Social Rocket
YouTube is not only discovered via Google and its own recommendation loop. AI systems are increasingly citing YouTube as a source – creating an additional layer for creators and brands alongside classic search, social reach and recommendations.
New data compiled by Digiday on 23 September 2026 shows YouTube as the most-cited platform in several major AI products. An important limit applies, however: A visible YouTube citation does not automatically mean the video generated the answer's content. In some cases it is simply recommended as further reading.
YouTube appears in many AI systems
According to Meltwater data cited by Digiday, YouTube was the most-cited platform in August across eight AI products studied: Claude, ChatGPT, Copilot, Google AI Mode, Google AI Overviews, Gemini, Perplexity and Grok.
This makes YouTube more relevant for brands and creators beyond the platform itself. A well-structured video can appear not only in YouTube Search or recommendations, but also as a source in a generative answer.
This fits a development we are already seeing with creator AEO and AI citations: creators are beginning to treat recurring visibility in AI answers as an additional distribution channel and, in some cases, even a pitching channel.
YouTube is growing especially strongly in Google's AI search
Tinuiti data, based on an analysis with Profound, shows according to Digiday that YouTube's share of citations in Google AI Mode rose more than fourfold between January and April 2026. In AI Overviews, its share more than doubled in the same period.
From a platform perspective this is plausible: Google can draw on titles, descriptions, transcripts, chapters, timestamps and other context signals on YouTube. For marketers it remains important to measure these effects cleanly. A citation can bring traffic, but it does not necessarily mean the video was central to the answer.
Our overview of AI visibility and AEO measurement explains why mentions, visible citations, actual source usage and referral traffic are separate metrics.
Long-form seems easier for AI to pick up than Shorts
A particularly interesting signal concerns video format. Digiday references OtterlyAI data, circulated via a post in the developer community DEV. In this dataset, 94 percent of the observed YouTube citations went to long-form videos, while Shorts accounted for 5,7 percent. Videos between ten and 20 minutes made up 32,1 percent of citations.
These figures are not a universal YouTube benchmark and do not come directly from YouTube. As a directional indicator, they are nonetheless plausible: longer videos offer more linguistic context and can be navigated more granularly with chapter markers, timestamps, clear titles and detailed descriptions.
This does not mean creators should artificially lengthen videos. Structure matters more. A ten-minute video with clearly separated sections is easier for humans and machines to understand than 20 minutes without discernible structure.
Tutorials and explanatory content have a natural advantage
YouTube has been strong for years in tutorials, reviews, explainer videos and other how-to formats. These are exactly the kinds of content that match the questions users put to AI assistants.
For creators this yields a practical AEO logic:
- Titles should name concretely which question or problem the video solves.
- Descriptions should place relevant terms in natural context rather than consisting only of social CTAs.
- Chapters and timestamps help make individual sub-questions findable.
- Statements backed by data, sources or traceable examples give a video more verifiable context.
- Transcripts should be clear enough that core statements are not only understood visually.
This complements but does not replace standard YouTube optimisation. We explain how the platform matches viewers by interests and satisfaction in our guide to the YouTube algorithm 2026.
AI citations can disappear quickly
Another point argues against AEO hype. Scrunch data, cited by Digiday, shows for thousands of sponsored YouTube videos and millions of citation events between May and July 2026 a marked decline after first appearance: the citation rate fell from 18,4 to 10,5 percent in the observed month.
This underlines that AI discovery is volatile. New videos, new sources and changing retrieval systems can quickly displace previous citations. For creators, a single screenshot from ChatGPT is therefore not a reliable KPI.
A recurring monitoring approach makes more sense: Which topics are cited over several weeks? Which models draw on the same videos? Does this produce measurable referral traffic, brand search volume or new demand?
What brands and creators can take from this
YouTube is thus developing into more than a video platform. Its content increasingly forms part of the information layer from which AI search selects sources and refers users.
In practice, however, this does not mean "now only optimise for LLMs". The most robust strategy remains producing useful videos for people and structuring them technically and editorially so clearly that search and AI systems also understand what they are about.
For products that need explaining, tutorials, reviews and expert content in particular, this additional discovery layer may become valuable. It should, however, be viewed as a complement to watch time, audience fit and conversion – not as a replacement.