DeepSeek's 20% MAU Slide: Why Brands Should Rebalance Their China AI Bets in 2026

For eighteen months, DeepSeek was the name every global brand marketer learned first. It was the model that rattled Silicon Valley, the app that topped download charts on three continents, and — for teams building a China AI visibility strategy — the obvious place to start. That reflex is now out of date. According to QuestMobile's June 2026 native-AI-app ranking (released July 14), DeepSeek's monthly active users fell to 129.82 million, down 20.3% year over year — a rare contraction in a market that is still growing overall. Over the same window, ByteDance's Doubao reached 382.3 million MAU, a "断层第一" (tier-defining first place) that QuestMobile pegs at roughly 172% annual growth. The gap between the two is no longer a lead; it is a chasm.

For brands allocating limited GEO (generative engine optimization) budget across China's fragmented AI landscape, this single data point reorders priorities. The question is no longer "Are we visible in DeepSeek?" It is "How much of our China AI reach still runs through a platform that is shedding consumers?"

The consumer reach story and the reasoning story have split

DeepSeek's decline is easy to misread. The model itself has not gotten worse — its reasoning and coding capabilities remain among the strongest in the Chinese field, and its API continues to power a long tail of third-party apps, enterprise tools, and in-car assistants. What is shrinking is the consumer-facing DeepSeek app: the standalone destination where an ordinary shopper types "which running shoes are best for flat feet" or "recommend a reliable mid-range SUV."

That distinction matters enormously for brand marketers, because GEO visibility is only valuable where consumers actually ask brand-adjacent questions. A model that thrives as invisible infrastructure — answering coding prompts, running behind an enterprise chatbot — generates very few of the discovery moments a brand cares about. Doubao, by contrast, is where consumer intent is concentrating. Its growth is fed directly by ByteDance's Douyin flywheel: content discovery, search, and increasingly commerce all funnel back into the same recommendation surface. When 382 million people open Doubao and a meaningful share ask product questions, that is where brand citations convert into consideration.

The takeaway is uncomfortable for anyone who built their 2025 China AI plan around DeepSeek: reach and reasoning have decoupled, and brand budget should follow reach.

What the June 2026 tier looks like

Two of the three leaders are gaining consumer ground, and both — Doubao and Qwen — sit inside ecosystems (Douyin/Doubao and Alibaba's commerce stack) that connect an AI answer to a purchase. DeepSeek, brilliant but ecosystem-light, has no equivalent path from "recommended" to "bought." In a market where AI answers increasingly end in transactions, an unmonetizable consumer surface is a surface that platforms and users both drift away from.

Why DeepSeek is losing the consumer, not the developer

Three forces explain the slide, and each has a brand implication.

First, novelty normalized. DeepSeek's early MAU spike was fueled by global curiosity and a viral moment. Curiosity-driven users are the least sticky; once the novelty faded, many folded their AI usage back into whatever app was already on their home screen — for most Chinese users, that is Doubao or a super-app assistant.

Second, distribution asymmetry. Doubao is promoted relentlessly inside Douyin, one of the most-used apps in China. Qwen rides Alibaba's properties. DeepSeek has no comparable consumer distribution engine, so it must win users one download at a time — expensive, and increasingly futile against embedded rivals.

Third, the pivot to depth over breadth. DeepSeek's roadmap has leaned into reasoning, agents, and enterprise deployment. That is a defensible business — but it is a B2B business, and it pulls the brand story away from the everyday consumer prompts that generate brand citations.

For a marketer, the implication is not "abandon DeepSeek." It is "reclassify it." DeepSeek is shifting from a consumer discovery channel into a reasoning and enterprise-vendor channel. If you sell to consumers, its priority in your GEO stack drops. If you sell B2B — enterprise software, industrial components, professional services — DeepSeek may actually be rising in relevance, because that is where its remaining, stickier users live.

The budget reallocation, concretely

Most brands running China GEO still weight their effort roughly evenly across the top models, a habit left over from the 2025 "cover everything" era. The June data argues for a sharper allocation. A consumer brand rebalancing for the second half of 2026 should be shifting the center of gravity toward Doubao — optimizing the Chinese-language evidence pages, structured product data, and third-party citations (Xiaohongshu, SMZDM, Zhihu) that Doubao's retrieval actually pulls from — while treating DeepSeek as a maintenance-tier presence rather than a growth priority. Qwen deserves a rising share precisely because its commerce-linked intent converts.

This is not a call to zero out DeepSeek. A −20.3% year still leaves nearly 130 million monthly users, and DeepSeek remains a default engine inside many downstream apps a brand cannot see directly. The point is proportion: the platform that most deserved the first dollar of your China GEO budget in early 2025 no longer deserves it in mid-2026.

The risk of over-correcting

There is a wrong way to read this data, and it is worth naming. A −20.3% headline can tempt a marketing team to swing from "DeepSeek-first" to "DeepSeek-never" in a single planning cycle. That over-correction ignores two facts. First, DeepSeek is deeply embedded as a backend model inside apps, browsers, and devices a brand will never optimize for directly — its influence on what Chinese users see extends well past its own app's MAU. Second, consumer habits are not the whole market. If your buyers are procurement managers, developers, or specialists who ask detailed comparison questions, DeepSeek's reasoning strength keeps it directly relevant to your category.

The disciplined move is a proportional shift, not a purge. Think in terms of budget share, not binary presence: a consumer brand might move DeepSeek from, say, a quarter of its China GEO effort down to a tenth, while Doubao absorbs most of the difference. That keeps a maintenance-tier presence — enough to correct misinformation and stay cited — without over-investing in a shrinking consumer surface.

What this signals about China AI more broadly

DeepSeek's contraction is the clearest evidence yet that China's consumer AI market is consolidating around ecosystems, not standalone models. A great model is no longer enough; without a distribution engine and a path to commerce, even a category-defining product loses consumer ground. Doubao's Douyin loop and Qwen's Alibaba stack are structural advantages that no amount of model quality can offset on the consumer side.

For brands, the lesson generalizes beyond any one platform: track where consumer AI intent is concentrating, and re-check the map every quarter. The rankings that felt permanent a year ago were not. The teams that win China AI visibility in 2026 will be the ones who treat their model-priority list as a living allocation, revised against fresh MAU and citation data — not a fixed plan drawn up when DeepSeek was the only name anyone knew.

Takeaway for Brand Marketers

• Reclassify DeepSeek by audience. Consumer brands: demote it from priority to maintenance tier. B2B brands: it may be holding or rising in value as its user base concentrates toward professional and enterprise use.
• Move the center of gravity to Doubao. With 382M MAU and a Douyin-to-commerce loop, it is where consumer discovery and conversion now live. Prioritize Chinese-language evidence pages and the third-party sources (Xiaohongshu, SMZDM, Zhihu) Doubao cites.
• Raise Qwen's weight. Alibaba's commerce-linked intent turns AI answers into purchases — high-value visibility for retail and consumer brands.
• Make allocation dynamic. Re-run your model-priority split against fresh QuestMobile MAU and your own brand-citation tracking every quarter. A −20.3% swing in a single leader is proof the map moves fast.

Related: see how these platforms score international brands on the hubGEO /brands tracker, updated across China's six major AI models.