In March 2026, OtterlyAI published the first large-scale study on YouTube citations in AI engines: 100 million citations analyzed over 30 days, six platforms covered. The headline finding surprised everyone.
40.83% of YouTube videos cited by AI had fewer than 1,000 views. 36% had fewer than 15 likes. Subscriber count: near-zero correlation (r = -0.03) with citation frequency.
What AI engines reward on YouTube isn't popularity. It's structure. And that completely changes the content strategy for 2026.
YouTube has overtaken Reddit as the top source for AI engines
In January 2026, Adweek published an analysis consolidating data from four independent sources. The finding: YouTube now appears in 16% of LLM responses compared to 10% for Reddit. That's a complete reversal from mid-2025, when Reddit dominated social citations from AI engines.
But YouTube didn't just overtake Reddit. According to BrightEdge, YouTube is cited 200 times more than any other video platform in AI responses, including TikTok, Vimeo, and Twitch. ChatGPT and Perplexity, which have no obligation to favor Google properties, cite YouTube almost exclusively in their video category. This isn't favoritism. It's architecture.
Google AI Overviews cites YouTube in 29.5% of its responses. Google AI Mode includes it in 16.6% of results. And YouTube's share of social citations in AI engines climbed from 18.9% to 39.2% between August and December 2025 according to Superlines, while Reddit fell from 44.2% to 20.3% over the same period.
Why LLMs read your videos but don't watch them
This is the most important point to understand. GPTBot, PerplexityBot, and AI crawlers don't read video files. They don't press play. They analyze the text surrounding your video: the transcript, the description, the title, the chapters.
In a RAG (Retrieval-Augmented Generation) pipeline, the AI engine breaks down the user's question, searches for relevant text passages in its indexed sources, and assembles a response. If your video has no accessible transcript, it doesn't exist for this pipeline. If your description is 15 words, the model has nothing to extract. If your chapters are called "Introduction" and "Conclusion", the bot has no idea what the video contains.
A poorly structured YouTube video with 500,000 views is less visible to AI than a video with 200 views whose title is a direct question, whose description summarizes key points, and whose chapters work like H2 headings in a blog post.
The finding that changes everything: views don't matter
The OtterlyAI study is the largest analysis ever published on this topic. 100 million citation instances, 30 days of collection, six AI platforms covered. Here's what it found about the YouTube videos actually cited by AI engines.
40.83% of cited videos had fewer than 1,000 views at the time of citation. 36% had fewer than 15 likes. The median cited channel had fewer than 41 total videos. Subscriber count had a correlation of r = -0.03 with citation frequency, which is statistically zero.
What this means in practice: a brand launching its YouTube channel today, publishing a 12-minute video structured with descriptive chapters on its main topic, can be cited by Perplexity or Google AI Overviews the following week. No subscribers. No favorable algorithm. No promotion budget. The only condition is structure.
A single well-structured explainer can out-cite a 500,000-subscriber channel publishing Shorts. That's the new playing field.
The 5 signals that predict AI citation on YouTube
OtterlyAI's correlation analysis precisely identifies what separates cited videos from those ignored by AI engines.
1. Description length (r = 0.31)
This is the signal most correlated with repeated citation. A description of at least 300 words that summarizes key points, uses the exact terms your audience types into LLMs, and reads like a blog post summary. Not a marketing pitch. A text document the bot can extract directly. The first 200 words get priority attention from the crawler, so put substantive information at the top.
2. Chapters with descriptive timestamps
YouTube chapters function like H2 headings for AI bots. A chapter titled "How to measure your AI share of voice" is directly citable and extractable. A chapter titled "Part 2" tells the model nothing. Only 31% of cited videos in the OtterlyAI study had chapter structure, meaning 69% of your competitors already appearing in AI responses aren't yet optimized on this signal.
3. Long-form format (10 to 20 minutes)
94% of YouTube citations in AI responses go to long-form videos. Shorts account for 5.7%, and almost exclusively on Google surfaces (AI Overviews and AI Mode). ChatGPT, Perplexity, Copilot, and Gemini have near-zero Shorts inclusion. The most cited duration bracket is 10 to 20 minutes (32.1% of citations), followed by 5 to 10 minutes (26.1%).
4. Title as a direct question
A title that mirrors the exact question your audience asks in LLMs is directly aligned with the RAG mechanism. "How to measure your AI visibility in ChatGPT in 2026" triggers extraction. "My GEO strategy for this year" triggers nothing. Think of every video title as a prompt someone types into Perplexity.
5. A clean, accessible transcript
YouTube auto-generates transcripts, but auto-generated transcripts are often poor quality: no punctuation, errors on technical terms, sentences cut at the wrong point. A manually edited or corrected transcript is structurally more extractable by LLMs. For domain-specific terms like GEO, LLM, or AI share of voice, manual correction is particularly impactful.
AI engines don't behave the same way with YouTube
OtterlyAI's data reveals significant fragmentation across platforms. Perplexity generates 38.7% of all YouTube citations in AI engines and is the most reactive to new publications. Google AI Overviews generates 36.6% and particularly values timestamped videos, as they function as multi-citation assets: 78% of timestamped videos cited by Google are cited more than once, across two to five different chapters. ChatGPT generates only 4.4% of YouTube citations, and Gemini and Copilot less than 1% each.
The strategic implication is clear. If you're starting your YouTube GEO strategy today, optimize first for Perplexity and Google AI Overviews. Those are the two engines that will generate your first measurable citations, often within days of publication for Perplexity.
7 concrete actions to optimize a YouTube video for AI engines
Every action below is standalone and directly applicable. Start with the ones that apply to your existing videos before creating new ones.
1. Title every video like a search prompt
Before publishing a video, type the planned title into Perplexity. If Perplexity generates a direct response on that topic, that's a strong signal the title aligns with what AI engines are looking for. If Perplexity can't find relevant information, the title is probably too vague or too marketing-focused.
2. Write a description of at least 300 words
The description must include the key points covered in the video, the exact terms of your domain, and a summary of the main conclusion. No slogans. No marketing call-to-action first. The bot reads the first 200 words as priority. Put substantive information at the top.
3. Structure your chapters like article H2s
Each chapter should answer a specific question. Ideal format: "00:00 Introduction", "01:30 Why YouTube is cited by AI engines", "04:15 How to optimize your description", "08:00 Mistakes to avoid". This format lets AI bots extract the relevant section directly without having to process the entire transcript.
4. Manually edit your transcript
In YouTube Studio, go to Subtitles, select the auto-generated transcript, and correct any mis-transcribed technical terms. Add missing punctuation. One hour of work on a transcript can have a measurable impact on Perplexity citations within days.
5. Make videos 10 to 20 minutes on a single topic
Resist the temptation of 2-minute "easy to consume" videos. AI engines are looking for depth. A 15-minute video that covers a topic exhaustively is structurally more citable than a series of 5 three-minute videos on the same subject.
6. Link each video to its corresponding blog post
In the YouTube description, add the link to your blog post on the same topic. And in the blog post, embed the video. This cross-linking combines the authority signals of both formats and reinforces your topical entity in both ecosystems, SEO and GEO.
7. Rework existing videos before creating new ones
If you already have a YouTube channel with videos on your main topic, start by adding chapters and improving descriptions. That's the highest effort-to-impact ratio action available. An already-indexed video, properly structured, can go from zero to regular citations without you republishing anything.
The window is still open
Only 31% of YouTube videos cited in the OtterlyAI study had chapter structure. That means among all the videos already appearing in AI responses, 69% don't yet have this basic optimization signal. Brands adding this structure now are building a lead in a space that isn't saturated yet.
It's the same window that existed for brands that understood SEO in 2012, or those that invested in blog content when competition was still low. The difference in 2026 is that the channel is new and the rules are still barely known. Brands acting on YouTube for AI engines now will be in the training data of the next LLM update cycles.
YouTube is no longer just a video platform. It's the second-largest source of AI citations, just behind websites, and the only video platform that genuinely matters for LLMs.
How to measure whether YouTube is boosting your AI visibility
The manual check is straightforward. Take the title of your most recent video and type it into Perplexity. Look at the cited sources. If your YouTube video doesn't appear, look at which source is cited instead and analyze its structure. That's your direct benchmark.
In Google Analytics 4, create a segment dedicated to referral traffic from youtube.com combined with AI traffic (chatgpt.com, perplexity.ai, claude.ai). This segment gives you an indirect signal of your YouTube presence's impact on your LLM citations over time.
The limit of this approach is the same as any manual audit: it's a snapshot. Vizible AI automatically tracks your share of voice in ChatGPT, Gemini, Claude, Perplexity, Mistral, and DeepSeek every day, including when citations come from your YouTube videos. You see precisely which sources AI engines use to cite you and how that evolves over time.
FAQ: YouTube and AI engine visibility
Are YouTube Shorts cited by AI engines?
Barely. OtterlyAI's study across 100 million citations shows Shorts account for only 5.7% of YouTube citations in AI engines. And those 5.7% are almost exclusively concentrated on Google surfaces (AI Overviews and AI Mode). ChatGPT, Perplexity, Copilot, and Gemini have negligible Shorts inclusion. For an AI visibility strategy, long-form videos of 10 to 20 minutes are the format to prioritize.
How do you know if your YouTube video is cited by Perplexity or ChatGPT?
Manual method: type the exact title of your video into Perplexity in private browsing and look at the cited sources. For ChatGPT, enable web browsing and run the same search. If youtube.com followed by your channel URL appears in the sources, your video is being cited. For systematic tracking, GA4 lets you create a youtube.com referral traffic segment that, combined with AI traffic data, gives an indirect signal of your videos' impact on LLM citations.
Does subscriber count influence AI citations on YouTube?
No. OtterlyAI's March 2026 study shows a correlation of r = -0.03 between subscriber count and citation frequency in AI engines. That's statistically zero. 40.83% of cited videos had fewer than 1,000 views. The median cited channel had fewer than 41 total videos. What matters to LLMs is content structure, not channel popularity.
What YouTube video length is most cited by AI engines?
The 10 to 20 minute bracket concentrates 32.1% of YouTube citations in AI engines, followed by 5 to 10 minutes (26.1%) and 20 minutes or longer (17.6%). Videos under 5 minutes represent a marginal fraction outside Shorts. The ideal for an AI citation strategy is a 12 to 15 minute video on one precise topic, with a clear chapter structure.