We were wrong about YouTube and AI answer engines

What did we get wrong about video citations?

When we started researching how answer engines handle long-form video, we began with a clear working hypothesis. We expected video to be poorly served by conversational search products compared to traditional web pages. Our premise was that language models, trained and evaluated primarily on text, would naturally lean toward articles, documentation, and written guides, while video content would rarely surface in answer engine responses.

That premise was wrong. The evidence gathered across independent industry studies shows that video is cited heavily across modern conversational platforms. Video is not being bypassed.

We also ran our own citation test, on one search-grounded engine, and part of its result cuts against our original pitch as well. What it found is in The Dark Archive report.

What does the industry data actually show?

The scale of video citations across conversational platforms is documented in several independent evaluations.

According to coverage of Meltwater data reported by Digiday, YouTube was the most-cited platform across eight answer engines in August 2026.

Search-specific evaluations show a similar pattern. In an analysis by Ahrefs of domains cited in AI Overviews, YouTube led all domains by capturing 22.9% of citations. More than one in five citations in those overviews pointed directly to YouTube URLs, ahead of every other domain in that analysis.

However, that citation volume is not distributed evenly across all tools. In a study of YouTube citations across answer engines by OtterlyAI, the YouTube citations were concentrated: Perplexity accounted for 38.7% of them, Google AI Overviews 36.6%, Google AI Mode 19.6% and ChatGPT 4.4%. YouTube citations, in other words, come mostly from a few engines.

That same research revealed an encouraging pattern for independent creators. High view counts are not a prerequisite for being referenced. The same OtterlyAI citation study found that 40.83% of cited videos had fewer than 1,000 views. In our reading, this suggests that relevance matters more than popularity when an engine picks a video. Smaller channels can and do earn citations when their material answers a specific question.

Why does citing YouTube leave creators empty-handed?

If answer engines cite YouTube so frequently, it is tempting to conclude that video creators have solved their discovery problem. Why bother maintaining an independent written website when your videos are already cited?

In our reading, that conclusion overlooks who actually benefits from platform citations.

When an answer engine references a video, it provides a link back to youtube.com. The visitor who clicks that link lands on YouTube. The platform retains the user session, serves accompanying advertisements, and keeps the audience inside its own recommendation loop. The creator who produced the research gets no direct visit to a site of their own.

This creates an asymmetry in value. The creator provides the substantive knowledge that answers the query, but the platform captures the traffic and audience relationship. A creator whose presence exists entirely on YouTube remains vulnerable to platform policy shifts, monetization updates, and algorithmic changes. In our view, relying entirely on platform citations is renting visibility without building equity.

Where does video still lose ground to written text?

The second limitation is that video citation remains highly platform-dependent.

As the research demonstrates, different conversational tools treat video differently. While Perplexity and Google search surfaces cite video frequently, tools like ChatGPT quote video far less often. Creators who rely exclusively on video are visible in some tools while remaining absent from others. Our companion note on which AI engines cite video looks at the one controlled test we found, in which several assistants cited only the written versions.

In that same test, even the most video-friendly engine cited the written versions more often than the videos. A written article published on an owned domain allows text-first engines to discover, evaluate, and quote the creator directly. When a creator publishes both a video and a corresponding written essay, they become discoverable across both categories of assistants.

Technical shortcuts do not replace the text. As we showed in our analysis of why schema is not the AI citation lever many claim it to be, adding structured code to a site without substantial written content provides nothing for an engine to quote. Answer engines cite substantive explanations, not technical markup.

What should creators do next?

The discovery that answer engines cite YouTube heavily is good news for video creators, but it is incomplete on its own. It proves that the knowledge inside long-form video is valuable enough to be cited. It also proves that creators are currently leaving that citation equity on YouTube rather than capturing it for themselves.

Video and text are not rivals here. On the engines that cite video often, the two can support each other. On the engines that cited only text in the one controlled test we found, the article is the only version of your work that got quoted.

We think creators should approach their body of work with a dual strategy:

• Keep producing video. Video content is clearly recognized and cited by several leading answer engines, and smaller channels earn citations alongside large ones.
• Publish written versions on an owned domain. Creating a written archive ensures that text-first engines can quote your work while driving referral traffic directly to your own site.
• Own the audience relationship. Converting video insights into readable articles turns platform discovery into something you own.

The full breakdown of our study of creator websites, our crawler access audit and our citation test is in The Dark Archive report.

To evaluate your own channel and see how your web presence performs, run the free Dark Matter Check.