ChinAI #331: Chinese Public Perceptions and Usage of AI (2025 survey)
Greetings from a world where…
The Farseer Trilogy is epic
…As always, the searchable archive of all past issues is here. Please please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay support access for all AND compensation for awesome ChinAI contributors).
Feature Translation: Chinese Public Perceptions and Usage of Generative AI
Context: Last month, to explore the public’s usage habits and attitudes toward generative AI, Tencent Research Institute conducted an online survey of 3,570 Chinese adults (link to original Chinese). This is not a nationally representative survey; rather, it captures the views of highly educated young adults. To make this clear, 85% of the survey respondents had a college degree or higher, whereas just 10% of China’s labor force (age 25-64) has at least a bachelor’s degree. They did collect a gender-balanced sample from a variety of industries, so the below findings probably reflect the views of young, highly educated digital natives in China.
Key Takeaways: Adoption patterns — wide, localized, but “wait-and-see” attitude regarding willingness to pay. Let’s go through these three points in turn.
96% of respondents had used AI-generated content tools, with two-thirds (67.7%) of those being daily users.
Chinese companies accounted for the top three most-used products: ByteDance’s Doubao, DeepSeek, and Tencent Yuanbao. Tencent Research Institute attributes this trend to network access restrictions (e.g., you have to find mirror sites or use a VPN to access Claude or ChatGPT), superior Chinese-language capabilities of local services, and synergies with the ecosystem of Chinese applications (e.g., WeChat’s integration of DeepSeek)
I was disappointed that the survey results don’t show the actual bar graphs for this question; instead, they just provide a word cloud, which omits the exact figures for the most popular generative AI apps.
Two more interesting points about age groups and GenAI product usage. “Among respondents aged 20-29, ChatGPT’s usage rate was significantly higher than among respondents in other age groups,” the article states. And, Moonshot AI’s Kimi places in the top three among respondents under 20.
In the section on commercialization prospects, the Tencent Research Institute survey found that about 60% of respondents stated that they were “watching from the sidelines [正在观望], but would pay if it’s good” (see image below). Among the users that were open to paying for generative AI services, the most preferred payment option was “monthly payment” and under 100 RMB/month. The study authors theorizes that consumers see the generative AI product market like video streaming services (for context, the monthly fee for iQiyi [China’s Netflix] is around 25 RMB/month).
The Chinese public has mixed views about AI’s societal impact.
About 46% of respondents feel “both excited and concerned” about the future of AI. The three main risks: 1) 60% of respondents expressed concern about the “proliferation of false information and fake news”1; 2) 60% also fear job displacement; and 3) 47% cited the risk of “privacy breaches.”
One item that caught my attention relates to the so-called “enthusiasm gap” between the Chinese public and Western publics, which is worth unpacking further. Overall, the majority of respondents (72%) believe that generative AI will have a “primarily positive impact.” This seems to support this overall assumption that the Chinese public is more optimistic about AI than people in other countries.
But, what happens when we dig a little deeper into this purported optimism gap? Again, I would be very skeptical about taking online surveys from China as representative of the entire population. For instance, one survey cited as evidence for this gap finds that 81% of the Chinese public agrees that the benefits of AI outweigh the risks (the figure was 41% for the US). However, turn to page 75 of the methodology section, and you’ll see this revealing statement:
“Samples from emerging economies (Brazil, India, China, and South Africa) represented considerably more university educated people than their respective general populations (using OECD 2021 education data as a comparison). A higher representation of educated people is common in survey research from the BICS countries. For instance, Edelman (2022) and Ipsos (2022) both note that online samples in Brazil, India, China and South Africa are more educated, affluent, and urban than the general population.”
To me, here’s the bottom line. Regarding findings from online surveys like this week’s feature translation and the above one, they reflect the views of highly educated adults in China, very likely concentrated in the coastal provinces — not the general Chinese public. Thus, I would be very wary of using these survey findings to support the claim that AI will diffuse more quickly throughout the entire Chinese economy.
FULL TRANSLATION: Chinese Public Perceptions and Usage of Generative AI
ChinAI Links (Four to Forward)
Should-read: Compute is not the answer to AI sovereignty
A GovAI summer fellow, Hamish Low, has published a thoughtful piece about how the UK should approach its compute buildout and AI sovereignty. He concludes, “Given a sufficiently focused and aggressive industrial policy we can build powerful positions in key future nodes of the AI value chain and ensure that Britain stays in the room when that crisis comes. (Mutual) Interdependence (with the U.S.) can be much more powerful for the UK than compute infrastructure.”
Should-read: The U.S. Needs A Generative AI Intensity Index
This is a very cool project from SeedAI that tackles the challenges in measuring the diffusion of generative AI. Joshua New, Marina Meyjes, and Austin Carson propose an index to “measure the digital processes created during real-world use of generative AI systems, offering a direct proxy for their adoption across the economy.”
Should-read: Grant Delays Threaten Cultural and Language Studies Programs
On September 30, the Department of Education ended seven decades of (bipartisan) support for initiatives such as the National Resource Center (NRCs) program, dedicated to language and area studies. GW has two NRCs, including one that supports students to pursue intensive language study in East Asia. Again, what are we doing here? Reporting for Inside Higher Ed, Johanna Alonso provides the broader context.
Must-read more closely later but putting it here now: DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning
DeepSeek researchers published an open access Nature paper that details their process for training the R1 model. H/t to Miles Brundage for flagging that it includes a 10-page safety report that talks about their efforts to control risks. More on this in future issues.
Thank you for reading and engaging.
*These are Jeff Ding’s (sometimes) weekly translations of Chinese-language musings on AI and related topics. Jeff is an Assistant Professor of Political Science at George Washington University.
Check out the archive of all past issues here & please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay for a subscription will support access for all).
Also! Listen to narrations of the ChinAI Newsletter in podcast format here.
Any suggestions or feedback? Let me know at chinainewsletter@gmail.com or on Twitter at @jjding99
Tencent Research Institute recently published a report on AI-generated misinformation, which ChinAI translated: https://chinai.substack.com/p/chinai-322-100-cases-of-ai-disinformation?utm_source=publication-search
Thank you - your suggested reads are also really great. The Hamish Low article is super.
"I would be very wary of using these survey findings to support the claim that AI will diffuse more quickly throughout the entire Chinese economy."
Strongly agree.
I've worked with survey companies in China before - massive problem that online survey takers tend to be significantly more educated and higher-income than the national median. In fact, many survey firms divide the population into income groups (lower/middle/upper), and the "low-income category" is <10,000-12,000 Yuan a month (which is around the 80th percentile nationally)!
When you do weight by age, I've always guessed that neglected populations (esp. rural elderly) are often fake respondents (people pretending to be in that demographic) for fairly obvious supply and demand reasons. Even if they aren't, then they're unrepresentative in other ways, because you have to be a very distinctive 70 year-old villager to be accessing these online survey platforms.
Unless someone's on the ground interviewing face-to-face, you have to assume that you're always missing at least the "bottom half" of China's population in surveys.