YouTube AI visibility: 20%, despite the geo-block.

The geo-block paradox

YouTube has been blocked in Mainland China since 2009. A consumer searching DeepSeek from Shanghai cannot click a YouTube link in the answer and watch the video. By any obvious commercial logic, the engine should not surface YouTube — the citation is not actionable for the user.

But the engine does surface YouTube, at 20% of responses. Three explanations are consistent with the data:

• Training-time exposure. DeepSeek's training corpus included Western web data fetched before the engine was deployed in production. YouTube content — channel descriptions, video titles, transcripts where available, comment-thread summaries — was part of that corpus. The engine learned that YouTube channels and videos are common artefacts of the brand-discovery and product-review information landscape, and surfaces them at retrieval time even when end-users cannot directly access them.
• Cross-language re-publication. YouTube content is routinely mirrored, transcribed and discussed on Mainland-CN platforms — Bilibili clips, Weibo discussion threads, Zhihu posts that link to or describe YouTube videos. The engine may be citing the original YouTube URL because it is the canonical anchor for content that has been re-published across the substrate it can access.
• Citation hygiene. Engines tend to credit primary sources rather than the platforms that re-host them. If an engine knows a piece of content originated on YouTube, it will preferentially cite the YouTube URL even if the consumer accessed the content through a different platform.

What this means for global brands

The geo-block paradox has a practical implication: YouTube content is part of the AI-citation surface for Mainland-CN consumers even though it is not part of the consumer's direct browsing surface. A brand that ships substantive YouTube content — product reviews, channel-owned explainers, third-party tutorials, executive interviews — is contributing to the source pool that DeepSeek (and likely Qwen and Doubao) draw from when answering Mainland-CN consumer questions.

Three caveats matter:

• The citation is not actionable for the consumer. The brand cannot expect Mainland-CN viewers to click through and watch. The contribution is to the answer's evidence base, not to direct viewer traffic.
• Channel-owned content has different weight from third-party content. The 20% figure is aggregate; we have not yet decomposed it into "brand-owned channel" vs "third-party reviewer". Anecdotally, third-party review and tutorial content surfaces more often than brand-owned promotional content — engines tend to weight outside-in voices over self-claims.
• Subtitle and metadata quality matter more than production value. AI engines cannot watch a video; they parse titles, descriptions, transcripts (where available), and comment-thread structure. A well-described video with a thorough caption and accurate auto-transcript is more citable than a beautiful video with a one-line description.

What to actually ship on YouTube for Mainland-CN AI surfacing

A YouTube strategy aimed at Mainland-CN AI visibility looks different from one aimed at Western SEO or direct-viewer growth. The optimisation targets are textual, not video-craft:

1. Long-form, descriptive titles. Not "Bag Review" — "Hermès Birkin 25 review: leather wear, hardware patina, resale considerations after 5 years". The title is the strongest single signal for an engine deciding whether to cite.
2. Substantive descriptions. 300+ words covering what the video actually contains, named entities mentioned, products compared, and links to related references. The description is text the engine reads in full.
3. Accurate captions and auto-transcripts. Edit YouTube's auto-transcript for accuracy. Misspellings of brand names in the transcript propagate into AI training and retrieval. A 30-minute video with a well-edited transcript is a 30-minute searchable text document.
4. Cross-link to and from durable web pages. YouTube descriptions linking to a brand's content on its own site (not paid social, but actual encyclopedia-style pages) build the entity graph. Inverse links — from blog posts to YouTube videos — reinforce the same graph.
5. Pursue third-party coverage, not just self-published. A video on a recognised reviewer's channel about your brand is more citable than the same content on your own channel. Industry reviewers, comparison channels, professional photographers (for handbags / watches / cars), expert demonstrators (for skincare / cookware) — these are the citation targets.

Caveats and uncertainties

The 20% rate is from one engine (DeepSeek) on one category (luxury handbags) with n=1,620 responses, May 2026 panel. Reliability statistics: top-5 source-attribution κ = 1.00 (perfect across re-runs); top-15 κ = 0.46 (moderate). YouTube consistently appeared in top-5, so the headline rate is high-confidence; but specific channels appearing in long-tail positions are less stable. We have not yet decomposed the rate per engine — Qwen and Doubao panels are queued.

We are not measuring sales, conversion, or attributable revenue from YouTube-mediated AI citations. Whether brands that surface this way close more business in Mainland China is a separate question. Anecdotally — and we flag this as anecdote, not measurement — we have seen brand-discovery-stage clients report Mainland-CN buyers referencing AI-relayed YouTube content during the consideration process. That observation is consistent with the citation-surface hypothesis but not a substitute for measurement.