382 Million Users, Under $1 Million a Day: What Doubao's Failed Paywall Reveals About Brand Visibility in China's AI Search (2026)

In June 2026, Doubao hit 382 million monthly active users — more than the second- and third-place apps, Qwen and DeepSeek, combined. In the same month, according to reporting from Chinese outlet Huxiu (citing LatePost), Doubao's C-end business generated less than RMB 1 million ($140,000) in daily revenue, while its daily compute bill ran into the tens of millions of RMB. That gap — the largest AI audience in China producing almost no direct revenue — is the real story behind Doubao's paywall experiment, and it changes how brands should think about where AI visibility actually pays off in 2026.

The Paywall That Didn't Convert

On June 24, 2026, ByteDance rolled out three paid tiers for Doubao: Standard (RMB 68/month), Enhanced (RMB 200/month), and Premium (RMB 500/month), aimed at professional use cases in coding, finance, design, and data analysis. It wasn't the first signal — pricing had leaked via Apple's App Store in May, and that announcement alone triggered Doubao's first-ever monthly active user decline: a 1.81% drop, roughly 6.07–6.1 million users, in May 2026, according to AI产品榜 (AI Product Rankings) data cited by Eastmoney.

The obvious follow-up question — did the paid tiers stick, or did Doubao just scare off its most price-sensitive users — now has an answer. By June, MAU had not only recovered but hit a new high of 382 million, up 172.1% year-over-year according to QuestMobile. Analysts read this as evidence that the churn was shallow: light, casual users left, while the core high-frequency base stayed. That's a reasonable read, but it also exposes the ceiling on Doubao's monetization strategy — the users who stayed are largely the same ones who were never going to pay in the first place.

The Numbers Behind the 382 Million

ByteDance's overall large-model business has reportedly crossed $4 billion in annualized run-rate revenue (ARR), more than the combined ARR of China's other major model companies. But per Huxiu's breakdown, an estimated 90% of that is not consumer-facing. It comes from Volcano Engine, ByteDance's cloud/MaaS platform, which holds a reported 49.5% share of China's public-cloud MaaS market and processes roughly 180 trillion tokens a day, and from Seedance, Doubao's video-generation model, which reportedly does over RMB 1 billion in monthly revenue at a 70% gross margin.

The consumer subscription business looks thin by comparison. Industry-wide, C-end AI payment conversion in China runs 1%–3%. Doubao sits at the low end, 1%–1.8%, versus 2%–3% for Kimi and DeepSeek. Morgan Stanley's estimate, cited in the same report, puts Doubao's annualized subscription revenue at $101 million to $1.5 billion depending on conversion assumptions — even at the high end, that's under 40% of the $4 billion ARR figure, and the model requires a 3% conversion rate Doubao is nowhere near hitting.

The usage-depth numbers matter as much as the conversion rate. QuestMobile's June 2026 data shows DeepSeek and Kimi users spending 10+ minutes per session grew 7.8 and 8.4 percentage points year-over-year, respectively, while Doubao's grew just 2.2 points. Doubao has the biggest reach; DeepSeek and Kimi are building the deeper, more research-oriented sessions where users compare options, ask follow-up questions, and — for GEO purposes — are more likely to explicitly evaluate and recommend brands rather than just chat.

Why Doubao Is Pivoting to B2B, Not C2C

The clearest signal of where ByteDance actually expects to make money came on July 30, 2026, when the company folded its enterprise collaboration product, Feishu (Lark), into the Doubao product organization, with Feishu's head now reporting to Doubao's product lead. Feishu already sells to enterprises; in Q2 2026, ByteDance said 90% of Feishu's new customers also purchased its AI features. Merging the two effectively repositions Doubao's growth story around enterprise and API revenue — the same revenue lines already dominating ByteDance's model business — rather than consumer subscriptions.

Huxiu's analysis frames the four layers of logic behind the C-end paywall this way: it tests price tolerance without fully alienating free users, it signals to enterprise buyers that "people already pay for Doubao's capability" (useful in MaaS sales conversations), it segments high-frequency users, and it shifts a share of inference cost from ByteDance onto the users who consume the most compute. None of the four is "make money from subscriptions." As one investor quoted in China Business Journal put it, domestic users' willingness to pay is capped, and building it further "requires long-term product upgrades and user education" — a multi-year project, not a Q3 pivot.

What This Means for Brand Visibility

For international brands running GEO programs in China, the Doubao paywall data reframes two assumptions.

First, the free tier is not a shrinking, lower-value surface — it's still where over 97% of Doubao's audience lives, and where ByteDance has the least incentive to restrict AI-generated answers, since restricting free access is what caused the May churn. Brand visibility work aimed at Doubao should continue to target the mass, free-tier conversational surface rather than assuming premium features (deeper research modes, higher usage caps) meaningfully change how brands get recommended. hubGEO's brand tracking treats Doubao's free-tier answer behavior as the baseline for exactly this reason — see current brand scores at hubgeo.vercel.app/brands.

Second, Doubao's own leadership has effectively told the market where its business model is going: e-commerce commission (via the Douyin mall integration) and enterprise/API revenue, not subscriptions. That lines up with what we've tracked separately in Doubao's commerce loop — categories like beauty, where Doubao routes queries directly into Douyin's shopping ecosystem, are the ones ByteDance has the clearest financial incentive to keep optimizing for citation quality and conversion. Brands with an active Douyin commerce presence are plugging into the part of Doubao's business that ByteDance actually needs to work; brands relying solely on organic content elsewhere are competing for space in a feature ByteDance has openly said isn't the profit center.

Third, the usage-depth gap is a leading indicator, not a footnote. DeepSeek and Kimi's session-depth growth outpacing Doubao's suggests that as Chinese AI users mature past simple Q&A into comparison-and-decision tasks — the exact behavior that produces brand recommendations — that behavior is growing faster on the platforms already positioned for research and reasoning. A GEO program built only around Doubao's scale risks under-investing in the platforms where deeper, recommendation-generating queries are increasingly happening.

Takeaway for Brand Marketers

Don't read Doubao's 382 million MAU as 382 million equally valuable exposure opportunities, and don't read the paywall as evidence that Doubao is shifting toward a smaller, higher-value paid audience — the data says the opposite. The free tier remains the dominant surface by design, e-commerce integration (not subscriptions) is the monetization path ByteDance is actually building toward, and the platforms with deeper session growth — DeepSeek and Kimi — are where more decision-stage, brand-comparison behavior is concentrating. Budget allocation across China's AI platforms should track engagement depth and commerce integration, not raw MAU, and brands selling through Douyin's ecosystem have a structural advantage in Doubao specifically that a generic content strategy won't replicate on Kimi or DeepSeek.

Related: See current brand-by-brand AI visibility scores across Doubao, Kimi, DeepSeek, Qwen, Wenxin, and Hunyuan at hubGEO's Brands directory.