Knowledge Mapping and Frontier Analysis of Artificial Intelligence Applications in Mental Health Interventions: A CiteSpace‑Based Bibliometric Analysis

For decades, the world has faced a quiet crisis in mental health care. There are far more people struggling with anxiety, depression, and other conditions than there are trained therapists to help them. The shortage is global, and the waiting lists are long. In recent years, a new kind of technology has emerged to fill this gap: large language models. These are computer systems trained on vast amounts of text that can understand and generate human language with surprising fluency. They can hold conversations, offer emotional support, and even simulate the techniques of a therapist. The promise is that these tools could act as a bridge, providing immediate, accessible care to millions who currently have none. But as these digital helpers multiply, a critical question remains: are they actually working, and how do we know which ones are safe?

To answer this, a team of researchers from China and the United States decided to take a step back and look at the entire landscape of scientific literature on this topic. Instead of testing a single chatbot or running a new experiment, they gathered and analyzed 1,685 published studies from the Web of Science database, a massive collection of academic research. They used a specialized software tool called CiteSpace to map out how this field has grown, who is doing the work, and what questions are being asked. The result is a clear picture of a field that is exploding in size but still searching for its footing in the real world.

The story of this research field is one of sudden acceleration. For nearly forty years, from 1982 until the early 2020s, the number of studies published each year was tiny, often hovering between one and a dozen. Researchers were exploring the idea, but progress was slow and scattered. Then, in 2023, everything changed. That year, the number of publications jumped from 22 to 65. The growth did not stop there; it surged to 183 papers in 2024, nearly 600 in 2025, and reached 644 by mid-2026. This dramatic spike coincides perfectly with the public release of advanced generative artificial intelligence tools, like the chatbots that can now write stories and answer complex questions. The data suggests that the arrival of these powerful new technologies has ignited a frenzy of academic interest, driving the total volume of research in the past four years to account for 93 percent of all the work ever done in this area.

When the researchers looked at who is driving this explosion, they found a distinct pattern of global effort. China produced the most papers, with 534 articles, followed closely by the United States with 438. Other nations like England, Japan, and South Korea also contributed significantly. However, the map of collaboration tells a different story than the map of production. While China has the highest number of papers, its researchers are not as deeply connected to the global network as their American counterparts. The United States sits at the very center of the collaboration web, acting as a hub that links researchers from many different countries together. In contrast, Chinese research, while prolific, remains somewhat isolated in this network. This suggests that while many countries are producing work, the United States is currently the primary engine for connecting ideas and fostering international cooperation in this specific field.

The institutions leading the charge are mostly major universities in the United States and Europe. Harvard University, the University of California system, and the University of London are among the top producers of research. These groups are not just building better chatbots; they are rigorously testing them. They are investigating whether these artificial intelligence tools can accurately diagnose conditions like depression or obsessive-compulsive disorder. They are running clinical trials to see if talking to a computer program can actually improve a patient's symptoms compared to traditional care. They are also deeply concerned with safety, studying how these systems might fail when a user expresses suicidal thoughts or how they might handle sensitive cultural differences.

Despite the rapid growth in numbers, the researchers found that the field is still in its early stages of maturity. The most common topics in the literature focus on the technology itself: how to make the models better at understanding human emotion, how to train them on specific therapies like cognitive behavioral therapy, and how to ensure they do not make up facts or give dangerous advice. There is a strong emphasis on "computational modeling," which means using math to understand how the human brain makes decisions and how these digital tools can mimic those processes. Researchers are also working hard to create better ways to measure if these tools actually work, moving beyond simple tests to see if they can be used in real clinics and community centers.

However, a significant gap remains between the laboratory and the hospital. The analysis shows that while there is a wealth of technical validation and small-scale testing, there is a lack of large-scale, real-world studies that prove these tools are safe and effective for the general population. Most of the current work is still about proving that the technology can work, rather than proving that it does work for everyone in everyday life. The researchers note that issues like privacy, the risk of patients becoming too dependent on a machine, and the need for human oversight are still being sorted out. There is no single, standardized way yet to evaluate these tools, and the path to getting them approved for widespread clinical use is not yet clear.

The authors conclude that the field is currently driven by the excitement of new technology, but it needs to shift toward building a solid foundation for the future. They suggest that the next step is not just to build more models, but to bring researchers from different countries and disciplines together to create large, long-term studies. They call for the development of clear rules for how humans and artificial intelligence should work together in therapy, ensuring that the technology supports the patient without replacing the essential human connection. The data shows a field that is growing faster than its infrastructure, a place where the potential is immense, but where the journey from a promising idea to a reliable, life-saving tool is still very much underway.