropic’s own Vision API documentation lists image inputs and does not list a native video format [6]. Claude’s web search tool returns cited sources, but the vendor documentation does not describe native video processing [6].

My read is that when Claude does cite a video source, it reaches it through a written companion or a third-party page that references the video, not only through the video’s own transcript. The data pool is thin enough that I treat this as inference rather than a finding; one of Stockebrand’s observations is useful here. Brands with saturated, market-leading positions tend to appear on Claude without specific optimization, while every other brand needs a deliberate strategy [2]. The practical implication for a pre-Series B founder is the same as it is for ChatGPT. Pair the video with a text layer.

So which engines will cite your video, and which need a written companion?

Build the video for Perplexity and Google. Build the written companion for ChatGPT and Claude.

EngineShare of YouTube citations [1]How it reaches the videoWhat a founder adds
Perplexity38.7%RAG-native search; description + auto-transcript [2][4]Long description, corrected transcript, chapter titles buyers would type
Google AI Overviews36.6%Index-grounded retrieval; chapters surface by timestamp [1][5]Chapter list from 00:00, VideoObject markup, question-shaped chapter titles
Google AI Mode19.6%Query fan-out; favors chapters and specific passages [1][2]Same video stack; biography and topic sections pick it up
ChatGPT4.4%Bing-powered live search; prefers text passages [2][8]Written companion: Medium post, concept explainer, watch page
ClaudeNot reported in [1]; sparse elsewhere [2]Brave Search; conservative citation; no documented video format [2][6]Established author credentials and authoritative written companion

The parent video pillar frames the full video stack as Citable Surfaces. The cross-pillar Alexander Berger three-engine case study is the clearest single example of the stack working across three of these four engines. The sibling spoke on turning one episode into thirty content assets covers the repackaging work that gives ChatGPT and Claude something to cite.

Two warnings carry over. Ahrefs tracked 1,885 pages that added JSON-LD schema and found no major citation uplift on any AI platform [14]. SparkToro found that the chance of ChatGPT or Google’s AI giving the same brand list twice was under 1 in 100 [13]. Build the video stack because the mechanisms support it, not because any single run will confirm it.

That is also the posture we take at TwoSetAI: we turn one podcast conversation into lasting AI-search presence for AI founders whose buyers check AI before they check the website. If your category is one where YouTube already carries citation weight, the video is the asset that pays. If your buyers check ChatGPT first, the written layer is where you earn the citation. Most founders need both. When your buyers ask AI, be the answer.

Who is TwoSetAI?

TwoSetAI is one of the top AI podcast and video studios in Silicon Valley, and the team behind this article, hosted by Angelina Yang. The show has featured 100+ AI founders and leaders on how real AI products get built, sold, and trusted. Alongside the podcast, TwoSetAI runs the Founder Visibility Program that turns each conversation into the articles, clips, and pages that make its guests easy to find. Watch full episodes on YouTube at youtube.com/@TwoSetAI, and find every guest and episode at twosetai.com.

Glossary

  • Reach Path (also called: engine retrieval path, citation path). The specific text layer an AI engine traverses to cite a video (title, description, chapters, transcript, or written companion). Different per engine.
  • Parametric mode (also called: training-data response, pre-retrieval response). An LLM answering from its training data without a web search. Any cited URL is reconstructed from training patterns and can be fabricated [2].
  • Live-search mode (also called: retrieval mode, RAG mode). An LLM issuing a web search mid-answer, fetching pages, and returning cited sources [2].
  • Written companion (also called: episode companion post, text-layer companion). A text page (Medium post, concept explainer, watch page) that pairs with a video and gives text-heavy engines a passage to cite.
  • Query fan-out (also called: subquery decomposition). A technique Google uses to break a query into parallel subqueries, each retrieved separately, with results stitched into a synthesized answer [2].
  • Citable Surface (also called: retrievable video text, video citation surface). Any piece of text attached to a video that an engine can retrieve and quote: transcript, chapter titles, description, watch page.

About Angelina Yang

Angelina Yang is VP of AI, a fast.ai fellow, and the host of the TwoSetAI podcast, where she has interviewed 100+ AI founders and leaders about how real AI products get built, sold, and trusted. Find her on LinkedIn, talk to her AI clone on Delphi, watch the podcast on YouTube, or browse twosetai.com.

Frequently Asked Questions

Does ChatGPT cite YouTube videos? Rarely. ChatGPT supplied 4.4% of all YouTube citations in OtterlyAI’s 100-million-citation study, far below Perplexity’s 38.7% and Google’s combined 56.2% [1]. In our August 2026 audit, When ChatGPT did not cite a YouTube video, it reached our guests through third-party authoritative posts (founder profiles, twosetai.com pages, Medium posts) [3]. For ChatGPT, pair the video with a written companion.

Which AI engine cites YouTube the most? Perplexity, at 38.7% of all YouTube citations in OtterlyAI’s data [1]. Google’s AI Overviews was second at 36.6% and Google AI Mode third at 19.6% [1].

Why does Perplexity cite YouTube more than ChatGPT? Perplexity runs a real-time search on every query, and its retrieval reads a video’s description and auto-generated transcript [2][4]. ChatGPT’s live search runs on Bing and tends to prefer text passages from pages like Wikipedia, which supplied 12.1% of ChatGPT citations in Analyze AI’s 83,670-citation study [2].

Does Google AI Mode read YouTube transcripts? It reads text attached to the video. Peec AI documented Google AI Mode building an answer directly from a video description [4]. OtterlyAI’s data showed chapter-level citations only on Google’s AI surfaces (73% AI Overviews, 27% AI Mode) [1].

Does Claude cite YouTube videos? Rarely. Claude cites conservatively by design: its Wikipedia citation rate was 0.1% versus ChatGPT’s 12.1% in Analyze AI’s study, and it cited LinkedIn zero times across that period [2]. Anthropic’s Vision API documents image inputs and does not list a native video format [6].

What percentage of Google AI Overviews citations come from YouTube? About 23% on aggregate across queries, with Wikipedia second at 18%, per Surfer SEO’s analysis of 36 million AI Overviews and 46 million citations, as cited in Stockebrand’s O’Reilly book [2]. That is an average; your category may be higher or lower.

Do I need a written article for ChatGPT to cite my video? In our experience, yes. In our clearest three-engine case, ChatGPT named a TwoSetAI episode “Best overall” but reached it through Medium posts, not through YouTube [11]. The written companion is what ChatGPT cited.

Does Claude read video content directly? No native video format is documented in Anthropic’s Vision API [6]. Claude’s web search tool returns cited sources through Brave Search; a video that is cited is typically reached through a text page that references it [2].

Are these YouTube citation shares stable over time? No. Google AI Overviews answers for the same query change about 70% of the time, and 45.5% of citations swap between runs, per Ahrefs research in Stockebrand’s book [2]. SparkToro found that the chance of ChatGPT or Google’s AI giving the same brand list twice is under 1 in 100 [13].

References

[1] Tousseyn, R. (2026). “The YouTube Citation Study 2026.” OtterlyAI, updated March 2, 2026. https://otterly.ai/blog/youtube-ai-citation-study-2026/

[2] Stockebrand, R. (2026). Answer Engine Optimization. O’Reilly Media, July 2026 (Chapter 6, Platform Optimization, citing Surfer SEO 2025’s analysis of 36 million AI Overviews and 46 million citations; Analyze AI’s 83,670-citation study; Profound’s 680-million-citation dataset; Ahrefs’ AI Overviews volatility research; Seer Interactive’s Perplexity freshness study). https://learning.oreilly.com/library/view/answer-engine-optimization/0642572275549/

[3] TwoSetAI (2026). Guest search audit, checked August 10 to 25, 2026, incognito; screenshots on file. Internal data; protocol in the Methodology note.

[4] Rudzki, T. (2026). “YouTube in AI search: #2 most-cited domain, transcripts included.” Peec AI documentation, March 2026. https://docs.peec.ai/research/youtube-in-ai-search

[5] Google (2026). “Video understanding.” Gemini API documentation, Google AI for Developers, last updated October 9, 2026. https://ai.google.dev/gemini-api/docs/video-understanding

[6] Anthropic (2026). “Vision.” Claude API documentation, accessed October 13, 2026. https://platform.claude.com/docs/en/build-with-claude/vision

[7] Perplexity (2026). “How does Perplexity work?” Perplexity Help Center, accessed October 13, 2026. https://www.perplexity.ai/help-center/en/articles/10352895-how-does-perplexity-work

[8] OpenAI (2024). “Introducing ChatGPT search.” October 31, 2024. https://openai.com/index/introducing-chatgpt-search/

[9] YouTube (2026). “Video Chapters.” YouTube Help, accessed October 13, 2026. https://support.google.com/youtube/answer/9884579

[10] YouTube (2026). “Use automatic captioning.” YouTube Help, accessed October 13, 2026. https://support.google.com/youtube/answer/6373554

[11] TwoSetAI (2026). Alexander Berger (COO, Bolt.new) episode and three-engine check, incognito, August 2026; watch page https://www.twosetai.com/watch/alexander-berger/. https://www.youtube.com/watch?v=8C3C4Vhknx8

[12] Scanlon, K. (2026). “In Graphic Detail: LLMs keep citing YouTube in search results.” Digiday, September 23, 2026 (reports Meltwater, Scrunch, and Weglot data). https://digiday.com/marketing/in-graphic-detail-llms-keep-citing-youtube-in-search-results/

[13] Fishkin, R. and O’Donnell, P. (2026). “New Research: AIs are highly inconsistent when recommending brands or products.” SparkToro, January 28, 2026. https://sparktoro.com/blog/new-research-ais-are-highly-inconsistent-when-recommending-brands-or-products-marketers-should-take-care-when-tracking-ai-visibility/

[14] Linehan, L. (2026). “We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved.” Ahrefs, May 11, 2026. https://ahrefs.com/blog/schema-ai-citations/

[15] Google (2025). “AI features and your website.” Google Search Central, last updated December 10, 2025. https://developers.google.com/search/docs/appearance/ai-features

Methodology note. Last verified 2026-10-13. The engine share numbers in this piece come from third-party vendor studies. OtterlyAI’s YouTube Citation Study reports percentages across more than 100 million AI citations and labels its shares as directional [1]. Stockebrand’s book cites Surfer SEO, Analyze AI, Profound, Ahrefs, and Seer Interactive, each with its own methodology; I report the figures as he states them [2]. Peec AI, Scrunch, and Weglot are vendor sources with commercial interests in AI-visibility measurement; treat their work as directional [4][12]. The TwoSetAI audit [3] covered a cohort of past guests at roughly seven queries per guest (one Google name search, two each on Google AI Mode, Perplexity, and ChatGPT), in incognito windows between August 10 and 25, 2026; the recent-cohort figures are the most recent subset of that audit. The Alexander Berger results were checked in August 2026 and may differ on your run. Claude citation behavior for YouTube specifically is inferred from adjacent evidence (sparse Wikipedia and LinkedIn citation shares, absence of a native video format in Anthropic’s Vision API) rather than from a direct study of Claude YouTube citation shares [2][6]. “Reach Path,” “Citable Surface,” and the other capitalized terms are TwoSetAI’s own naming, not industry standards. Nothing here guarantees a citation, a ranking, or a business outcome.