Artificial intelligence chatbots are changing the way people seek help for anxiety, depression, and other emotional struggles. Recent research shows that while Americans prefer continuous conversation and dislike automated systems’ use of emojis, Chinese users readily accept this technology regardless of their communication style. The research results were published in Computers in Human Behavior.
Digital tools allow people to feel comfortable discussing sensitive information by providing anonymity and immediate responses. Even common platforms not intended for medical use are becoming everyday mediums for people to share their daily stresses and deeper psychological concerns.
With so many people turning to these digital tools, researchers are trying to understand why people find non-human conversation helpful. Text-based interactions lack the physical cues found in regular conversations, such as a sympathetic tone of voice or a warm facial expression.
Emojis are digital images that attempt to bridge this gap by adding an emotional tone to text. A smile may make a casual text seem friendly, but it can also make professional advice seem less trustworthy. In high-stakes situations involving personal well-being, users may request a more serious tone.
Another factor is how people seek help, which psychologists link to different ways of coping with stress. Emotion-focused coping occurs when people vent or share their deep personal struggles in order to manage their negative emotions. Problem-focused coping occurs when a person gathers facts and asks specific questions to find direct solutions.
Finally, automated systems can maintain different levels of true conversation. Some interactions are just one turn, where the user asks a question and the system returns one answer. Others are multi-turn interactions in which the system remembers context and responds to follow-up questions over longer periods of time.
Jihye Lee, a communication researcher at the University of Texas at Austin, led a study examining how these three factors change user trust. The research team included Zinan Darren Yang and Weijia Shi from the University of Texas at Austin, and Yan Liu from Shanghai University.
They set up an online experiment that aimed to compare responses in two different cultural settings. The team attracted 394 participants from the United States and 401 from China.
Researchers created simulated transcripts featuring humans speaking to ChatGPT about mental health concerns. To make the conversation realistic, the team sourced the first human messages from real posts on Reddit and popular Chinese support forums.
The team tweaked the transcripts in three ways before showing them to participants. First, we told the system whether to include emojis in the response or use plain text.
Next, we changed the types of user prompts to reflect the two different coping styles. Half of the recordings showed humans sharing vulnerable feelings. The other half shows humans asking direct questions about dealing with anxiety and depression.
Third, the researchers varied the depth of the conversations. Some participants read out a single interaction between human and machine. Others read an expanded dialog in which a human asks additional questions and a machine provides step-by-step guidance based on previous answers.
After reading the assigned transcript, each participant completed the questionnaire. They rated the quality of the information, the emotional support they felt the machine provided, and the likelihood of using the system themselves.
The results revealed significant differences in how users in the two countries evaluated interactions. For American participants, the presence of emojis had a significantly negative effect.
When the machine used emojis, American users gave the answers lower ratings for quality of information. Emojis also tarnished the machine’s reputation when the humans in the transcription shared deep emotional struggles. In highly emotional scenarios like this, using emojis made Americans less likely to say they would use the service.
However, interactivity turned out to be a very positive feature for the American group. Transcripts showing continuous back-and-forth conversations received higher user ratings in all categories.
American participants felt that extended interactions included higher quality information and provided more psychological comfort. The longer conversation style also increased the willingness to recommend tools to friends in need.
The reactions of Chinese participants told a completely different story. No matter how their records were changed, their survey scores remained completely stable.
Chinese users didn’t care whether the machine used emojis, what prompts started the conversation, or how many turns the conversation took. Of all groups tested, Chinese participants consistently reported the most favorable opinions of technology.
They also showed the highest overall intention to use digital tools for their own health needs. The researchers proposed that Chinese users may interpret digital interactions based on the entire context, rather than small stylistic details like digital grimaces.
China has also largely integrated automated services into its national infrastructure and daily life. This high degree of social integration may give Chinese users a stronger baseline of trust in digital tools compared to American users.
The authors noted several limitations to their study. They chose to use a common conversational model rather than a clinical tool that has passed rigorous medical testing.
The experiment only measured subjective opinions about the conversation. It was not tested whether talking to a machine actually reduces symptoms of depression and anxiety over the long term.
The team also faced translation hurdles when preparing learning materials. The artificial intelligence naturally gave longer and more detailed responses in Chinese than in English.
To keep transcripts comparable in length and meaning, the researchers translated the English responses into Chinese using ChatGPT and then had three bilingual Mandarin native speakers review the translations to verify semantic accuracy and contextual relevance. Future research will need to find new ways to precisely match languages without losing cultural nuances.
Because Chinese participants already had very high baseline levels of trust, their survey responses often reached the top of the measurement scale. This high starting point may have made it difficult for researchers to detect small changes in opinion.
In future projects, the researchers hope to test whether having more emojis makes a difference. While using one digital smile may be acceptable, using 12 digital smiles can seriously compromise the reliability of the system.
Our findings suggest that technology companies designing medical applications should avoid a universal approach. The communication styles programmed into these services may need to be specifically adapted to the cultural expectations of their users.
The study, “AI Chatbots in Mental Health: How Emojis, Prompt Types, and Interactivity Shape User Perceptions in the United States and China,” was authored by Jihye Lee, Zinan Darren Yang, Weijia Shi, and Yan Liu.
