If you follow GEO discussions, you’ve probably heard the claim: YouTube is one of the most cited sources in AI search, so every brand should be making videos.
I’ve spent the past three weeks tracking 20 custom prompts across different AI search tools. Three weeks is too short for big conclusions, but one pattern kept repeating, and it doesn’t quite match the common advice.
Short answer: In my 20-prompt tracking sample, YouTube appeared as a direct source only in Google’s AI search. Larger datasets show a more nuanced picture: Google and Perplexity cite YouTube heavily, while ChatGPT uses it much more selectively and Gemini rarely does. YouTube can support AI search visibility, but its value is highly platform-dependent.
What I tested
This is an ongoing measurement, not a one-off experiment. The setup: 20 custom English-language prompts in the SEO, GEO and AI marketing niche, run once a week across four tools: Google’s AI results (AI Overviews and AI Mode), ChatGPT, Claude and Gemini. Perplexity joins as the fifth from the next round onward. Search-enabled modes are used where available.
At the time of writing, the tracking has been running for three weeks (July–August 2026), so each prompt has been run three times per tool. That’s a short baseline, and I’ll say so again below. One more caveat worth stating up front: AI products change their retrieval systems and model versions constantly, so these are observations from a specific window, not permanent truths.
Part of the prompt set asks for direct recommendations. Two prompts in particular should surface YouTube if anything does:
- Best YouTube channels to learn GEO / AI SEO
- Best GEO (generative engine optimization) experts to follow
Here’s what happened:
| AI surface | Cites YouTube directly? | What it cites instead |
|---|---|---|
| Google (AI Overviews / AI Mode) | Yes. Videos embedded in the answer itself; for the channel prompt, 4 of 9 cited sources were direct links to YouTube videos | — |
| ChatGPT | No | Articles that list YouTube channels |
| Claude | No | Articles that list YouTube channels |
| Gemini | No | Articles that list YouTube channels |
In Google, both prompts produced answers with YouTube videos embedded directly in the response. In the channel recommendation prompt, nearly half of the cited sources (4/9) pointed straight to YouTube videos.
In ChatGPT, Claude and Gemini, the same prompts were answered from listicle articles about YouTube channels, not from YouTube itself. The same pattern held across the rest of my prompt set: YouTube videos were essentially absent everywhere except Google.
The most interesting row in that table is Gemini. It’s Google’s own product, and in my sample it still didn’t cite YouTube. Whatever drives YouTube’s visibility, it’s not “Google the company.” It’s specifically the AI surfaces built into Google Search.
What the large-scale data says
Because my sample is small, I checked whether bigger datasets show the same thing. They mostly do, with one important nuance.
Otterly.ai’s YouTube Citation Study 2026 analyzed over 100 million AI citations across six AI search platforms. The headline numbers put the “YouTube is huge in AI search” narrative in perspective:
- Social media and video platforms combined account for only 5.54% of all AI citations. Brand domains (52.2%) and news media (20.3%) dominate.
- Within that social slice, Reddit leads with 46.4% and YouTube takes 31.8%.
More importantly, YouTube citations are heavily concentrated on specific platforms. Otterly’s breakdown of where YouTube citations come from:
- Perplexity: 38.7%
- Google AI Overviews: 36.6%
- Google AI Mode: 19.6%
- ChatGPT: 4.4%
- Microsoft Copilot: 0.5%
- Gemini: 0.2%
Google’s search surfaces and Perplexity generate over 90% of all YouTube citations in that dataset.
The nuance comes from BrightEdge’s analysis of how Google AI Overviews and ChatGPT use YouTube differently. Google AI Overviews cites YouTube at roughly 30x the absolute volume of ChatGPT. But ChatGPT’s YouTube use isn’t zero; it’s selective. About 60% of ChatGPT’s YouTube citations are tied to instructional, how-to queries, nearly three times the instructional share seen in Google AI Overviews. For ChatGPT, YouTube isn’t a general source. It’s specifically where it points people who want to learn how to do something.
My own observations match the large datasets for Gemini and mostly for ChatGPT; my prompt set simply may not have included the kind of how-to queries where ChatGPT reaches for video. Perplexity is a blind spot in my data. I hadn’t included it in this prompt set, and based on Otterly’s numbers it’s the most interesting platform I missed. It’s now in the tracking rotation.
What this doesn’t prove
An honest caveat before conclusions. My own data so far is 20 prompts, run weekly for three weeks, in one language and one niche. That’s an early baseline from an ongoing measurement, not a study. It can’t tell you:
- whether these patterns hold across industries or languages
- whether the platforms will behave the same next month, since citation behavior changes constantly
- anything about why the models behave this way
And the large-scale studies come from GEO tool vendors, who have an obvious interest in the topic being important. The datasets are real and large, but keep the incentive in mind.
What the combination does support: YouTube’s AI visibility is real, but it’s concentrated in Google Search’s AI surfaces and Perplexity, with ChatGPT using it narrowly for instructional queries. “AI search” is not one channel.
The most useful finding from the larger dataset: popularity barely matters
If you make videos, the most encouraging part of Otterly’s dataset isn’t about platforms at all. It’s this:
View counts, likes and subscriber numbers showed near-zero correlation with citation frequency (r ≈ -0.03). In fact, 40.8% of AI-cited videos had fewer than 1,000 views, and 35% of cited channels had under 10,000 subscribers.
One important limitation Otterly states itself: the dataset consists of videos that were already cited, so the findings best explain repeated citation behavior, not why a video got picked up in the first place. With that in mind, the accurate version of the takeaway is: popularity doesn’t seem to predict how often a cited video keeps getting cited. The data is more consistent with reference value and structure mattering than with traditional popularity signals.
For smaller creators, that’s good news either way. A small channel with well-structured content can keep earning citations alongside much larger ones.
How to make videos AI systems can actually cite
So what should you do if video is part of your content strategy? Based on the citation data, here’s what appears to move the needle.
1. Make long-form reference content, not clips
94% of AI citations in Otterly’s data went to long-form videos; Shorts got 5.7%. The most cited length cluster was 10–20 minutes. AI systems favor videos that behave like references: tutorials, explainers, walkthroughs, comparisons, case studies.
Think of it this way: would this video work as a source someone could quote? A 30-second clip rarely would. A structured 15-minute tutorial does.
2. Prioritize how-to and demonstration topics
BrightEdge’s analysis of which queries trigger YouTube citations found three dominant categories:
- Instructional content: how-to queries are the single biggest driver
- Visual demonstrations: techniques, software walkthroughs, physical demonstrations
- Product reviews and comparisons: roughly a quarter of cited videos
These are also the queries where video has a natural advantage over text: showing how to do something beats describing it. And notably, instructional content is the one category where even ChatGPT reaches for YouTube.
3. Add chapters: they multiply your citations
This is the most concrete, most overlooked tactic. In Otterly’s data, Google’s AI surfaces treat a chaptered video not as one source but as several: 78% of timestamped videos were cited multiple times, typically from 2–5 different chapters. All timestamped citations came from Google AI Overviews (73%) and AI Mode (27%). No other platform cited video segments.
Chapters work like subheadings on a webpage. YouTube’s own video chapters guidelines spell out the formatting requirements:
- first timestamp at 00:00
- at least three timestamps in ascending order
- each chapter at least 10 seconds long
- chapter titles that describe the content, like good H2s
If you also publish videos on your own site, Google Search Central’s video structured data documentation covers the separate key moments markup for web pages.
One well-structured 15-minute video with clear chapters can earn more citations than five separate short videos.
4. Write descriptions for machines
Description length was one of the few metadata factors that correlated with repeat citations (r ≈ 0.31). Treat the description as machine-readable metadata:
- a plain-language summary paragraph
- the key entities, tools and terms the video covers
- links to related resources
- a correctly formatted chapter list
5. Build evergreen references, then keep them current
Otterly found a weak positive correlation (r ≈ 0.3) between recency and citation frequency. That doesn’t mean chasing upload frequency. It means the winning pattern is an evergreen reference video that gets maintained: update the description and chapters when facts change, add an “updated for 2026” note, and keep your best reference videos current instead of publishing new ones at random.
In fast-moving niches, AI tools being the obvious example, a video whose facts have expired quietly drops out of citations.
The conclusion: YouTube is not a universal GEO channel. It’s a highly platform-specific one.
Here’s how I’d summarize both my own observations and the large-scale data, platform by platform:
| Platform | YouTube GEO opportunity |
|---|---|
| Google AI Overviews | High |
| Google AI Mode | High |
| Perplexity | High (in large-scale data; not yet in my own tracking) |
| ChatGPT | Low overall, but real for instructional and how-to queries |
| Gemini | Very low in current citation datasets |
| Claude | Insufficient evidence; absent in my sample |
Today, the strongest opportunity is Google Search and Perplexity. ChatGPT uses YouTube much more selectively, mostly for how-to content, while Gemini barely cites it in the large-scale dataset I reviewed.
If you want visibility in ChatGPT, Claude or Gemini for anything other than instructional queries, text is still doing the work. Those models answer from articles, including, somewhat ironically, articles that list YouTube channels. If you want to be the source for those answers, write the article.
If your goal is visibility in Google’s AI surfaces or Perplexity, video is a real and underused lever, especially structured, chaptered, long-form how-to content. Since Google still handles the overwhelming majority of search traffic, that’s not a small prize.
Treating “AI search” as one uniform channel is the actual mistake — YouTube just makes it unusually easy to see.
This is an ongoing measurement
The tracking continues weekly: 20 prompts, five platforms from here on (Perplexity joins the rotation), same questions every week, watching what actually changes. I’ll publish what shifts, including the findings that contradict this one.
If you’re seeing different patterns in your own data, I’d like to hear about it.
Read more:
SEO vs. AEO vs. GEO: What’s Actually Different (and What Isn’t)
Do FAQs Matter for AI Search? I Built an FAQ Generator, Then Checked the Evidence