Editor's Note:

As the Chinese New Year red envelope craze, which had initially driven a surge in daily active users for AI applications, drew to a close, a newly released industry report from QuestMobile punctured the industry's illusion of prosperity with a set of stark statistics: the average 30-day retention rate for mainstream AI applications in China stands at a mere 12.8%. This means that out of 100 users lured by marketing tactics or cash incentives to download the apps, fewer than 13 will continue using them for a full month, while the remaining 87 will either quietly uninstall them or let them languish unused on their phones. The AI frenzy that began in 2025 ultimately proved to be a fleeting phenomenon. Unlike the stable retention rates exceeding 80% witnessed during the mobile payment boom, the retention curve for AI applications resembles a steep cliff, rapidly declining from a peak on the first day of download and plummeting to the bottom within just 30 days.

As an observer who has long tracked the evolution of the technology industry, I aim to dissect the structural challenges hindering AI from retaining users and explore the path toward true 'usefulness' beyond the superficial 12.8% figure.

The Retreat of the Craze and Industry Differentiation Behind the Steep 12.8% Decline in Retention Rates

The QuestMobile Generative AI Industry Monitoring Report for January-February 2026 reveals that the retention curve for mainstream AI applications in China exhibits a typical 'cliff-like drop.' Specific data shows that the average first-day retention rate for AI applications reaches 65%, buoyed by extensive marketing campaigns and red envelope subsidies from major platforms during the Chinese New Year period. However, this initial enthusiasm faded quickly, with the 3-day retention rate plummeting to 42%, the 7-day retention rate dropping to just 25%, the 15-day retention rate falling below 18%, and finally, the average 30-day retention rate settling at 12.8%—a decline of over 80% from the first-day retention rate.

More notably, even among the leading players, retention rates show significant differentiation. Report data indicates that Doubao APP, leveraging its ecological advantages, stands as the only mainstream AI application with a 30-day retention rate exceeding 40%, averaging 44.5%. Kimi and DeepSeek follow in the second tier with retention rates of 32.1% and 30.8%, respectively. In contrast, many AI applications from smaller vendors have 30-day retention rates below 5%, with some products seeing retention rates drop below 1% once red envelope subsidies end.

This data clearly portrays a typical scenario in the early stages of industry explosion: after the tide of marketing-driven users recedes, a large number of homogeneous, poorly designed products quickly expose their shortcomings, while a few products focusing on specific scenarios and delivering excellent experiences begin to build competitive barriers. The differentiation in retention rates signals that the industry is shifting from 'wild growth' in traffic acquisition to 'intensive cultivation' in value competition.

Three Core Challenges: The 'Triple Gates' of Interaction, Scenarios, and Technology

After interviewing numerous users and practitioners, I found that the reasons for users uninstalling AI applications are highly concentrated in three areas, which act as three hurdles that must be overcome to retain users long-term.

Interaction logic defies human intuition, increasing usage costs. Many AI product designs exhibit a 'technology-driven' rather than 'user-driven' approach. Users are required to input complex instructions, navigate multi-level menus to find functions, and even grant excessive privacy permissions, forming a stark contrast to the simple, 'one-click' experience of mobile payments. QuestMobile data shows that 37.2% of users have uninstalled apps due to cumbersome interactions or excessive privacy permission requests.

Demand scenarios lack focus, leading to the dilemma of being 'versatile yet useless.' Most current general-purpose AI applications claim to be all-powerful in their promotions but fail to truly and consistently address users' core pain points in any single scenario. A survey by Clawdbot revealed that 42.5% of users cited 'inability to find continuous usage scenarios' as their reason for uninstalling. While 'red envelope tasks' during the Chinese New Year temporarily boosted daily active users, user churn accelerated once the campaigns ended, exacerbating the disconnect between scenarios and functions.

Unstable technical experience erodes user trust and patience. Frequent issues such as inconsistent content quality, occasional lag and errors, and semantic understanding biases severely impact usability. Additionally, concerns over copyright risks in generated content and data privacy and security further heighten user apprehensions, particularly among users with commercial or professional needs.

Benchmarking and Reflection: The 'Usefulness Threshold' That AI Urgently Needs to Surpass

As AI applications struggle in the quagmire of low retention rates, the mobile payment industry's stable retention rates exceeding 80% offer a profound comparative example. Of course, directly comparing generative AI applications in their exploratory introduction phase with mobile payment tools in their mature adoption phase involves generational differences in technological maturity, user habits, and ecological completeness. However, the success path of mobile payments precisely provides a crucial conceptual framework for AI applications to bridge the gap from 'novelty' to 'regular use'—namely, they must surpass the 'usefulness threshold.'

Mobile payments succeeded by precisely targeting the high-frequency, rigid demand scenario of 'payments,' offering an extremely simple and smooth experience, and deeply integrating into e-commerce, social, and life service ecosystems to form a closed loop of 'payment-scenario-service,' transforming users from 'willing to use' to 'cannot live without.'

In contrast, many current AI applications fall short in these three aspects: scenarios are broad and vague, interactions are complex and inefficient, and ecological support is weak. However, not all AI applications are mired in dilemmas. Those deeply cultivated in vertical fields, such as AI assistants for lawyers, translation tools for cross-border e-commerce, and accounting software for accountants, generally have higher retention rates than the industry average because they precisely address the professional pain points of specific groups.

This fully demonstrates that the 'usefulness threshold' for AI applications does not necessarily equate to 'daily high frequency.' For productivity tools, their value may lie more in their irreplaceability or extreme efficiency improvements in critical scenarios. Therefore, the key to surpassing the threshold lies in whether they can provide stable value (quality, efficiency, experience) significantly superior to traditional solutions in specific scenarios and enable users to form reliable expectations.

Conclusion

The average retention rate of 12.8% serves as a mirror, reflecting both the industry's initial fervor and bubble and users' desire for genuine value. This 'retention battle' is essentially an ultimate test of AI products' 'glamorous technology' versus 'plain usefulness.' The journey of mobile payments shows that victory does not belong to the earliest starters but to those who can most precisely target scenarios, meticulously refine experiences, and integrate into ecosystems.

For the AI industry, abandoning the illusion of 'versatility' and thoroughly delivering value in vertical fields is the true starting point for navigating through the current retention fog and winning the future. This race, which began as a traffic frenzy, will ultimately become a protracted battle for value creation.

Differentiation has already emerged, and the real competition is just beginning.