Why MAU Lies About AI Brand Exposure in China: The Query-Frequency Index for 2026

Here is the number that should reset how international brands allocate their China GEO budget: Doubao users open the app 54.8 times a month, while Qwen users open it 19.8 times. That 2.8x gap means a brand cited by Doubao's model gets surfaced to a real human far more often than the same citation inside Qwen — even before you count the raw user base. Yet almost every China AI visibility plan we see still divides spend by monthly active users (MAU), a metric that quietly overstates the reach of low-engagement platforms and undersells the ones people actually live in.

This piece introduces a simple correction we call the Query-Frequency Exposure Index — MAU multiplied by monthly sessions per user — and shows how it reshuffles the priority order that brand marketers have been working from all year.

The MAU trap

MAU answers "how many people could theoretically see my brand." It says nothing about "how many times they'll actually ask a question where my brand might appear." For generative engine optimization (GEO), the second question is the one that matters, because AI visibility is earned per-query: your brand is either cited in the answer or it isn't, and that dice roll happens every single time a user prompts the model.

QuestMobile's Q1 2026 data makes the distortion concrete. Across China's AI-native apps, MAU reached 440 million, with Doubao at roughly 345 million, Qwen (Tongyi) at 166 million, and DeepSeek at 127 million. On MAU alone, the story is "Doubao leads, but Qwen is a serious number two worth heavy investment."

The engagement data tells a different story. In the same quarter, average monthly sessions ran to 54.8 for Doubao, 41.7 for DeepSeek, and just 19.8 for Qwen. Industry-wide, users averaged 87.1 sessions and 173.3 minutes per month — up 55.3% and 41.4% year over year — confirming that 2026 is the year China's AI market shifted from raw user growth to genuine daily-habit depth. Doubao's daily active users hit 140 million by May 2026; Qwen, despite a 3 billion yuan (~$417M) coupon campaign in February that briefly lifted DAU from 7 million to 58 million, saw much of that traffic churn once the subsidies ended.

Recomputing the leaderboard

If we weight each platform by how often its users actually query, the ranking changes shape. Multiply MAU by monthly sessions per user to get total monthly query volume — the real number of chances your brand has to be cited.

(Shares calculated across these three leading assistants. Query volume = MAU × monthly sessions.)

The reshuffle is dramatic. On MAU, Qwen looks like a clear second place deserving roughly a quarter of your GEO attention. On query-weighted exposure, Qwen collapses to 12% — less than a third of what its user count implies — while Doubao expands from 54% to nearly 69% of all brand-citation opportunities. DeepSeek, the most engaged platform relative to its size, holds its ground and arguably deserves more weight than its modest MAU suggests.

In plain terms: a brand that splits its optimization effort evenly between Doubao and Qwen because "they're the top two" is pouring roughly a quarter of its work into a channel that generates one-eighth of the actual exposure.

Why Doubao's users query so much more

The frequency gap is not an accident of measurement — it reflects fundamentally different product roles. Model comparisons through mid-2026 consistently place Doubao as the everyday lifestyle-and-voice assistant (ByteDance leans on Douyin's content flywheel and social features to build a daily habit), DeepSeek as the technical and reasoning workhorse used repeatedly within work sessions, and Kimi as a specialized long-document tool. Qwen's user base, inflated by subsidy-driven acquisition, skews toward lighter, more occasional use.

For brands, product role maps directly onto citation context. Doubao's high-frequency lifestyle queries — "recommend a moisturizer for oily skin," "best coffee chain near me," "which SUV under 300k" — are exactly the commercial-intent prompts where being cited translates into consideration. That is precisely why Doubao's outsized query-weighted share is not just a bigger number but a higher-quality number for consumer brands.

The stickiness compounding effect

There is a second-order reason frequency matters more than the table alone shows. On a low-frequency platform, a user might encounter your category once a month; if your brand isn't cited that one time, you're invisible until the next cycle. On a high-frequency platform, the same user asks category-adjacent questions repeatedly, giving a well-optimized brand multiple reinforcing impressions — and giving a poorly-optimized brand repeated reminders of its absence. Frequency doesn't just multiply exposure linearly; it compounds brand memory through repetition. A 2.8x session gap can mean a much larger gap in actual brand recall.

This also reframes how to read subsidy-driven MAU spikes. When Alibaba's coupons drove Qwen's DAU up eightfold in February, the headline looked like a visibility windfall. But acquired-then-churned users generate few durable queries, so the exposure value of that spike was far smaller than the user count implied — and it faded fast. Brands that chased the spike with a Qwen-first content push were optimizing for a number that was already evaporating.

A reallocation worksheet

Turning this into an actual budget move takes three steps. First, pull the current session-frequency figures for each platform you're active on — Doubao 54.8, DeepSeek 41.7, Qwen 19.8 per month as of Q1 2026 — and multiply by each platform's MAU to get its query volume. Second, convert those volumes into percentage shares, as in the table above, to see each platform's true slice of citation opportunity. Third, compare that share against how your team currently splits GEO effort: content production, source seeding on platforms like Zhihu and Xiaohongshu, and structured-data work.

Most brands we review discover a two-way mismatch. They over-invest in Qwen because its MAU rank flatters it, and they under-invest in DeepSeek because its user base looks small. The fix is rarely to abandon a platform — coverage still matters — but to move the marginal hour and the marginal content asset toward the platform with the higher query-weighted return. For a consumer brand, that almost always means Doubao gets the flagship content and the deepest source-seeding effort, DeepSeek gets a focused technical-credibility layer, and Qwen gets maintenance-level coverage rather than a campaign.

A caveat on the math: these session-frequency figures are platform averages, not category-specific. A luxury or automotive brand whose buyers cluster in high-intent research sessions may see DeepSeek and Kimi punch above even their engagement-weighted share, because those users ask longer, more deliberate questions. Use the index as a default allocation, then adjust for where your specific category's queries actually concentrate. The point is not that one formula replaces judgment — it's that MAU should never be the number you start from.

What this means going forward

The broader pattern behind these numbers is that China's AI market has entered its habit-formation phase. With industry-wide sessions up 55.3% year over year, the platforms that win aren't just acquiring users — they're becoming daily utilities. For brands, that raises the stakes on being present where the habit is forming, because a citation on a daily-use platform compounds into recognition in a way a citation on an occasional-use platform never will. Expect the query-weighted gap between Doubao and the rest to widen, not narrow, as engagement deepens — which means the cost of misallocating today's GEO budget grows every quarter you leave it unfixed.

Takeaway for Brand Marketers

1. Stop budgeting GEO by MAU. Budget by query-weighted exposure. Multiply each platform's user base by its session frequency before you decide where optimization effort goes. On current data, that pushes Doubao from roughly half to roughly two-thirds of your priority, and shrinks Qwen to a supporting role rather than a co-lead.

2. Give DeepSeek more weight than its size suggests. Its high engagement means each of its 127M users is worth disproportionately more citation opportunities — especially for B2B, technical, and research-adjacent brands whose buyers live inside those work sessions.

3. Treat Qwen as a coverage play, not a primary channel — for now. It still reaches 166M people, so you want a baseline presence, but the low query frequency means it shouldn't command a Doubao-level content investment until its engagement deepens.

4. Discount subsidy-inflated user numbers. When a platform buys users with coupons, wait to see whether query frequency holds before you reallocate. Durable exposure comes from habit, not from one-time acquisition.

The brands winning China's AI search in 2026 are not the ones present on the most platforms. They're the ones concentrated where the questions actually get asked — over and over, every day.

Related: see how brand scores break down model-by-model on our /brands tracker.