> External research summary > This article is an English editorial synthesis of QuestMobile’s Q1 2026 AI Application Insights, published on April 21, 2026. All quantitative findings are attributed to QuestMobile unless otherwise stated. The underlying audience measurements have not been independently validated in this article.

Research Source

Abstract

QuestMobile’s first-quarter 2026 report describes a Chinese AI application market that has moved beyond its initial phase of rapid audience acquisition. By March 2026, the category-level monthly active audience for AI-native applications had reached approximately 446 million users. Doubao, Qwen, and DeepSeek were the three largest individual applications, with monthly active user counts of approximately 345 million, 166 million, and 127 million respectively.

The report argues that competition is shifting toward engagement quality, retention, scenario coverage, and ecosystem integration. Average usage frequency and time spent increased materially, while new user growth expanded into older and lower-tier-city populations. Spring Festival promotional campaigns accelerated adoption, but post-campaign retention exposed meaningful differences among platforms.

QuestMobile also identifies a second competitive layer built around “Claw” agent ecosystems. In this model, a system-level agent orchestrates applications, skills, data, and infrastructure. The competitive unit expands from a standalone AI application to a broader stack that includes user entry points, skill marketplaces, local or cloud deployment, operating-system access, developer participation, and hardware integration.

Executive Summary

Core Market Metrics

The category-level audience represents a deduplicated market total. Individual application MAUs overlap and should not be added together.

Key Findings

1. The market reached mass consumer scale. AI-native applications reached more than 446 million monthly active users by March 2026. The category had entered an accelerated expansion phase by November 2025 and added approximately 135 million users over the following four months.

2. Engagement grew alongside audience size. The average user opened AI-native applications 87.1 times and spent 173.3 minutes in the category during March. Both metrics increased significantly from November 2025, indicating deeper habitual use rather than acquisition alone.

3. Doubao maintained the strongest combination of scale and engagement. Doubao ranked first in MAU, average monthly usage frequency, average activity rate, and March seven-day retention among the leading general-purpose AI applications.

4. Qwen achieved exceptional campaign-driven growth. Qwen rose from sixth place in November 2025 to second place by March 2026. Its February download surge was substantially larger than those of Doubao and Yuanbao, although its activity rate and post-acquisition daily audience remained lower than Doubao’s.

5. DeepSeek retained a strong usage position without comparable promotional growth. DeepSeek ranked third in MAU and second in monthly usage frequency. Its average activity rate declined slightly year over year, suggesting that high awareness and utility did not translate into the same engagement acceleration seen at Doubao.

6. Spring Festival campaigns changed the distribution of AI traffic. Alibaba, ByteDance, and Tencent used transaction, content, hardware, and social ecosystems to acquire users. Doubao, Qwen, and Yuanbao collectively added approximately 130 million users during the promotional window.

7. The next competitive layer is ecosystem orchestration. QuestMobile positions Claw agents as system-level orchestrators that connect user intent with skills, applications, infrastructure, and hardware. This shifts competition from app usage toward scenario invocation, ecosystem depth, and execution reliability.

1. Market Expansion and Engagement Growth

1.1 AI-Native Applications Reached 446 Million Monthly Active Users

QuestMobile divides the market’s recent development into two phases.

During the earlier accumulation period, the category expanded from approximately 142 million monthly active users in January 2025 to approximately 304 million in October 2025. This represented an increase of about 162 million users.

The accelerated phase began around November 2025. Category-level monthly active users increased from approximately 311 million in November 2025 to 446.15 million in March 2026. The category added 134.95 million users during this period, equivalent to growth of 43.4%.

The growth curve suggests that AI-native applications moved from early-adopter adoption into broad consumer distribution. Spring Festival marketing contributed to the acceleration, but the market had already entered a faster expansion phase before the February campaign peak.

1.2 Engagement Increased with Scale

In March 2026, the average AI-native app user recorded:

• 87.1 usage sessions per month
• 173.3 minutes of monthly usage time

Compared with November 2025, monthly usage frequency increased by 36.9%, while monthly usage time increased by 30.3%.

This is an important shift in category quality. Audience growth without corresponding engagement can indicate temporary promotion-led acquisition. In this case, both frequency and time spent increased, suggesting that a meaningful share of newly acquired users continued to use AI products after installation.

The growth in usage frequency was slightly stronger than the growth in time spent. This pattern may indicate that AI applications were becoming more integrated into repeated, task-oriented interactions rather than being used only for long exploratory sessions.

1.3 The Audience Expanded Toward Older and Lower-Tier-City Users

The March 2026 category audience remained male-skewed:

The male share increased by 4.5 percentage points year over year, corresponding to an increase of approximately 122 million male users.

The report also identifies two expansion directions:

• Older users, particularly the post-1960s cohort
• Users in third-tier and lower-tier cities

Post-1960s users accounted for 7.0% of the audience and increased by 1.7 percentage points year over year. QuestMobile estimates that this cohort added approximately 16.6 million users.

Third-tier cities accounted for 23.0% of the audience, fourth-tier cities for 17.7%, and fifth-tier or lower cities for 9.1%. These segments all increased their share. QuestMobile estimates that users from third-tier and lower-tier cities increased by approximately 91.3 million.

The category was therefore expanding beyond young users in major urban centers. This broadening creates demand for simpler onboarding, voice interaction, practical life services, education, health information, and transaction-oriented use cases.

2. Competitive Landscape of AI-Native Applications

2.1 March 2026 Monthly Active User Ranking

QuestMobile’s March 2026 ranking excludes AI assistants embedded by smartphone manufacturers, such as Huawei Celia and OPPO Xiaobu. It focuses on standalone AI-native applications.

Doubao had more than twice the MAU of Qwen and approximately 2.7 times the MAU of DeepSeek.

Qwen recorded the largest first-quarter increase, adding approximately 126.33 million monthly active users between January and March. Doubao added approximately 101.07 million. The remaining top-ten applications recorded much smaller absolute gains.

2.2 Ranking Changes Show a Rapid Redistribution of Traffic

Doubao remained first from October 2025 through March 2026.

DeepSeek held second place through January 2026, then moved to third after Qwen’s February surge. Qwen rose from sixth in November 2025 to fifth in December, fourth in January, and second in February.

Tencent Yuanbao remained third through January and moved to fourth after Qwen overtook both Yuanbao and DeepSeek.

The ranking changes show that user distribution was highly sensitive to ecosystem-led acquisition. Qwen’s rise did not require a gradual multi-quarter progression. A large promotional and distribution event changed the ranking within a single month.

2.3 Scale and Engagement Did Not Move in Parallel

QuestMobile reports both average monthly usage frequency and average activity rate. Average activity rate is defined as average DAU divided by MAU during the quarter.

Doubao led both frequency and activity rate. Its position indicates that it combined broad reach with relatively deep recurring use.

DeepSeek ranked second in usage frequency despite ranking third in MAU. Its 41.7 monthly uses per user indicate strong functional utility among active users. However, its activity rate declined by 1.6 percentage points, while Doubao, Qwen, and Yuanbao all improved.

Qwen’s 19.8 monthly uses and 17.1% activity rate were considerably lower than Doubao’s. Its acquisition strength was therefore greater than its engagement depth during the same quarter.

2.4 The Leading Applications Served Different Engagement Models

The data suggests three distinct competitive positions:

These positions support QuestMobile’s broader argument that user count alone no longer explains competitive quality. Products are differentiating through scenario relevance, ecosystem access, and the frequency with which users return to complete real tasks.

3. Spring Festival Campaigns as an Acquisition Engine

3.1 Three Internet Groups Used Different Ecosystem Advantages

QuestMobile characterizes the Spring Festival competition through three distinct strategies.

Alibaba: AI Plus Transactions

Alibaba used approximately RMB 3 billion in cash incentives through a Spring Festival campaign. Qwen was connected to e-commerce and local-life services.

The strategy used transactions as the activation mechanism. The product’s longer-term position was a life assistant capable of connecting search, payment, travel, shopping, food delivery, and other Alibaba ecosystem services.

ByteDance: Full-Stack AI, Content, and Hardware

ByteDance used Doubao as a central entry point, supported by content distribution and hardware-related exposure. The campaign included approximately 10 million physical prizes and a major Spring Festival media partnership.

The strategy strengthened consumer awareness while connecting Doubao to visual interaction, content creation, and shopping scenarios.

Tencent: AI Plus Social Distribution

Tencent used approximately RMB 1 billion in cash incentives through Yuanbao and distributed the campaign through a relay-style WeChat red-packet mechanism.

The strategy used Tencent’s social graph to accelerate adoption. Yuanbao’s longer-term positioning connected office productivity with social interaction.

3.2 Qwen Produced the Largest Download Spike

Monthly downloads for the three campaign leaders were:

Qwen’s February downloads reached approximately 173.7 million. Its weekly peak was approximately 102.8 million downloads.

Doubao showed a smaller campaign spike and a more stable download baseline. QuestMobile interprets this as evidence that Doubao had entered a phase supported by organic growth and brand recognition, rather than depending exclusively on promotional acquisition.

3.3 User Acquisition Was Followed by a Retention Test

The deduplicated audience across Doubao, Qwen, and Yuanbao increased from approximately 279.6 million in January to 418.7 million in February, then decreased slightly to 413.0 million in March.

The February surge therefore produced an immediate increase in category reach, followed by a modest post-campaign correction.

Daily active users after the Spring Festival peak stabilized at approximately:

Doubao retained a substantially larger daily audience after the campaign. Qwen preserved a meaningful share of its acquired users, but its DAU settled far below its February promotional peak.

3.4 Seven-Day Retention Differentiated Acquisition Quality

Doubao’s seven-day retention improved throughout the quarter and reached 60.6% in March.

Qwen’s retention fell to 32.6% during the February acquisition peak, then recovered to 40.0% in March. The pattern is consistent with a campaign that brought in a large number of lower-intent users, followed by partial filtering and stabilization.

Yuanbao’s retention remained relatively stable, but below Doubao and Qwen.

The data demonstrates that large-scale user acquisition and durable user value are separate outcomes. Promotional distribution can change MAU rankings quickly, while retention reveals how effectively the product converts campaign traffic into repeated use.

4. Product Positioning and Ecosystem Differentiation

4.1 Doubao: Visual Interaction and Commerce

QuestMobile describes Doubao as a product driven by visual interaction and e-commerce.

Its highlighted capabilities include:

• Real-time video calls
• Object recognition
• Natural voice interruption
• Shopping links that allow products discussed in the conversation to be purchased inside the Doubao experience
• Continued development of model reasoning and coding capabilities
• Integration with AI phone initiatives

The commercial logic connects multimodal interaction with transaction conversion. Doubao can use visual and conversational context to identify user needs, then connect those needs to content or commerce.

4.2 Tencent Yuanbao: Office and Social Connectivity

Yuanbao is positioned as a connector across office and social scenarios.

QuestMobile highlights:

• Integration with the WeChat and QQ ecosystems
• Office-productivity use cases
• Cross-system AI social interaction
• Collaborative creation with contacts
• Meeting transcription and summarization
• WeChat-context understanding
• Mini-program transaction testing

This positioning uses Tencent’s communication graph as the primary ecosystem asset. Yuanbao’s differentiation comes from connecting people, conversations, work, and services rather than competing only through general model capability.

4.3 Qwen: A Life-Service Assistant

Qwen is positioned as a “life manager” supported by Alibaba’s service ecosystem.

QuestMobile highlights access to:

• Taobao
• Alipay
• Fliggy
• Shopping and local-life services
• Food delivery
• Ticket and hotel booking
• Financial information
• Spreadsheet generation and editing
• Itinerary organization

Qwen’s strategic advantage is the ability to connect conversational intent with transaction and service execution across Alibaba’s ecosystem.

The resulting product model is broader than question answering. It aims to complete multi-step consumer tasks that would otherwise require users to move across several applications.

5. Audience Profiles and GEO Implications

5.1 The Three Leading Campaign Products Had Different Audiences

The March 2026 audience profiles of Doubao, Qwen, and Yuanbao differed materially.

Qwen had the strongest male skew. Doubao had the most balanced gender distribution among the three.

Qwen also had a relatively high share of users born in the 1990s, while Yuanbao showed comparatively stronger representation in major and developed urban markets. Doubao had broader distribution across age and city tiers.

5.2 QuestMobile Connects Audience Differences to GEO Strategy

The report uses GEO to describe brand visibility and source optimization within generative AI systems.

Its central claim is that GEO cannot be implemented as a uniform cross-platform tactic. Each assistant serves a different audience and may prefer different source structures.

Examples proposed by the report include:

• Qwen’s male-skewed audience may respond more strongly to technical specifications and structured product parameters
• Yuanbao’s stronger presence in developed cities and its relationship with the WeChat ecosystem may increase the value of well-structured WeChat Official Account content
• Platform-specific source preferences should influence how brands structure and distribute authoritative information

This section is primarily strategic interpretation rather than direct behavioral measurement. The report provides audience-profile evidence, then infers how brands may adapt content for AI-mediated discovery.

6. From Applications to Claw Agent Ecosystems

6.1 QuestMobile’s Definition of the Claw Model

QuestMobile uses “Claw” to describe a class of agent systems organized around a system-level orchestrator and an extensible skill ecosystem.

The architecture contains four major layers:

The report compares this structure with an operating system plus an application store.

In the traditional model, the user identifies the required application and manually completes the workflow. In the Claw model, the user expresses an objective and the system selects and coordinates the required skills.

The interaction shifts from “the user finds an app” to “the task finds the required system capabilities.”

6.2 Competition Expands from User Attention to Scenario Invocation

Traditional applications compete for user time, sessions, and screen presence.

A skill-based agent ecosystem competes for invocation rights. A platform creates value when it becomes the default execution path for a task, even when the user does not directly open the underlying application.

QuestMobile therefore separates competition into two connected entry points:

The strategic path proposed by the report is:

Awareness → User Reach → Value Conversion → Commercial Execution

Spring Festival red-packet campaigns primarily strengthened awareness and reach. Claw systems are intended to convert that reach into repeatable execution and commercial value.

6.3 Skills Become the Functional Interface of Applications

The report argues that major Chinese applications are becoming skill providers.

Examples include:

Office and Collaboration

• Tencent Docs for document creation and editing
• Tencent Meeting for meeting summaries
• Tencent Maps for route planning and point-of-interest search
• Feishu for meeting records and multidimensional tables
• DingTalk for approvals, attendance, and group management

Creation and Production

• Meitu for image creation, illustration, intelligent cutout, and AI styling
• Baidu Miaoda for generating applications, mini-games, and web pages with payment functionality

Life Services and E-Commerce

• Fliggy for hotel and ticket search and booking
• E-commerce services for order handling, product search, and logistics queries after authorization

Content Connection and Distribution

• Social and content platforms such as QQ, Weibo, Xiaohongshu, Bilibili, Douyin, and Kuaishou
• Cross-platform message aggregation
• Content retrieval and analysis
• One-time creation with multi-platform distribution

This transition changes the role of an application. The application remains a consumer destination, but its services can also be invoked by an external agent as modular capabilities.

7. Structure and Economics of the Claw Ecosystem

7.1 Four-Layer Ecosystem Model

QuestMobile describes the domestic Claw ecosystem as a four-layer structure.

User Entry Layer

The main participants are:

• Individual users
• Developers
• Device manufacturers

Platform and Skill Layer

This layer provides:

• Hosting platforms
• Skill marketplaces
• Enterprise capabilities
• Application and service integrations

Cloud and Infrastructure Layer

This layer provides:

• Compute
• Deployment
• Data infrastructure
• Stable and secure execution environments

Open-Source Framework Layer

This layer provides:

• Agent execution frameworks
• Core runtime capabilities
• Extensible orchestration foundations

The ecosystem is supported by compute supply, storage and memory, network connectivity, security and compliance, skill orchestration, and algorithmic support.

7.2 Business Models Differ by Layer

The report’s model implies that no single layer captures all value.

Application providers control user relationships. Skill platforms control access and distribution. Cloud and infrastructure providers control resource efficiency and execution reliability.

7.3 Ecosystem Metrics Should Mature over Time

QuestMobile proposes a staged measurement framework.

Short-Term Metrics: Scale and Coverage

• Number of skills
• Number of active developers
• Number of agent deployments
• Number of user entry points

Mid-Term Metrics: Quality and Commercial Efficiency

• Skill invocation success rate
• Skill reuse rate
• Average response latency
• Orchestration complexity
• Enterprise customer count
• Skill transaction volume
• Average skill transaction value
• Agent MAU

Long-Term Metrics: Intelligence and Interoperability

• Share of multimodal skills
• Autonomous orchestration index
• Memory recall accuracy
• Cross-vendor skill interoperability

The market was still concentrated on short-term scale indicators at the time of the report. QuestMobile concludes that the ecosystem had not yet reached a mature phase defined by execution quality, autonomous intelligence, and interoperability.

8. Early Market Status of Claw Agents

8.1 Mainstream Products and Deployment Models

QuestMobile’s April 2026 snapshot includes the following representative products:

The market already included local, cloud, web, desktop, and mobile deployment models. Ecosystem binding was a major point of differentiation.

8.2 User Scale Remained Small and Volatile

QuestMobile’s weekly active user snapshot from March 23 to April 12, 2026 showed:

These products remained small relative to mainstream AI-native mobile applications.

The short observation window and large percentage movements indicate an experimental market with unstable adoption. Product presence and developer activity were expanding, but durable consumer demand had not yet been established.

9. Three Conditions for a Defensible Claw Ecosystem

QuestMobile identifies three connected requirements.

9.1 A Standalone Product Is Insufficient

Foundation-model capabilities are converging, which reduces the durability of differentiation based only on model intelligence.

The user experience increasingly depends on how effectively the agent can access resources and complete complex tasks. The relevant competitive asset is therefore the breadth and reliability of the ecosystem that the agent can orchestrate.

9.2 Users Select an Ecosystem Through the Product

A request such as “find and book a restaurant” may require:

• Conversational interpretation
• Maps and location
• Restaurant data
• Reservation infrastructure
• Payment
• Reviews and feedback

Each stage depends on an ecosystem partner or first-party service.

An AI assistant without these connections may provide recommendations, but it cannot complete the workflow. The value difference is created by execution coverage rather than answer quality alone.

9.3 Ecosystem Depth Determines Commercial Value

QuestMobile proposes a positive feedback model:

More connected services → Higher user stickiness → More developer participation → Greater ecosystem depth

A deeper ecosystem provides more data and more complete services. This increases user dependence, attracts additional developers, and further expands the ecosystem.

The strength of this loop depends on execution quality. A large skill count alone will not create a moat if invocation fails, latency remains high, permissions are confusing, or cross-service workflows are unreliable.

10. Claw as an AI Hardware Catalyst

QuestMobile extends the Claw model beyond software and identifies four hardware directions.

Fixed Deployment

Dedicated AI computers designed for continuous operation can provide local execution, persistent storage, independent compute, and stronger isolation.

Examples cited by the report include Lenovo YOGA AI Mini and Think AI Tiny.

Portable Agent Devices

Portable “Claw boxes” can package the agent, models, and skills into a device that works after startup with limited configuration.

The report cites 360’s ClawBox as an example.

Upgrades to Existing Computers and Phones

Existing hardware can be adapted through system-level agent integration.

The report points to laptops and smartphones that embed system agents capable of coordinating on-device applications and services.

Embodied Intelligence and Wearables

Robots, AI glasses, and other wearable devices can use an agent as the intelligence layer that connects perception, task understanding, and execution.

In this model, Claw is not a separate consumer application. It becomes a runtime for intelligent hardware.

Strategic Interpretation

The Market Is Entering a Retention and Execution Phase

The first competitive phase focused on awareness and acquisition. Large campaigns proved that Chinese internet groups could move hundreds of millions of users into AI applications.

The next phase is measured by:

• Whether users return after incentives end
• Whether the product solves recurring tasks
• Whether the assistant can execute across services
• Whether ecosystem access improves completion rates
• Whether engagement can support sustainable monetization

Doubao entered this phase with the strongest combination of scale, activity, frequency, and retention. Qwen demonstrated the strongest acquisition leverage. DeepSeek retained high functional usage. Yuanbao’s opportunity depends on converting Tencent’s social and office graph into recurring agent workflows.

Ecosystem Access Is Becoming a Product Capability

Model quality remains important, but the report treats ecosystem access as a first-class product capability.

An assistant connected to maps, payment, shopping, travel, communication, and productivity systems can complete tasks that a standalone model cannot. The product’s effective capability is therefore determined by both intelligence and accessible resources.

This is particularly relevant to system-level assistants and device manufacturers. Operating-system access provides stronger control over applications, permissions, sensors, local data, and cross-device continuity. Internet platforms provide large service and skill ecosystems. Future competition is likely to involve both collaboration and control disputes between these layers.

Campaign Growth and Product Value Must Be Evaluated Separately

Qwen’s February surge demonstrates the power of transaction ecosystems and financial incentives. Its lower activity rate and lower retention than Doubao show that campaign-driven MAU does not automatically translate into equivalent habit depth.

A complete evaluation should separate:

• Acquired users
• Retained users
• Daily active users
• Usage frequency
• Time spent
• Completed tasks
• Commercial conversion
• Cross-service execution success

MAU alone can overstate competitive strength during major promotional periods.

The Claw Market Was Still Pre-Maturity

The Claw section describes a strategically important direction, but the measured products remained small and volatile.

The market had established:

• Multiple deployment models
• Early skill connections
• Local and cloud execution
• Enterprise and consumer experiments
• Initial device integration

It had not yet demonstrated:

• Large-scale consumer adoption
• Stable retention
• Consistent skill quality
• Mature business models
• Broad cross-platform interoperability
• Reliable autonomous execution

The report’s long-term ecosystem thesis should therefore be read as a strategic framework rather than evidence of a mature, commercially proven market.

Methodology and Interpretation Notes

Definition of an AI-Native Application

QuestMobile defines an AI-native application as an application that uses AI as its central product driver and has been designed or substantially reconstructed around AI capabilities.

The category excludes smartphone-vendor assistants such as Huawei Celia and OPPO Xiaobu.

The reported market totals should therefore not be interpreted as the complete audience for all AI assistants in China.

Activity Rate

Activity rate is calculated as:

Average DAU during the measurement period ÷ MAU during the measurement period × 100%

The Q1 activity rate is the average of the January, February, and March activity rates.

Seven-Day Retention

Seven-day retention represents the average share of daily active users who reopen the same application on the seventh day within the monthly measurement period.

Data Scope

The primary audience data covers January 2025 through March 2026.

Some Claw product metrics cover March 23 through April 12, 2026, which extends beyond the main first-quarter observation window.

Analytical Limitations

1. The report is mobile-app centered. Its primary audience measurements do not fully capture browser-based use, API usage, embedded assistants, enterprise deployment, or system-level AI interactions.

2. Individual application MAUs overlap. The MAU ranking should not be added to estimate total category reach.

3. Promotional periods distort short-term comparisons. Spring Festival campaigns produced unusually large installation and activation spikes.

4. Product categories differ. General assistants, search products, educational applications, professional advisors, and creative tools serve different intents. Direct comparisons of frequency or retention should consider these differences.

5. The Claw framework combines measurement and strategic interpretation. The architecture, business models, and long-term indicators are QuestMobile’s analytical model. They are not all directly validated by market-scale performance data.

6. GEO recommendations are inferential. Audience profiles are measured, while recommendations about content structure and source preference are strategic conclusions derived from those profiles.

Conclusion

QuestMobile’s Q1 2026 data shows that China’s AI-native application market reached mass consumer scale while also improving engagement. By March, the category had approximately 446 million monthly active users, with higher usage frequency and time spent than in late 2025.

Doubao held the strongest overall position across scale, engagement, and retention. Qwen used Alibaba’s ecosystem and Spring Festival incentives to produce the quarter’s most significant user-acquisition event. DeepSeek maintained high usage frequency, while Tencent Yuanbao continued to build around social and office integration.

The market’s competitive logic is expanding. Consumer AI products are no longer evaluated only by model capability or MAU. Retention, repeated task completion, service access, transaction execution, and ecosystem orchestration are becoming equally important.

QuestMobile’s Claw framework captures this shift. The framework describes an agent layer that receives intent, selects skills, coordinates applications, and uses infrastructure or hardware to complete tasks. The market remained early and fragmented in April 2026, but the direction points toward competition over operating-system access, skill distribution, developer ecosystems, and scenario invocation.

The report’s central implication is that the next phase of AI application competition will be determined by the ability to convert broad audience reach into reliable, repeatable, and commercially valuable execution.