From Plugins to Native Apps: What Has Changed in China's AI Entry-Point Landscape (2024–2026)?

China's generative AI market is undergoing a structural shift in user entry points. Early on, users mostly encountered AI features within existing products like search engines, office suites, and content platforms. Subsequently, standalone AI-native apps, in-app AI, mobile AI assistants, and agents have each expanded their user base. Drawing on multiple QuestMobile AI industry reports from 2025–2026, the CNNIC report 'Development Report on Generative AI Applications (2024)', and the 55th 'Statistical Report on China's Internet Development', this analysis examines China's generative AI entry points from four dimensions: user scale, usage frequency, entry point forms, and service connection methods, observing how they evolved from simple information interaction tools into a multi-entry parallel structure during the first half of 2024 through 2026.

Research Background

2024 was a period of rapid user adoption for generative AI in China. According to the CNNIC 55th 'Statistical Report on China's Internet Development', as of December 2024, 331 million people in China had heard of generative AI (23.5% of the population), and 249 million had actually used it (17.7% of the population). At that time, user interaction with generative AI mainly centered on information acquisition and production assistance: 77.6% used it to answer questions, and 45.5% used it as an office assistant. Moving into 2025–2026, market changes are reflected not just in user numbers, but more significantly in where users enter AI, how long they use it, and whether AI can further invoke external applications and services. The structure of entry points has therefore become a critical dimension for observing the evolution of the generative AI market.

AI Traffic Gravity Is Shifting from 'Add-on Features' to High-Frequency Native Entry Points

The most obvious change is that AI-native apps have achieved high usage frequency and depth. According to QuestMobile's '2026 AI App Market Semi-Annual Report', as of May 2026, the monthly active users (MAU) of China's AI-native apps reached 499 million, a year-on-year increase of 85.4%, with monthly per-capita usage reaching 92.7 times. This indicates that for a portion of users, AI is no longer just an occasional feature invoked within other products, but an independent entry point that can be opened and used repeatedly.

This contrasts sharply with the usage structure in 2024. In December 2024, CNNIC reported 249 million actual generative AI users, primarily for answering questions and office assistance. By May 2026, QuestMobile statistics showed native apps nearing 100 monthly uses per capita, suggesting the change isn't just more generative AI users, but a shift in user interaction habits from 'finding AI within other products' to 'entering AI directly'.

Concurrently, usage time for traditional internet applications is also seeing structural changes. QuestMobile's 2026 semi-annual report shows that in May 2026, 28% of industry apps saw simultaneous declines in both usage sessions and time spent. The traditional search sector saw a 19.1% year-on-year decline in monthly per-capita usage sessions and a 13.5% decline in monthly per-capita usage time. While this data indicates a redistribution of user time, it doesn't directly imply traditional search or apps will be replaced by AI. Traditional search, vertical apps, and AI-native applications still serve different tasks, but the share of AI-native entry points in high-frequency interactions is expanding.

China's AI Entry Points Are Not Converging into a Single Path but Forming a Parallel, Specialized Structure

The growth of native apps doesn't mean all AI traffic will consolidate into standalone applications. According to QuestMobile's '2025 Q3 AI Applications Industry Report', as of September 2025, China's mobile AI active users reached 729 million, and PC active users reached 200 million. Within the mobile sector, AI-native apps, in-app AI, and mobile manufacturer AI assistants coexist, each performing differently in user scale and usage depth.

For example, QuestMobile data shows that in Q3 2025, the monthly per-capita usage time for in-app AI grew 93.8% year-on-year, indicating deeper engagement from existing users. However, more than half of these applications saw negative user growth during the same period. AI-native apps show similar divergence, with nearly 60% experiencing negative growth in Q3 2025. This means analyzing AI entry points requires distinguishing between 'how many people are using' and 'how deeply existing users engage'.

Mobile AI assistants represent another logic. The QuestMobile 2025 Q3 report shows that within one year, AI assistants from six major phone manufacturers added approximately 65 million users (up 13.9% YoY), reaching a total of 535 million. Unlike standalone apps, these entry points leverage operating system permissions and device capabilities, enabling cross-app information retrieval and task coordination. The PC sector was in an adjustment phase during the same period, with 75.3% of web-based AI applications and 75.0% of client-based AI applications experiencing negative growth, as their usage scenarios concentrate more on complex work and deep tasks.

Therefore, AI entry competition in 2026 is closer to a 'multi-entry collaboration.' Native apps suit continuous, general-purpose interaction; in-app AI integrates tightly with existing business processes; mobile assistants have system-level connectivity; and PC continues to handle complex tasks. Different entry points compete for user time while also dividing labor based on permissions, scenarios, and task complexity.

Entry Competition Is Extending from 'Answering Questions' to 'Executing Tasks'

A second, deeper change is the expanding functional boundary of AI entry points. According to CNNIC's 55th report, as of December 2024, there were 249 million actual generative AI users, with 77.6% using it for answering questions and 45.5% as office assistants. The typical interaction at this stage remained 'user requests, AI outputs information or content'.

By 2025–2026, agents began pushing this relationship further into task execution. According to QuestMobile's '2026 AI App Market Semi-Annual Report' (using a recent 3-month metric), active users for office-delivery intelligent agents grew by 164.4%, and ecosystem-task agents by 115.3%. The former emphasizes producing direct work outputs like documents and analysis results; the latter orchestrates multiple services for queries, transactions, or cross-app operations.

The development of connection standards is also driving this change. QuestMobile's '2025 Q3 AI Applications Industry Report' showed over 5,400 MCP (Model Context Protocol) services available on the ModelScope community. The significance of standardized connections like MCP isn't just adding a new chat capability, but enabling models to connect data, tools, and external services in a relatively unified way. Simultaneously, GUI agents try to understand interfaces and execute operations, converting flows that previously required step-by-step user clicks into agent-orchestrated processes.

This extends the role of AI entry points from 'information gateways' toward 'service gateways'. Users may not just face a dialog box for Q&A, but an interaction layer that understands needs, selects services, and executes partial tasks. Competition now isn't just about model answer quality, but also about which systems AI can connect to, what capabilities it can invoke, and what types of tasks it can accomplish.

Traditional Apps Are Experiencing Two Structural Changes: 'Skill-ification' and 'Content Supplying'

As agents connect to more services, the role of traditional websites and apps in the AI environment changes. One path can be understood as 'skill-ification': functions previously encapsulated within apps, such as queries, calculations, bookings, or other features, are being decomposed into capabilities callable by agents. Users may not need to fully open an app each time; they might express their need through an AI entry point, with backend services providing the actual capability.

The other path is 'content supplying'. Professional content accumulated on traditional websites and apps, previously accessed mainly through search results, recommendation feeds, or direct app opens, may now enter new distribution pathways as AI organizes and invokes external information. According to public cases in QuestMobile's '2026 AI App Market Semi-Annual Report', Autohome's professional content was structure-invoked via AI-native apps, indirectly reaching approximately 14.97 million users, equivalent to 24.9% of its own app traffic. Pacific Auto's content, accessed via AI entry points, reached 113.8 times its own traffic.

It's crucial to distinguish between 'own traffic' and 'AI indirect reach'. The former typically means direct visits to a website or app; the latter may occur without users actively visiting the original content platform. These public cases show some professional content can reach users via AI information curation pathways, but this doesn't mean all structured content will automatically get AI citations, nor can a single technical adjustment be causally linked to citation outcomes.

For websites, this means content structure, brand entities, product relationships, and service information not only need to be user-friendly but also minimize ambiguity for machine understanding. The focus of website optimization is therefore expanding from traditional page display to whether content can be stably recognized, decomposed, and associated.

Trend Observation: Entry Competition Is Shifting Toward Content, Functionality, and Service Connectivity

Combining data from 2024–2026, China's generative AI entry competition has passed through several distinct stages. In 2024, the main variables were whether users started using generative AI and what questions it could answer. In 2025, native apps, in-app AI, mobile assistants, and PC products began forming clearer entry-point specialization. Entering 2026, high-frequency usage of AI-native apps and agent task-execution capabilities expanded further, with competition extending from information output to external service connectivity. China's AI entry competition is shifting from 'which app users open' to 'what content, functionality, and services AI can invoke.'

This doesn't imply that only one of native apps, in-app AI, mobile assistants, or agents will ultimately survive. They possess different system permissions, interaction costs, and suitable tasks. Native apps handle high-frequency general interactions; in-app AI aligns closely with existing business flows; mobile assistants have device-level capabilities; agents orchestrate content, tools, and services further. What's more worth observing in 2026 is how these entry points redistribute tasks among themselves, rather than seeking a single form that completely replaces the others.

For internet enterprises, the definition of traffic also needs adjustment. Previously, website visits and app MAU could fairly directly reflect whether users accessed owned digital assets. As AI takes over information filtering, content curation, and even service connection, some brand discovery and content reach may occur outside enterprises' own pages. The changes addressed by GEO thus originate not just from search results pages, but from the broader shift in digital information distribution as AI increasingly participates in information filtering, service invocation, and task execution.

Research Implications

At this stage, it's not necessary to predict which AI entry point will become long-term dominant. A more practical approach is to re-examine whether existing digital assets can adapt to a multi-entry environment. First, ensure core entity information—brand, products, services, and use cases—remains consistent with relatively clear relationships. Second, check whether website content can be easily decomposed and understood, avoiding important information residing only in complex visual pages or isolated modules lacking context.

Simultaneously, enterprises can start distinguishing two types of digital assets: one category is content supply serving as an information source (product documentation, professional knowledge, service descriptions, scenario information); the other is business capabilities potentially invoked via APIs or skills in the future. Optimization approaches differ: the former focuses on whether machines can accurately recognize content and entity relationships; the latter also requires considering service interfaces, permissions, and task execution methods.

Data analysis metrics also need gradual expansion. As AI indirect reach increases, solely monitoring direct website visits or app MAU may not fully capture a brand's presence in AI environments. While maintaining traditional search and traffic metrics, enterprises should also track brand presence, content citations, and indirect discovery across different AI entry points to assess whether structural entry-point changes are affecting their information distribution pathways.

Related Questions

Through which primary entry points did Chinese AI users access generative AI in 2026?

Based on public data, the primary entry points include AI-native apps, in-app AI within traditional apps, mobile AI assistants, and PC web/client applications. With agent development, task-oriented entry points that directly connect tools and services are also expanding.

What is the difference between AI-native apps and in-app AI in traditional apps?

AI-native apps use AI interaction itself as the primary product entry point, suitable for continuous and high-frequency use. In-app AI leverages existing application scenarios, embedding AI capabilities into specific flows like search, office work, or content. The user needs they address are not identical.

Why are in-app AI and mobile assistants still important despite the growth of AI-native apps?

Different entry points have different scenarios and permissions. QuestMobile Q3 2025 data shows in-app AI monthly per-capita usage time maintained high year-on-year growth, and mobile AI assistants reached a total scale of 535 million, indicating users haven't concentrated on a single entry point.

How will agents change the way users interact with traditional apps?

As agents gain more tool and system invocation capabilities, some operations that previously required users to open an app and complete steps may be transformed into AI understanding the need and invoking relevant services for execution. Some functions of traditional apps may therefore increasingly exist as skills or interface capabilities.

Why might website content gain new indirect reach through AI?

AI may invoke external professional content when answering and organizing information, so users don't necessarily need to directly visit the original website to access relevant info. The automotive industry case in QuestMobile's 2026 semi-annual report already shows coexistence of owned app traffic and AI indirect reach.

How should enterprise websites adapt to the evolution of AI from a search entry point to a task entry point?

Start by focusing on entity information consistency, content structure clarity, and machine readability. Simultaneously, distinguish which digital assets are content supply and which business capabilities might be standardized for future invocation. Data analysis should also progressively distinguish between direct visits and AI-driven indirect discovery and reach.