Doubao Universe 2026: Four Surfaces Where Brand Visibility Is Now Decided
Doubao has 345 million monthly users and more than 200 million daily users, yet by one account it earns under RMB 1 million a day, almost all of it e-commerce commission, while compute costs run to tens of millions. That gap explains why ByteDance is turning its chatbot into what the press now calls the "Doubao Universe": cars, phones, office software, a shopping layer and Douyin content search. For international brands, the implication is blunt. Doubao is no longer one place to be recommended. It is at least four, and each one rewards different signals.
This briefing is based on a 36Kr/Zimubang analysis dated 24 September 2026, QuestMobile's H1 2026 rankings, and our own reading of what it means for the brand scores tracked on GEO Hub.
The economics behind the pivot
According to the 36Kr analysis, Doubao's app passed 200 million DAU in H1 2026 with 345 million MAU, first among domestic AI apps. QuestMobile's June ranking, published on 4 August, shows 382 million MAU for Doubao, against 167 million for Qwen and 130 million for DeepSeek. The two Doubao figures differ by month and methodology, so treat both as directional.
The more important number is the revenue line. The same report says daily revenue is below RMB 1 million and comes almost entirely from commissions on e-commerce, while compute costs reach tens of millions of RMB per day. Dialogue is the expensive part: text plus image, voice and video generation can consume dozens of times the tokens of a plain chatbot. The article's conclusion is that the more users chat, the larger the losses, because Doubao cannot yet pass costs on to consumers.
This is our interpretation rather than ByteDance's stated position, and ByteDance has disputed parts of the reporting on internal reorganisation. A Doubao PR head said on 24 September that only 11 people were affected by a team restructuring. But the strategic direction, keeping chat as a base while investing in surfaces that monetise, is consistent across the product moves below.
Pillar 1: Control (cars, phones, devices)
The report lists several launches that place Doubao inside hardware. Volcano Engine and Deep Blue's S07 AI laser edition carry an end-to-end Doubao model for voice control and route planning. SAIC Roewe's Jiayue 07 will ship with a Doubao assistant, SAIC Audi plans similar models, and Tesla China pushed a Doubao real-time dialogue model in a mid-August software update. A Doubao phone built with Nubia launched on 16 September with system-level cross-app agent operation. A separate 36Kr piece notes the second-generation phone uses the Nubia NaviX Ultra and a strategy of trading protocol access for scale.
For brands, a voice query in a car or an agent action on a phone produces one answer, not a results page. Where a desktop user might scan five options, a driver hears one or two names. We have covered the in-car case in our Tesla in-car AI analysis and the agent-phone case in China agent-phone execution queries. The new point here is scale: control surfaces are now a stated pillar of the company's strategy, not an experiment.
Pillar 2: Office (Feishu and Doubao Work)
Doubao AI is being embedded across Feishu modules, and Doubao Work has a prominent place in the Feishu client. The report speculates that Feishu could eventually be merged into Doubao Work, though that is speculation. For B2B brands, the relevant fact is that vendor research increasingly happens inside work tools, where an employee asks an assistant to shortlist suppliers. Our earlier note on workplace agents as a routing layer applies directly. Brands that appear in Doubao's public answers may still be absent from the narrower, enterprise-context answers.
Pillar 3: Transactions (the shopping layer)
This is where Doubao earns money today. Douyin mall links appear in AI answers, and Doubao has added ordering, payment and order management so that a purchase can finish without leaving the app. The navigation entry "Doubao Helps You Choose" arrived in April and was upgraded in May to "Ask Doubao Before Buying". The report notes users may now perceive it as a "Doubao mall" rather than a Douyin feature.
Named brand partners include China Merchants Bank, Feihe, Luckin Coffee, Peking Union Medical College Hospital and New Oriental, across banking, retail, food and beverage, healthcare and education.
Two consequences follow. First, a recommendation here is partly a commercial result: where a model earns commission, brands with live Douyin shop inventory have a structural advantage over brands that exist only as editorial mentions. Second, organic visibility and paid presence are likely to blur, a trend we discussed in Doubao's paid-ad rollout. We cannot prove from public data how commission influences ranking, and the article does not claim it does. But brands should assume commerce readiness is a signal.
Pillar 4: Content (Douyin search and AI-generated media)
The fourth pillar combines AI-generated content with search over Douyin's existing video and posts. The report gives few specifics, but the direction matters for brand GEO: Douyin content becomes both source material and answer format. A brand with a strong Douyin presence, including official accounts, creator reviews and product demonstrations, has more material for Doubao to draw on. This extends the Doubao-Douyin flywheel we have tracked.
What this means in a comparison
| Pillar | Query type | Answer format | Signal that likely matters most | Brand risk |
|---|---|---|---|---|
| Control | Voice, task execution | One or two names | Clear Chinese entity, in-car and device partnerships | Absence from the single answer |
| Office | Vendor and tool research | Shortlist in work context | B2B documentation, Feishu-compatible content | Invisible to enterprise users |
| Transactions | "Which should I buy" | Ranked picks with purchase links | Douyin shop presence, reviews, price accuracy | Losing to brands with live inventory |
| Content | Discovery, how-to | Video and post-based answers | Douyin creator and official content | Thin Chinese video footprint |
Caveats
Several figures here come from secondary reporting. The revenue and compute-cost estimates are attributed in the 36Kr piece to industry reporting rather than company disclosure, and the 36Kr page metadata shows a later date (10 October) than the byline (24 September). MAU numbers differ between QuestMobile (382 million, June) and the article (345 million, H1 average). The article also argues the ecosystem's real purpose is to secure proprietary non-text data for model training, which is the author's view. Competitors, including Alibaba with its QwenBook tablet and Quark AI glasses, are chasing the same entry points, so Doubao's advantage may not last.
Takeaway for Brand Marketers
- Audit Doubao by surface, not as one score. A brand with a strong chat score can still be missing from the shopping layer or from voice answers. Compare your per-model scores on the GEO Hub brands dashboard, then test queries that mimic each pillar.
- Treat Douyin shop and content as GEO assets. Live inventory, accurate prices and creator content feed both the transaction and content pillars.
- Plan for one-answer outcomes. In cars and on agent phones there is no second page. Prioritise being the clear category answer for your two or three hero queries.
- Check for partner and device deals early. Automotive and device partnerships, such as the Doubao deployments in SAIC and Tesla China, are where brand mentions can be embedded by design.
- Expect monetisation pressure. With compute costs far above revenue, free recommendation space is likely to shrink. The window for organic gains is open now and may narrow.
Related: See how international brands score across all six Chinese AI models on the GEO Hub brands dashboard.
Sources: 36Kr / Zimubang, "Doubao Universe" analysis; 36Kr / Tingtong Tech, big-tech AI hardware; QuestMobile H1 2026 ranking as reported by Tencent News.