Citibank on China’s AI Application: "High Penetration, Shallow Usage," Key Turning Point Approaching!

Citibank on China’s AI Application: "High Penetration, Shallow Usage," Key Turning Point Approaching!

```

Citigroup Research's latest survey shows that AI applications in China are at a critical transition period of "widespread adoption with shallow usage", but the behavioral traits of heavy users have already sent clear accelerating signals— the inflection point for deep AI integration into economic life is approaching, and its substantive effects on efficiency, the labor market, and corporate profitability are likely to be released more rapidly.

According to Wind Chaser Trading Desk, a special survey conducted by Citi Research Innovation Lab in March 2026 on 1,800 respondents shows that 70% use AI chatbots daily, with an average duration of 49 minutes, but only 5% are heavy users with over two hours of use per day. Citi's initial estimate is that AI brings about a 4.3% efficiency gain for respondents, which translates to about 1.8% for the entire economy.

The survey also reveals that China’s AI chatbot market is highly concentrated among leading internet platforms—ByteDance's Doubao leads with a 79% user penetration rate and a Net Promoter Score (NPS) of 54.4, far surpassing rivals like Tencent's Yuanbao (42%/28.3), DeepSeek (37%/36.1), etc. Citi notes that the diversified use cases of AI will drive users to use multiple products for different scenarios, making the risk of a single platform "winner-takes-all" lower than market fears, and that competition will remain fragmented in the long term.

On the policy front, the Ministry of Industry and Information Technology on April 10 released the "Interim Measures for the Management of Anthropomorphic Interactive AI Services," emphasizing the "people-oriented" principle and specifically addressing algorithmic exploitation, accelerating the formation of an AI governance framework. Citi expects substantial progress in deep fiscal reforms like consumption tax this year, with AI dividend distribution mechanisms gradually being established.

Widespread Adoption Surpasses Depth of Use, with "Shallow Usage" Clearly Evident

According to the China Internet Network Information Center (CNNIC) February 2026 report, by the end of 2025 there were 1.13 billion internet users in China, with an internet penetration rate of 80%. The user base for generative AI grew 142% year-on-year, reaching 602 million. However, average monthly usage is just about 183 minutes (around 6 minutes/day), making up a very small proportion of the average 4.6 hours of daily internet use.

Income data also supports this conclusion. According to Xinhua, citing Ministry of Industry and Information Technology data, core AI industry revenue in 2025 was about 1.2 trillion yuan, just 0.9% of annual GDP (140 trillion yuan); even including cloud computing and big data, it was just 1.6 trillion yuan, about 1.2%.

Citi's micro-level survey closely aligns with the above macro data. Of the 1,800 respondents who used an AI chatbot at least once in the past month, 83% used it less than an hour a day, and only 5% more than two hours—clearly showing a "high penetration, shallow usage" pattern.

Heavy User Behavior Reveals Inflection Point is Near

Citi believes that although overall usage depth is low, the behavioral traits of heavy users already point to an acceleration point.

In terms of frequency, among heavy users (over two hours/day), 89% use AI chatbots multiple times daily; among light users (less than 30 minutes/day), about half do not even use AI once every day. In terms of outlook, 46% of heavy users plan to significantly increase their usage in the coming year, compared to just 19% of light users.

The difference in usage scenarios is also noteworthy. 64% of heavy users use AI for work assistance, and 38% for creative thinking and brainstorming—far higher complexity than among light users. Citi points out that heavy users also have significantly more exposure to overseas AI products (such as ChatGPT), indicating they started using AI earlier and are typical early adopters.

The above data mutually corroborate Citi's assertion that AI adoption in China is evolving upward along a J-curve tipping point.

Preliminary Estimate: Economy-wide Efficiency Gains About 1.8%

As AI adoption is still in an early stage, Citi uses users' "trust" in AI’s answers as a proxy variable for efficiency— the higher the trust, the more time saved; low trust means time is wasted re-checking answers.

Within this framework, AI chatbots save respondents about 30.4 minutes/day on average, about 4.3% of waking hours. With a 42.8% generative AI market penetration, Citi extrapolates this efficiency gain to the whole economy, reaching a potential 1.8% improvement.

Heavy users benefit even more. Calculations show that users with over two hours of daily usage can save more than two hours per day—about 20% of waking time. At the same time, 38% of heavy users believe AI has already surpassed humans in reasoning and creativity, compared to just 13% of light users (less than 30 minutes/day). Citi believes that as usage rises—trust increases—usage scenarios upgrade, forming a virtuous cycle that will produce accelerating, non-linear productivity gains.

Dominance by Major Platforms, Specialization Determines Survival

Citi's survey shows that China's AI chatbot market is dominated by top internet platforms, but also that significant differentiation and specialized competition is emerging between products.

Doubao by ByteDance leads with a 79% user penetration rate and 63% as the "most commonly used product", with an NPS of 54.4. Tencent’s Yuanbao follows with 42% penetration and an NPS of 28.3. DeepSeek, with 37% penetration and 36.1 NPS, stands out, especially given its limited sales and marketing; Qwen by Alibaba and Ernie by Baidu follow with 24%/15.3 and 23%/18.4, respectively.

In terms of user choice drivers, Doubao wins for speed and ease of use (62% each), Yuanbao wins on trust (57%), DeepSeek for accuracy (66%) and logical reasoning (58%), while Tongyi leads for creativity and emotional tasks (a creativity score of 4.32, highest among the five).

Citi believes that AI’s diversified use cases will lead users to choose different products for different needs, making any single product becoming the “universal AI entrance” unlikely. This provides space for Alibaba, Tencent, and others to expand their AI applications within their own ecosystems, and indicates that future competition will become fiercer in vertical and agent-oriented directions.

Additionally, 45% of respondents have used overseas AI products, with ChatGPT being the most used (30%), followed by Google Gemini (13%) and Meta AI (8%). The main reasons for using overseas products are to seek “different answers” (45%), better functionality (44%), and greater accuracy (43%); penetration is notably higher among major cities and iPhone users.

Labor Market Impact Stays Emotional, Formal Adoption in Workplaces Lags

Citi’s previous macro analysis estimated that AI could impact about 31% of jobs in China, with about 9.6% (around 70 million) at direct replacement risk. But the current survey shows that this risk is still limited in practice.

In terms of sentiment, 69% of respondents believe AI will eventually replace a large amount of human jobs, but only 40% personally worry about their own jobs being affected—individual outlooks are much more optimistic than broader views. 59% believe AI can help increase income opportunities, reflecting that expectations of complementary effects dominate labor market sentiment, and overall optimism exceeds Citi’s earlier projections.

Usage scenario data shows AI is still mainly used for personal purposes: general search (71%), health consultation (51%), etc., predominate; work assistance (56%) and brainstorming (31%) rank high, but official adoption by employers is notably lagging—only 6% say their company or school designates specific AI tools, and even among heavy users, this is only 12%.

By age, those aged 18 to 24 worry most about job displacement by AI, with 58% concerned—the highest among all age groups; but this group also has the shallowest AI usage, with only 3% being heavy users (>2 hours/day). Citi notes that this matches its macro research, which found younger people and entry-level jobs more exposed to AI, thus more acutely worried about employment.

The survey shows that 58% of users prefer to keep using AI for free, in line with historic patterns of internet monetization. However, Citi believes the fact that 45% are willing to pay for advanced features reveals substantial monetization potential.

Willingness to pay increases with age, city tier, and for iPhone users. Of those willing to pay, 63% list "access to more advanced models" as their main motivator, followed by multimodal capabilities (file analysis, image generation, advanced voice interactions, 53%) and faster response speeds (51%).

The average acceptable monthly subscription is 48.3 yuan, higher than typical subscriptions for music or video content, and similar to monthly spending by online game users. Citi believes this demonstrates that users applying AI to complex tasks have a stronger willingness to pay, and that rational pricing for AI subscriptions should be higher than current market expectations.

 

~~~~~~~~~~~~~~~~~~~~~~~~

The above wonderful content is from Wind Chaser Trading Desk.

For more detailed analysis, including real-time commentary and frontline research, please join the【Wind Chaser Trading Desk Annual Membership】

Risk Disclosure and DisclaimerThe market involves risk, investment should be prudent. This article does not constitute personal investment advice, nor does it take into account the specific investment objectives, financial situation, or needs of any individual user. Users should consider whether any opinions, views, or conclusions contained herein are suitable for their particular circumstances. If you invest accordingly, the responsibility is your own. ```