Why Video Content Is Winning AI Search: The GEO Case for Long-Form Video
Angelina Yang, Founder & Host at TwoSetAI
Last updated: 2026-10-12
TL;DR
- The split: In OtterlyAI’s study of 100 million+ AI citations, 94% of YouTube citations went to long-form video and 5.7% to Shorts [1].
- Audience is not the signal: Views, likes, and subscribers correlated with citation at about zero (r from -0.02 to -0.03), and 40.83% of cited videos had under 1,000 views [1].
- Text around the video gets read: Description length correlated at r = 0.31, and 78% of timestamped videos were cited more than once, often across 2 to 5 chapters [1].
- Brand signal: Across 75,000 brands, YouTube mentions had the strongest correlation with AI visibility of any factor Ahrefs tested, about 0.737 [2].
- Our numbers: In our August 2026 audit, every participant in a recent Founder Visibility Program cohort appeared in Google’s video results and on at least one AI engine. When ChatGPT did not cite a YouTube video, it reached guests through third-party authoritative posts (founder profiles, twosetai.com pages, Medium posts), which is why the program ships a written editorial layer alongside each recording [3].
OtterlyAI analyzed more than 100 million AI citations across ChatGPT, Google AI Overviews, Google AI Mode, Perplexity, Microsoft Copilot, and Gemini. Of the citations that pointed to YouTube, 94% went to long-form video and 5.7% to Shorts [1]. If you are a pre-Series B AI founder deciding whether to spend founder hours on video, that split is the most useful number I know, because it says which kind of video the engines read.
This is the second of five pillar articles. The first covered the Citation Gap, the space between what AI says about your company and what is true. This one asks: why does video show up so often in AI answers, and how does a founder turn one conversation into a lasting AI-search presence? I host a podcast, so I have an interest in the answer. I have included the evidence that cuts against me.
Is video really winning AI search?
On specific engines, yes. Across all citations, video is a minority surface, and the studies agree on direction more than on rank.
- OtterlyAI (March 2026): Social and video domains were 5.54% of all citations. Within that slice, YouTube held 31.8%, second to Reddit at 46.4% [1].
- Meltwater (August 2026): YouTube was the most-cited platform across eight AI platforms, including ChatGPT, Perplexity, and Gemini, as reported by Digiday [4].
- Peec AI (March 2026): Across 30 million sources, YouTube was the second most-cited domain in AI search [5].
- Surfer SEO (2025), reported in Stockebrand’s O’Reilly book on AEO: An analysis of 36 million AI Overviews and 46 million citations found that, aggregated across queries, YouTube led Google AI Overviews citations at roughly 23 percent, with Wikipedia second at about 18 percent [26]. The figure is an average across query types, so some categories cite YouTube far more or less.
- Ahrefs (December 2025): Across 75,000 brands, YouTube mentions (a brand name in a video’s title, transcript, or description) correlated with AI visibility at about 0.737, ahead of every other factor tested, including branded web mentions at 0.656 to 0.709 [2]. Ahrefs notes that correlation is not causation [2].
The studies disagree on whether YouTube or Reddit comes first. None I found ranks YouTube below second.
The engines do not treat video alike. Of all YouTube citations in OtterlyAI’s data, 38.7% came from Perplexity and 36.6% from Google AI Overviews. ChatGPT supplied 4.4% [1]. Our audit saw the same pattern.
| Engine | Share of all YouTube citations in OtterlyAI’s data | Our most recent cohort (August 2026) |
|---|---|---|
| Perplexity | 38.7% [1] | Cited via the video and the watch page [3] |
| Google AI Overviews | 36.6% [1] | Not audited separately |
| Google AI Mode | 19.6% [1] | Cited via the video and the watch page [3] |
| ChatGPT | 4.4% [1] | When ChatGPT did not cite a YouTube video, it reached guests through third-party authoritative posts (founder profiles, twosetai.com, Medium) [3] |
| Microsoft Copilot | 0.5% [1] | Not audited |
| Gemini | 0.2% [1] | Not audited |
| Google video results (name search) | Not an AI engine | Every guest in our recent cohort [3] |
The first column is a share of YouTube citations, so it reflects how many citations each engine produces as well as how much it favors video. For context, YouTube was 22.7% of Perplexity’s social-media citations and 9.5% of ChatGPT’s [1]. The second column is the pattern we saw across a recent cohort, which maps directly to our editorial build: Perplexity and Google read the video; ChatGPT reads the written companion. Both exist in the program build, which is why the editorial layer is non-negotiable.
Why do AI engines cite long-form video and skip Shorts?
Nobody outside the labs can see inside an engine, so I stay with what documentation and studies support. The consistent finding is that engines reach a video through the text attached to it.
Peec AI found that AI answers draw on both a video’s description and its auto-generated transcript, and documented Google AI Mode building an answer directly from a description [5]. Video-RAG, a NeurIPS 2025 paper, improves long-video question answering by extracting text from a video’s audio, on-screen text, and detected objects and giving it to the model alongside the frames [6]. That is a research system, not any engine’s pipeline, but it shows where the value sits.
The vendors describe different things. Google’s Gemini API accepts public YouTube URLs, processes the audio and visual streams, and takes questions about a moment by MM:SS timestamp [7]. Anthropic’s API vision guide documents image inputs and lists no video format [8], and Claude’s web search tool answers with cited sources [9]. OpenAI describes ChatGPT search as answers with links to web sources [10], and Perplexity says each answer carries numbered citations [11]. My reading, and it is a reading: only Google documents native video input. Elsewhere, and in the citation studies, a video earns its place through text a search can retrieve.
A Short gives an engine a title, a line or two of description, and almost no spoken content. OtterlyAI’s 5.7% fits that [1]. I treat Shorts as reach, not GEO.
| Shorts and short clips | Long-form video (full episode) | |
|---|---|---|
| Share of YouTube citations | 5.7% [1] | 94% [1] |
| Cited-video length | Not applicable | Largest cluster is 10 to 20 minutes at 32.1%; 17.6% run over 20 minutes [1] |
| Text an engine can retrieve | A title and a brief description | Title, long description, chapter titles, and a transcript holding the whole argument |
| Chapter-level citation | No data found | Appeared only on Google’s AI Overviews (73%) and AI Mode (27%) [1] |
| How I use it | Human discovery on social feeds | The citable asset everything else points to |
The two episodes I cite below run 53 and 58 minutes [12][13], while OtterlyAI’s cited videos cluster at 10 to 20 minutes, so length alone does not decide it. In the Alexander Berger case below, Perplexity independently recommended the 21-minute highlight [13]. That is one case plus a correlation, so I treat a tighter cut as a hypothesis.
What parts of a video can an AI engine actually cite?
I call any text attached to a video that an engine can retrieve and quote a Citable Surface. A recording has four that matter.
The transcript. YouTube warns that automatic captions can misrepresent speech because of mispronunciations, accents, dialects, or background noise, and tells creators to review and edit them [14]. Speech recognition is strong: the Whisper authors report models approaching human accuracy and robustness after training on 680,000 hours of audio with transcripts [15]. For an AI founder the weak spot is vocabulary: model names, product names, acronyms. I correct the transcript by hand and publish it as text on a page.
Chapters. The first timestamp must be 00:00, there must be at least three in ascending order, and each chapter must run at least 10 seconds [16]. In OtterlyAI’s data, 78% of timestamped videos were cited more than once, often across 2 to 5 chapters, and only 31% of cited videos had timestamp signals at all [1]. For YouTube-hosted videos, Google says you set key moments with timestamps and labels in the description, and it prioritizes the ones you set [17]. I title each chapter as the question a viewer would type, such as “Can I make money building apps with AI tools?” at 46:22 of our Alexander Berger episode [13]. I have no data that question titles beat plain labels. It is a choice, not a finding.
The description. Description length was the strongest metadata correlate at r = 0.31, against r = 0.02 for title length, and the average cited description ran 334 words [1]. Put the summary, the named people and products, the chapter list, and links in it.
The watch page. Google recommends a dedicated page for each video, where the video is the main content [18]. Our watch-page template renders the chapter list and the full transcript as plain text, with VideoObject markup that marks each chapter as a Clip [13].
Treat markup as hygiene, not a lever. Ahrefs tracked 1,885 pages that added JSON-LD schema and found no major citation uplift on any platform, plus a 4.6% decline on AI Overviews that Ahrefs says may not be due to schema [19]. Those pages were already heavily cited, so the study says nothing about pages engines have not yet seen [19]. Google says AI Overviews and AI Mode need only an indexed, snippet-eligible page, with no special optimizations [20]. I add the markup because it is cheap.
Do you need an audience for video to work in AI search?
Not for the engines, and this matters most if your company has no audience yet. In OtterlyAI’s data, views correlated with citation at r = -0.03, likes at -0.02, and subscribers at -0.03. 40.83% of cited videos had fewer than 1,000 views, and 35% of cited channels had under 10,000 subscribers [1]. Ahrefs found the number of YouTube mentions seemed to matter slightly more than their reach [2]. One caution: OtterlyAI studied videos already cited, so it shows what repeat citation looks like, not what makes a video eligible [1].
Ahrefs counts any YouTube mention of a brand in a title, transcript, or description [2]. That includes other people’s videos, so a founder not ready to host can earn the same signal as a guest on a long conversation.
People still watch, too. YouTube was the service Americans used most often for podcasts in Edison Research’s 2025 Infinite Dial, cited by 33% of weekly podcast listeners, and 51% of Americans 12 and older had watched a podcast (5,020 people, fielded January 2025) [21]. a16z’s June 2026 conversation with Marc Andreessen and Ben Horowitz covers founders building audiences by speaking directly to customers and the public [22]. The human audience and the machine audience read the same recording.
What does this look like for real guests?
In August 2026 we searched for past TwoSetAI guests, about seven queries each, in incognito windows across Google, Google AI Mode, Perplexity, and ChatGPT [3]. A guest counted as present if any query surfaced them.
- 95 percent of our guests appeared on a Google surface, and 86 percent were cited by at least one AI engine [3].
- Every participant in a recent Founder Visibility Program cohort appeared in Google’s video results and on at least one AI engine. Google AI Mode and Perplexity leaned on the video and the watch page. When ChatGPT did not cite a YouTube video, it reached guests through third-party authoritative posts (founder profiles, twosetai.com pages, Medium posts) [3]. That is why the program ships a written editorial layer alongside each recording.
- In an earlier cohort, a few participants had no AI-engine citation [3].
No paid spend or coordinated amplification sat behind these results [3]. The recent cohort did better than an earlier one on every surface despite less time to be crawled. That points at packaging more than age, but it is not a controlled comparison [3].
The clearest single case is Alexander Berger, COO of Bolt.new. In August 2026, in incognito windows, ChatGPT named our episode “Best overall” and ranked it above interviews with Bolt’s CEO on other shows, citing Medium posts. Perplexity called it “the best fit” and independently recommended the 21-minute highlight. Google AI Mode cited the interview across multiple sections of his bio [13]. Behind it sat the full episode and a highlight on YouTube, Medium posts, Reddit posts, and a watch page [13]. One organic LinkedIn post reached 184,361 impressions with no paid promotion [13]. Those answers were reproducible in August 2026; yours may differ, so check rather than trust me.
Notice the split: ChatGPT’s answer came through written Medium posts, while Perplexity and Google cited the interview itself. Berger told me why this matters for founders: “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” [13]. Hear it on YouTube or read the write-up. Building is cheap. Being findable is the work.
Why does a founder’s spoken story beat a written summary?
A long conversation holds things a landing page rarely does: a specific story, a sequence, a named failure. Hector Hernandez, co-founder and COO of Speakeasy, told me about a salesperson on his team who set up an AI assistant, told it never to send an email without her knowing, and found it had emailed an existing customer anyway. “Even though I very much said you are never allowed to send an email without me knowing… it still went ahead and did it,” he said. When she asked, the assistant agreed it had been told not to. “It was kind of one of these things where scolding a puppy” [12]. Watch the moment or open the watch page.
No positioning deck can stand in for that. It is specific, attributable, and checkable on the recording. It is also the material GEO research rewards: the KDD 2024 paper that introduced generative engine optimization reported visibility gains of up to 40% from methods that include adding quotations, statistics, and cited sources [23]. A conversation produces all three without a copywriter. That is my reading of a paper that tested written pages, not transcripts. I call the result the Founder-Voice Moat: content in the founder’s own voice that a competitor cannot generate in an afternoon (pillar).
What does video not do?
Five limits, in the order I would worry about them.
- It does not win ChatGPT on its own. ChatGPT supplied 4.4% of YouTube citations in OtterlyAI’s data [1], and reached guests through third-party authoritative posts when it did not cite a YouTube video [3]. One practitioner reading Perplexity’s answer stream argued that Perplexity quotes video and ChatGPT cannot, from a single account he labels direction [24]. Pair every video with a written page. That is the Editorial Layer: human-authored, positioning-anchored content on top of the recording, which no monitoring dashboard writes for you (why).
- The evidence is correlational. OtterlyAI studied only videos already cited and says correlation does not establish causation [1].
- Citations fade. Scrunch tracked sponsored YouTube videos from May to July 2026 and found citations fell about 43% within a month, from a rate of 18.4% to 10.5% [4]. Plan a refresh.
- A citation is not always a source. Weglot found Claude and Gemini often embed YouTube videos as recommendations, which does not mean the model drew on the video for its answer [4].
- The output is noisy. In SparkToro’s 2,961 runs, the chance of ChatGPT or Google’s AI giving the same brand list twice was under 1 in 100 [25]. Re-run your checks. Google also does not guarantee that markup produces any specific video feature [18].
How do you turn one conversation into assets engines can cite?
This is the stack we build around one episode.
| Asset | What goes in it | Why it is there |
|---|---|---|
| Full episode on YouTube | The whole conversation | Long-form was 94% of YouTube citations [1] |
| Chapter list | Question-shaped titles, first at 00:00 | Chapter-level, repeat citations on Google’s AI surfaces [1]; Google prioritizes the key moments you set [17] |
| Long description | Summary, named entities, chapters, links | Strongest metadata correlate, r = 0.31 [1] |
| Edited transcript | Human-corrected, published as text | Auto-captions can misrepresent speech [14] |
| Watch page | Video, chapters, transcript, VideoObject markup | Google’s recommended structure [18]; markup is hygiene, not a lever [19] |
| Written companion | A quote-led article from the episode | Quotations lift visibility [23]; ChatGPT’s Berger answer cited Medium posts [13] |
| Mentions elsewhere | Medium, Reddit, and LinkedIn posts that point back | Branded web mentions correlate with AI visibility at 0.656 to 0.709 [2] |
The written companion can be a concept explainer, one of the page types in the 5-Category Architecture (pillar). One step sits before all of it. TwoSetAI runs proprietary positioning steps for every founder guest, drawing on classical PMM frameworks and adapted for AI-search citation. Those steps decide which claims the conversation must land. A recording that never says them gives an engine nothing specific to quote.
Should a pre-Series B founder invest in video?
Yes, if three things are true.
- Your buyers look you up first. Buyers, investors, and candidates run what I called the Counterparty Check in the first article: a quiet lookup in an AI assistant before they take your call. A recorded explanation of your category is something that lookup can find.
- You can talk for 45 minutes with stories and numbers. The citable material is specific, so general talk gives an engine little to quote.
- You will do the text layer. Chapters, a corrected transcript, a long description, and a written companion. The video alone is half the job.
Not yet, if you want reach and not retrieval, plan to count Shorts as GEO, or need a guaranteed citation. Nobody can promise one, because the lists change from run to run [25]. If you are not ready to host, be a guest on someone else’s long conversation, which Ahrefs counts as a mention [2].
So why is video winning AI search, and what should you do this week?
Because the engines that cite video most, Perplexity and Google’s AI Overviews and AI Mode [1], appear to reach it through the text attached to long, structured videos [5], and a long conversation with chapters and a corrected transcript supplies more specific, quotable text than most landing pages do. It does not need an audience [1]. It does not win ChatGPT by itself [1][3]. Video is a retrieval surface engines already read, and the written layer makes you the quotable passage.
This week:
- Open your latest long-form video. Check for at least three chapters starting at 00:00, a description of a few hundred words, and a human-corrected transcript on a page.
- Search your company name and category in Perplexity and Google AI Mode, in a private window, a few times each. Note whether a video appears and whose it is.
- If no long conversation about your product exists, book one.
That is what we do at TwoSetAI: we turn one podcast conversation into lasting AI-search presence for AI founders whose buyers check AI before they check the website. When your buyers ask AI, be the answer.
Next in this series: 94% of AI Citations of YouTube Go to Long-Form: What That Means for Founders, How AI Engines Parse Video, and From One Episode to Thirty Content Assets.
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
- Citable Surface (also called: retrievable video text, video citation surface). Any piece of text attached to a video that an AI engine can retrieve and quote: the transcript, chapter titles, the description, and a watch page.
- Chapters (also called: timestamps, key moments, video chapters). Timestamped sections listed in a YouTube description that split a video into labeled parts. Google’s AI Overviews and AI Mode are where OtterlyAI saw chapter-level citations [1].
- Watch page (also called: video landing page, video page). A page on your own site where one video is the main content, with its chapters and transcript as text [18].
- The Citation Gap (also called: AI-answer gap, brand invisibility in AI search). The space between what AI says about your brand and what your business actually is.
- Founder-Voice Moat (also called: founder-voice advantage). Content in the founder’s own voice that reads like a real person, not a template.
- Editorial Layer (also called: GEO editorial layer). Human-authored, positioning-anchored content on top of automated AI-visibility monitoring.
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 video content help with AI search visibility? Yes, on specific engines. In OtterlyAI’s study of 100 million+ AI citations, 38.7% of YouTube citations came from Perplexity and 36.6% from Google AI Overviews, while ChatGPT supplied 4.4% [1]. Ahrefs found YouTube mentions had the strongest correlation with AI visibility across 75,000 brands, about 0.737 [2].
What is the difference between video SEO and text SEO for AI search? Engines reach a video through the text attached to it, so the work is mostly text work: a long description, chapters, and a corrected transcript [5][1]. Google says a page needs only to be indexed and snippet-eligible to appear in AI Overviews or AI Mode [20]. A written companion still matters, because ChatGPT cites video least [1].
How do I get my YouTube video cited by ChatGPT, Perplexity, and Google AI Mode? Publish a full-length video with at least three chapters starting at 00:00, a long description, and a human-corrected transcript [16][14]. Add a written companion for ChatGPT. Nothing guarantees a citation, and answers vary run to run [25].
Are YouTube Shorts or long-form videos better for AI citation? Long-form. In OtterlyAI’s data, 94% of YouTube citations went to long-form video and 5.7% to Shorts [1]. Use Shorts for human reach.
Do I need a big YouTube channel to be cited by AI? No. Views, likes, and subscribers each correlated with citation at about zero, and 40.83% of cited videos had under 1,000 views [1]. That study only covered videos already cited, so it does not show what gets a video cited first [1].
Does a podcast help with GEO? A video podcast can. A long conversation supplies a transcript, chapters, and quotable stories. YouTube was the service Americans used most often for podcasts in Edison Research’s 2025 survey [21]. Publish the transcript as text and add a written companion.
Do YouTube chapters and timestamps help AI citations? They appear to. In OtterlyAI’s data, 78% of timestamped videos were cited again, often across 2 to 5 chapters, but only on Google’s AI Overviews and AI Mode [1]. Google says it prioritizes the key moments you set in a YouTube description [17].
Does ChatGPT cite YouTube videos? Less than Perplexity or Google’s AI surfaces do. ChatGPT supplied 4.4% of YouTube citations in OtterlyAI’s data [1], and it reached guests through third-party authoritative posts when it did not cite a YouTube video [3].
Does VideoObject schema markup improve AI citations? Not by much. Ahrefs tracked 1,885 pages that added JSON-LD and found no major uplift on any platform [19]. Google documents the markup for video features, so I add it as hygiene [17].
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] Linehan, L. (2025). “Top Brand Visibility Factors in ChatGPT, AI Mode, and AI Overviews (75k Brands Studied).” Ahrefs, December 12, 2025. https://ahrefs.com/blog/ai-brand-visibility-correlations/
[3] TwoSetAI (2026). Guest search audit, checked August 10 to 25, 2026, incognito, screenshots on file. Internal data; protocol in the Methodology note.
[4] 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/
[5] 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
[6] Luo, Y. et al. (2025). “Video-RAG: Visually-aligned Retrieval-Augmented Long Video Comprehension.” NeurIPS 2025; arXiv:2411.13093. https://arxiv.org/abs/2411.13093
[7] 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
[8] Anthropic (2026). “Vision.” Claude API documentation, accessed October 12, 2026. https://platform.claude.com/docs/en/build-with-claude/vision
[9] Anthropic (2026). “Web search tool.” Claude API documentation, accessed October 12, 2026. https://platform.claude.com/docs/en/agents-and-tools/tool-use/web-search-tool
[10] OpenAI (2024). “Introducing ChatGPT search.” October 31, 2024. https://openai.com/index/introducing-chatgpt-search/
[11] Perplexity (2026). “How does Perplexity work?” Perplexity Help Center, accessed October 12, 2026. https://www.perplexity.ai/help-center/en/articles/10352895-how-does-perplexity-work
[12] TwoSetAI (2026). Hector Hernandez (Co-founder and COO, Speakeasy) episode, 52:59, published October 5, 2026; quote at 8:14; watch page https://www.twosetai.com/watch/hector-hernandez/. https://www.youtube.com/watch?v=yhOLsoqSSEY&t=494s
[13] TwoSetAI (2026). Alexander Berger (COO, Bolt.new) episode, 57:51, and three-engine check, incognito, August 2026; chapter at 46:22, quote at 46:28; watch page https://www.twosetai.com/watch/alexander-berger/. https://www.youtube.com/watch?v=8C3C4Vhknx8&t=2788s
[14] YouTube (2026). “Use automatic captioning.” YouTube Help, accessed October 12, 2026. https://support.google.com/youtube/answer/6373554
[15] Radford, A. et al. (2022). “Robust Speech Recognition via Large-Scale Weak Supervision.” OpenAI; arXiv:2212.04356, December 6, 2022. https://arxiv.org/abs/2212.04356
[16] YouTube (2026). “Video Chapters.” YouTube Help, accessed October 12, 2026. https://support.google.com/youtube/answer/9884579
[17] Google (2026). “Video (VideoObject, Clip, BroadcastEvent) structured data.” Google Search Central, last updated September 24, 2026. https://developers.google.com/search/docs/appearance/structured-data/video
[18] Google (2025). “Video SEO best practices.” Google Search Central, last updated December 18, 2025. https://developers.google.com/search/docs/appearance/video
[19] 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/
[20] Google (2025). “AI features and your website.” Google Search Central, last updated December 10, 2025. https://developers.google.com/search/docs/appearance/ai-features
[21] Edison Research (2025). “The Infinite Dial 2025.” March 20, 2025 (survey of 5,020 people age 12+, fielded January 2025). https://ssrs.com/insights/the-infinite-dial-2025/
[22] Andreessen Horowitz (2026). “The New Rules of Media” (Marc Andreessen, Ben Horowitz, Gaby Goldberg, and Erik Torenberg). a16z Podcast, June 19, 2026. https://a16z.com/podcast/the-new-rules-of-media-marc-andreessen-ben-horowitz/
[23] Aggarwal, P. et al. (2024). “GEO: Generative Engine Optimization.” KDD 2024; arXiv:2311.09735. https://arxiv.org/abs/2311.09735
[24] Mohanadasan, S. (2026). “How Perplexity Actually Picks Sources (I Read the Stream, Not the Answers).” Search Engine Journal, July 29, 2026. https://www.searchenginejournal.com/how-perplexity-actually-picks-sources-i-read-the-stream-not-the-answers
[25] 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/
[26] Stockebrand, R. (2026). Answer Engine Optimization. O’Reilly Media, July 2026 (cites Surfer SEO’s 2025 analysis of 36 million AI Overviews and 46 million citations). https://learning.oreilly.com/library/view/answer-engine-optimization/0642572275549/
Methodology note. Last verified 2026-10-12. Engine behavior comes from each vendor’s own documentation, dated above. Third-party studies are cited with their dates, and several (OtterlyAI, Ahrefs, Peec, Meltwater, Scrunch, Weglot) come from companies that sell AI-visibility, analytics, or marketing products, so treat them as directional. Meltwater, Scrunch, and Weglot figures are as reported by Digiday and were not checked against the original reports. OtterlyAI’s write-up reports 94% of YouTube citations as long-form and 5.7% as Shorts and does not publish a minute cutoff for “long-form”; I use the word to mean a full-length video as opposed to a Short. Its engine percentages in the first table are shares of all YouTube citations, not each engine’s own citation mix, and they reflect citation volume as well as preference. The TwoSetAI audit [3] covered a cohort of past guests, not the full catalog: about seven queries per guest (one Google name search, and two each in Google AI Mode, Perplexity, and ChatGPT), run in incognito windows between August 10 and 25, 2026. A guest counted as present on a surface if any query for it surfaced them. The recent-cohort figures in this article are the most recent subset of that audit. The recent-versus-older comparison is directional, not controlled. The Alexander Berger results were checked in August 2026 and may differ on your run. Guest quotes come from the edited episode transcripts, with repeated filler words removed and ellipses where words are omitted. The Perplexity answer-stream observation comes from one account in one city, and its author labels it direction, not measurement. “Citable Surface,” “Counterparty Check,” “Founder-Voice Moat,” “Editorial Layer,” and the other capitalized terms are TwoSetAI’s own terms, not industry standards. Nothing here guarantees a citation, a ranking, or a business outcome.