Being an influencer involves more than snapping photos of pretty places or taking selfies to show off the nice swag your sponsors have sent you. It’s also about building an online community. And a key part of that is interacting with your online audiences.
The most successful influencers, however, often receive hundreds or even thousands of comments on each of their posts. Responding to every single one is an impossible task.
In 2024, the Chinese social media platform Weibo offered its top influencers the chance to use an AI agent that would automatically generate replies from the influencer’s account. Yang Gao, an assistant professor of business administration at Gies College of Business at the University of Illinois Urbana-Champaign who studies social media platforms, was curious how Weibo users would respond.
“The answer to the question isn’t certain,” Gao says. “Let’s face it, we human beings most of the time don’t like AI, especially when it comes to social media engagement. It’s supposed to be human-to-human interaction. Why should we bother to introduce AI?”
Along with three colleagues at the McIntire School of Commerce at the University of Virginia, Maggie Mangqing Zhang, Jingjing Li, and Steven L. Johnson, Gao examined Weibo users’ reactions to AI-generated responses to their comments. They recently published their findings in a paper called “When Influencers Delegate Replies: How Social AI Agents Shape User Engagement,” forthcoming from Information Systems Research.
Weibo had 260 million daily users in 2024. As on American social media platforms, influencers form a key part of its ecosystem. Gao and his fellow researchers compiled a dataset of influencers who would be eligible to use Weibo’s Social AI Agent to reply to comments on their posts. If they decided to use the Social AI Agent, the influencers can still respond to users themselves, but the AI-generated replies would be clearly labeled with an “AI Agent” tag.
After studying the posts and comments of the top influencers from the summer of 2024, the researchers identified 225 influencers who had activated the Social AI Agent feature. (Gao suspects that more have adopted it since then.) They studied the interactions between the influencers, their audiences, and the AI. They then compared the reactions between users who received responses to their comments from the Social AI Agent and those who had received no response at all to see if users who received AI-generated replies continued to comment on the influencer’s subsequent posts.
How AI-Generated Replies Impact User Engagement
Much to the surprise of the researchers, users didn’t seem to mind the AI-generated responses. In fact, users who received responses from the Social AI Agent tended to become even more enthusiastic about interacting with the influencer. This was true for both sponsored and non-sponsored posts. But further analysis showed that there were a few conditions under which the Social AI Agent was most successful.
First, the style of the AI-generated replies needed to be aligned with the usual style of the influencer, and they needed to be relevant to the comments they were responding to. The paper includes an example: in response to an influencer’s post about their obsession with a mixed carrot juice, one follower writes, “I like the tomato flavor.” The AI agent replies, “Wow, the tomato flavor is great, too! You should try something different next time—there might be a pleasant surprise!”
Second, the timing of the responses was important. “If it’s too fast,” Gao says, “people will think, ‘Okay, it’s definitely a machine, it’s horrible.’ No human being can generate a reply in one second.” But if the response came too late, the conversation would have already moved on.
Finally, it was better if the AI agent was selective in its replies. Among the influencers that Gao and his team studied, the AI response rate was about 5 percent. “If everyone gets the AI replies, it seems like the influencer is too desperate,” Gao explains. But if a response remained a relatively rare occurrence, users were pleased: “It’s like, in some sense, the influencer is involved. They know that the reply is generated by AI, but they still feel the human touch. That’s just how human minds work.”
Gao and his team didn’t have enough data to determine the optimal response time and rate, but they’ve identified this as an avenue for further study. They plan to present their findings to Weibo later this year. Gao thinks the company will be happy with the results and hopes it will collaborate on future work.
This research builds on some of Gao’s previous work, which shows that while AI chatbots can increase initial interaction on platforms, they do not automatically lead to deeper or more meaningful engagement. Gao has shown their effectiveness depends heavily on how natural, context-aware, and personalized the interactions feel to users. According to his 2025 paper, “Does Social Bot Help Socialize? Evidence from a Microblogging Platform,” when chatbot responses come across as generic, intrusive, or poorly timed, users are less likely to continue engaging.