Xiaohongshu's Video Pivot Could Quietly Break the AI Citation Pipeline Brands Rely On

Xiaohongshu just announced it has crossed 400 million monthly active users and over 8 billion daily searches, and it secured broadcast rights for the 2026 World Cup across the US, Canada, and Mexico. On the surface, that's a growth story. Underneath it, the platform is making a structural bet that matters far more to brand marketers than the headline numbers: it is deliberately shifting its content mix away from the short, text-heavy "notes" format that made it China's most-cited AI answer-engine source for product and brand queries, and toward mid-to-long-form video, podcasts, and 4K content.

That shift matters because Xiaohongshu (RED) is not just a discovery platform — it is raw material. Independent citation-source analysis of Chinese AI engines shows Xiaohongshu content getting cited in luxury and lifestyle brand answers at rates ranging from 17% to 96% depending on the product category and price tier, second only to the deal-and-review site SMZDM (80–99%) and Zhihu (32–98%). If the underlying content Xiaohongshu produces moves from scannable text to video, the question for GEO teams is simple: can DeepSeek, Qwen, Doubao, and Kimi still extract citable, entity-level facts from it — and if not, whose brand visibility quietly erodes first?

What Xiaohongshu is actually changing

The platform's 2026 roadmap, reported by Huxiu, lays out a deliberate pivot away from short-video completion-rate optimization:

• The recommendation algorithm has been redesigned to de-prioritize video completion rate as a ranking signal, extending the content discovery window from a matter of days to roughly 90 days.
• Xiaohongshu is allocating 5 billion traffic units in support across 15 content verticals to seed longer-form video, written long-form articles, and podcasts.
• Creator funding includes grants of up to ¥1 million per project for long-form video work, aimed at filling what the company describes as a three-to-four-year gap in "full-network" long-form creators.
• A YouTube-style video ad revenue-sharing model is in testing, alongside continued growth of "reverse funnel seeding" — brands targeting small, high-intent niche audiences rather than buying mass reach.
• Roughly 100,000 creators already operate as dual creator-and-sales-agents, tying content directly to transaction data.

None of this frames itself as an AI-search story. Xiaohongshu's own positioning is that AI lowers the barrier to content production, which is precisely why it feels safe accelerating into video now. But that framing is about creators, not about how downstream AI systems consume the platform's output.

Why this is a GEO problem, not just a content-strategy one

Chinese AI engines don't browse Xiaohongshu the way a human does. They rely on indexed, extractable text — captions, pinned comments, structured note bodies — to pull specific claims ("this serum reduced redness in two weeks," "this brand ranked in the top three for durability") that they can cite or paraphrase in an answer. That is exactly the content type Xiaohongshu built its citation weight on, and exactly the content type a video-first strategy structurally deprioritizes.

The risk isn't hypothetical. Source-citation analysis of DeepSeek and Qwen shows the two engines already handle sourcing very differently, which suggests they'll respond differently to a supply-side shift like this one:

The category data reinforces where this bites hardest. In the mass-luxury tier (goods under roughly ¥10,000), SMZDM dominates citations at near-universal rates and Xiaohongshu still plays a meaningful secondary role. But at the ultra-luxury tier (¥30,000 and up — Hermès-level goods), SMZDM's dominance already collapses to just 33% on DeepSeek, and citations shift toward forums, trade press, and auction-house archives. That tier is precisely where text-native, detail-rich Xiaohongshu notes — ingredient breakdowns, authentication tips, comparison threads — currently punch above their weight. Push that content into video, and the ultra-luxury citation gap that's already forming could widen faster than brands are tracking.

The "reverse funnel" wrinkle

Xiaohongshu's other big lever — reverse funnel seeding, where brands buy precision reach into small, high-intent niches rather than broad coverage — compounds the citation risk rather than offsetting it. Reverse funnel campaigns are designed to concentrate influence in a small number of deeply engaged creator accounts and comment threads, which is good for human conversion but bad for AI extraction diversity. If an AI engine's citation pipeline for a category depends on breadth of independent text sources to triangulate a claim (several notes independently mentioning the same ingredient or feature), a strategy that funnels engagement into fewer, more concentrated accounts can actually shrink the pool of independently citable text even before video enters the picture. Brands running reverse-funnel campaigns on Xiaohongshu should treat GEO citation tracking as a required companion metric, not an afterthought to conversion reporting.

There's also a role-mixing effect worth flagging: with roughly 100,000 creators now operating as both content creators and sales agents, more Xiaohongshu content is being produced with a direct transaction incentive baked in. That tends to push copy toward calls-to-action and away from the comparative, fact-dense language ("contains X% retinol," "scored higher than Y in independent testing") that AI systems are most likely to lift into a cited answer. A platform optimizing harder for commerce conversion and video engagement simultaneously is optimizing away from the exact content shape that built its AI-citation relevance in the first place.

How to measure the shift instead of guessing at it

Brand teams don't need to wait for a full year of MAU and citation reports to know whether this is hitting them. A workable tracking approach:

1. Pull a same-query baseline monthly. Run the same 10–15 category queries against DeepSeek, Qwen, Doubao, and Kimi every month, and log which domains and platforms get cited, not just which brands appear.
2. Tag citations by content format where visible. When a cited Xiaohongshu source is identifiable, note whether it reads as a text note or references video/podcast content. Early declines in text-note citation share, even before overall Xiaohongshu citation volume drops, are the leading indicator.
3. Cross-check against SMZDM and Zhihu share. If Xiaohongshu's citation share within your category drops while SMZDM and Zhihu hold steady or rise, that is consistent with a format-driven supply shift rather than a general loss of relevance for your brand.
4. Segment by price tier. Because ultra-luxury citation patterns already lean on forums and trade press rather than Xiaohongshu, the video pivot's impact will show up first and hardest in the ¥10,000–¥30,000 range, where Xiaohongshu notes currently do the most citation work.

The MAU context that makes this more urgent

This is landing at a moment when audience concentration in Chinese AI is already extreme. Doubao's monthly active users (382 million) exceed the combined total of the next several competitors, DeepSeek sits around 130 million with the highest engagement depth of any assistant, and Kimi has fallen to just 7.29 million MAU — despite comparable session-length retention to Doubao, because, as one Chinese analyst put it, "paper reading is a low-frequency need; nobody reads papers every day." Qwen holds a meaningful middle position, reported at roughly 170 million MAU in the most recent QuestMobile data.

In other words, brand GEO budgets are already being forced to concentrate on Doubao and DeepSeek because that's where the users are. If Xiaohongshu's video pivot also thins out DeepSeek's citation supply specifically — the engine most dependent on a narrow, high-authority source set — brands could be facing a double squeeze: fewer engines worth prioritizing, and a shrinking text-content supply feeding the most citation-concentrated one.

What brand marketers should do now

Takeaway for Brand Marketers:

1. Baseline your Xiaohongshu-sourced citations today. Before the format shift accelerates, check how much of your current brand visibility on DeepSeek and Qwen traces back to Xiaohongshu text notes versus other sources. This is your before-picture.
2. Don't let video content go caption-less. If your Xiaohongshu strategy is moving into long-form video and podcasts along with the platform, make sure every piece carries a full transcript, structured chapter markers, and pinned-comment summaries that restate specific, citable claims in plain text — the format AI extraction pipelines still favor.
3. Rebalance toward SMZDM and Zhihu now, not later. Both remain overwhelmingly text-based and already carry higher citation weight than Xiaohongshu in the highest-value categories. Treat them as the hedge while Xiaohongshu's content mix is in flux.
4. Watch DeepSeek citation share specifically over the next two quarters. Because DeepSeek concentrates citations in a small source set, it will show supply-side stress from Xiaohongshu's shift faster and more visibly than Qwen will — making it the best early-warning signal.
5. Don't assume Doubao is immune. Its Douyin-linked video-parsing advantage may let it absorb Xiaohongshu's pivot more gracefully, but that also means brands who've under-invested in Doubao-specific GEO now have a widening reason to catch up.

The platforms brands optimize for citations on are not static, and neither is the format those platforms reward. Xiaohongshu's video pivot is a business decision made for creators, monetization, and World Cup-scale audience growth — but it will land on GEO teams as a supply shock to the exact content type their brand-visibility strategy has been built on.

Related: see current brand-by-model visibility scores on the GEO Hub brand tracker.