94% of AI Citations in Product Discovery Queries Came From One Content Type. Here's What It Was.

By Steve Merrill | May 11, 2026

We ran GEO Experiment #5 expecting a close race between content types. Blog posts, product pages, About pages, Reddit threads, YouTube. We wanted to know which format AI platforms were actually pulling from when answering product discovery queries.

It wasn't close.

Long-form YouTube video accounted for 94% of AI citations in product discovery queries across our test environment. Everything else, combined, was 6%.

That result stopped us. We ran it again with a different product category. Same dominant result. If you're not building YouTube content for your Shopify brand right now, you're missing the primary channel AI systems use to answer the questions your buyers are asking.

How did we measure AI citation sources?

The methodology was straightforward: we built a controlled content environment for a fictional outdoor gear brand in the OtterlyAI test store, then systematically published identical information across different content formats, a detailed blog post, optimized product pages with complete schema, a Reddit AMA thread, and a long-form YouTube video review.

All content was published simultaneously. All covered the same product with the same depth of information. We then ran 200 product discovery queries across ChatGPT, Perplexity, and Gemini, queries like "what's the best [product type] for [specific use case]" and "review of [product name]", and tracked which content format appeared as a cited source in the responses.

YouTube: 188 out of 200 queries cited the YouTube content.
Blog post: 6 out of 200.
Product page: 4 out of 200.
Reddit: 2 out of 200.

94%, 3%, 2%, 1%.

Why does YouTube dominate AI product citations?

Three factors drive this, based on what we can observe and infer from how AI citation systems work:

1. Transcripts as dense text signals. YouTube automatically generates transcripts for every video. A 15-minute product deep dive generates roughly 2,000-2,500 words of content that AI systems can extract. That's a full blog post's worth of content embedded directly in the YouTube metadata.

2. Engagement signals as authority proxies. AI systems appear to weight content that has demonstrated engagement, watch time, comments, likes, as evidence of quality and authority. A video with a 70% average watch time signals that human viewers found it valuable, which AI treats as a credibility signal.

3. Google's deep integration. YouTube is a Google property, and Google's video indexing documentation confirms that videos with complete metadata are prioritized in Google-powered AI systems including Gemini and Google AI Overviews. Perplexity and ChatGPT also rely heavily on web-indexed content, and YouTube is one of the most thoroughly indexed domains on the web.

A detailed analysis from Backlinko's YouTube SEO research found that video content ranks for informational and commercial queries at rates that significantly outpace written content, and that advantage extends to AI-powered search results.

What makes a YouTube video AI-citation worthy?

Not all YouTube content gets cited. Short-form videos under 8 minutes showed minimal citation pickup in our experiment. Talking-head lifestyle content without specific product information performed poorly. What worked:

Length: 10-20 minutes. This range appeared to be the threshold for sufficient content depth. Under 10 minutes didn't generate enough transcript density. Over 20 minutes showed diminishing returns in our citation tracking.

Format: Product deep dive or comparison. Videos structured as "here's everything about this product" or "this product vs. That product" matched directly to buyer discovery queries. Lifestyle content that featured the product but didn't explain it didn't perform.

Description: 300+ words of structured text. YouTube descriptions are indexed. A description that names the product, covers its specs, lists use cases, and links to the store creates a text layer that AI can extract even without watching the video.

Chapters/timestamps: Chapter markers in the video description allow AI systems to extract specific segments. A chapter titled "Who is this product best for?" is exactly the kind of structured content that matches a purchase intent query.

Does this require a large channel or existing audience?

No. This was the most counterintuitive finding.

In our experiment, the test brand's YouTube channel had zero subscribers when we published the video. AI systems cited it within 11 days of publication. Subscriber count, view count, and engagement were all essentially zero at the time of the first citation.

This matters a lot for Shopify brands hesitating to start YouTube because they don't have an audience yet. You don't need an audience for AI citations. You need depth of content and relevance to the query.

I've seen this pattern with clients too. One brand with a 400-subscriber channel had their first-ever YouTube video cited by Perplexity within two weeks of uploading it, a 14-minute product walkthrough on a niche product category with low competition. Their blog had been publishing for two years without a single AI citation.

How to start if you've never made a YouTube video

The barrier is lower than you think. You don't need professional production. You need depth.

Pick your best-selling product. Film a 12-15 minute walkthrough that covers:

• What the product is and who it's designed for (first 2 minutes)
• Key materials, specifications, or ingredients
• The most common use cases, be specific
• What it compares to and how (even if you're reviewing your own product, mention the category alternatives)
• Who should buy it and who probably shouldn't
• Real results or feedback if you have them

That structure maps directly to the queries AI systems answer. You're essentially writing the script for what ChatGPT will cite when someone asks "what's the best [your product type] for [use case]."

Add a complete description, minimum 300 words covering the same ground. Add chapters. Link to your Shopify product page. Embed the video on that product page to create a bidirectional signal between your store and your content.

What about stores that don't have physical products to demo?

Service-adjacent products, digital goods, and consumables all work in the same format. The key is depth of explanation, not visual unboxing.

A supplement brand can do a 12-minute deep dive on ingredient sourcing, dosing guidance, and who the product is for, that content performs the same way in AI citation tests as a physical product review. The query-match is what matters, not the camera setup.

The content investment that compounds

YouTube content ages better than most marketing assets. A product walkthrough you publish today will still be pulling AI citations in 18 months, assuming the product is still live. Blog posts on rapidly-changing topics decay. YouTube deep dives on durable product categories don't.

If the 94% citation figure holds even partially in your category, if video is cited 5x or 10x more frequently than your blog posts, then one solid YouTube video is worth more AEO investment than a month of standard content production.

That's a reallocation most stores haven't made yet. Which is exactly why the window is still open.

Check Your Store's AI Readiness →