AI Transition: Survey Findings Suggest Tipping Point Is Near

A new Citi Research report from a team led by Head of Pan-Asia Internet Research Alicia Yap looks at the possibility that adoption of artificial intelligence (AI) is nearing a tipping point in China, and explores the potential impacts of such a shift.

We worked with Citi Research’s Innovation Lab to run a proprietary survey on the adoption of AI consumer products in China, covering 1,800 respondents with broad representation across age groups, genders and city tiers. The survey focuses on how people actually use AI chatbots today — and what that behavior might suggest about productivity, labor market implications and policy direction. 

The picture that emerges, as explored in our Macro-to-Micro report, is one of impressive breadth but limited depth. AI is already embedded in daily life for many users, yet overall usage intensity and formal workplace adoption remain relatively shallow. At the same time, several indicators suggest China may be approaching an inflection point where adoption and impact could accelerate.

Consumer engagement is already high. Nearly 70% of respondents use AI agents or chatbots daily, with average usage of 49 minutes per day. Most users rely on one or two apps, which underscores how competitive the space has become. 

And AI is becoming an all purpose tool: General search, content generation and work assistance are the most common uses, while travel planning and health related inquiries are also widespread. Patterns vary across demographics: Students gravitate toward homework help, working age adults toward productivity tasks, and older users toward health questions. Content creation is a particularly important use case, especially image generation, with video and other formats trailing behind.

Platform backed AI chatbots dominate consumer adoption, but the market isn’t homogenous. Users tend to differentiate providers by strengths such as speed and simplicity, trustworthiness, accuracy and logic, or creative and emotional capabilities. Nearly half of respondents have also experimented with foreign AI tools, particularly in larger cities, often in search of different features or better performance.

Despite high engagement, adoption remains shallow. Average monthly usage remains modest relative to overall internet time, and AI related revenues still account for only a small share of gross domestic product (GDP). Our survey data are consistent with this: Only a small minority of respondents qualify as heavy users, spending multiple hours per day with AI. These heavy users, however, behave very differently — and their behavior may matter disproportionately.

Heavy users are more engaged, rely more on AI for work and idea generation, and expect to increase usage more rapidly than the average user. They also display higher trust in AI’s outputs and are more willing to use it for higher order tasks involving reasoning or creativity. This pattern points to a J shaped adoption curve, where early adopters drive deeper usage that eventually spreads more broadly. In this sense, current adoption may understate future momentum.

Even at this early stage, efficiency gains appear meaningful. Using trust in AI outputs as a proxy for time saved, we estimate that respondents may already be saving about 30 minutes per day on average — equivalent to some 4.3% of waking hours. Adjusted for overall penetration, this suggests an economy wide efficiency gain of around 1.8%. For heavy users, the gains are far larger, reflecting a compounding effect between greater usage, higher trust, and more advanced applications. As adoption deepens, productivity impacts could accelerate.

By contrast, the labor market effects of AI remain largely potential rather than realized. Our earlier macro work suggested that AI could affect roughly 31% of jobs in China, with around 9.6% facing displacement risk. Survey evidence indicates little of this has materialized so far, largely because AI use is still more personal than institutionally driven. Employers are increasingly aware of AI’s relevance, but formal workplace deployment remains limited.

Sentiment around AI reflects this early stage. Most respondents believe AI will replace many jobs in general, but views are more divided when it comes to personal job security. At the same time, a majority believe AI can improve earnings opportunities. This combination of substitution anxiety and complementarity optimism closely mirrors our macro framework. Heavy users tend to hold both views most strongly, while younger respondents — despite lower adoption — show the highest job insecurity, consistent with greater exposure among junior roles.

Finally, policy is beginning to catch up with technology. AI governance in China is gaining momentum, with a clear emphasis on human centric deployment and safeguards against misuse. Beyond governance, policymakers appear focused on ensuring broader distribution of AI’s gains, including measures to improve work life balance and gradual progress on fiscal reforms. Taken together, the survey suggests an economy still early in the AI transition — but one that is steadily moving toward deeper adoption, greater productivity impact and more visible labor market implications.