Based on the inclusion criteria, the included studies from the search were listed in Table 2.
The results of this study are shown in Table 3a,d. Average utilization was calculated for each individual category. For surveys where respondents answered in an open-ended format regarding examples of AI utilization, the maximum value corresponding to mental health utilization was calculated.
Additionally, we present a weighted average of the estimated percentage of the population using AI for mental health, as shown in Figure 2 below.
Estimated proportion of the population using AI for mental health, by participant type and sample size.
In addition to percentage usage, we also investigated willingness to use AI chatbots for mental health support (see Table 4).
Table 4. Willingness to use AI for mental health support. Surveys that do not include a time of survey are dated based on the time of publication;
.
additional research
Three studies could not be coded in Table 2 due to different ways of measuring the use of AI for mental health, so we discuss these two studies here.Claude chatbot’s parent company, Anthropic, conducted a 2025 study using a collection of 4.5 million conversations users had with Claude.3
. The researchers first identified and flagged phrases and words that they determined counted as “emotional conversation.” Of the 4.5 million, 131,484 emotional conversations with Claude were found. This revealed that 2.9% of all conversations with Claude were emotional.OpenAI collaborated with MIT to conduct a 2024 study examining the user experience of ChatGPT.10
. The authors collected 4,076 responses, of which 2,333 were power (frequent) users and 1,743 were control (average) users. The authors then created classifiers (loneliness, vulnerability, self-esteem, problematic use, and potential dependence) that could identify users who perceived their conversations with ChatGPT to be emotional. The results show that over 18% of power users reported using expressions of affection in ChatGPT, less than 4.5% expressed a desire for emotion, and over 5% indicated they sought support. These same values among control users were <18%, <2.5%, and <12%, respectively.A recent and widely read study was published by Harvard Business Review in 2025.1
. For further information regarding the methodology of this study, the authors of this study were contacted and further details of the study were provided through personal communication. This study used Reddit forums to track and collect the use of AI in mental health. In this latest publication, researchers noted that mental health support will be the most common use case for AI by 2025.
harmAmong surveys that rated or asked respondents about harmful incidents caused by AI chatbots, the most common concerns and harms cited were data security and privacy. Albikawi et al. conducted a study in February 2025 and measured victimization using participant perception scales. The results showed that data storage and data privacy were cited as concerns by participants. Furthermore, the study by Rousmaniere et al. 9% of respondents reported harm or inappropriate responses from AI chatbots, and less than 1% noted that the AI encouraged harmful behavior. A recent study in the UK found that 11% of people using AI to support MH experienced the induction or worsening of psychotic symptoms, 11% were given harmful information about suicide, 11% felt anxious or depressed after using AI, and 9% experienced induction of self-harm or suicidal thoughts.11. Additional research also found that 41.2% of respondents reported receiving incorrect advice from a chatbot.12. Three surveys also cited privacy and security concerns, with the number of respondents reporting concerns ranging from 37% to 77%.13, 14, 15 . Finally, an additional survey stated that 27% of respondents expressed concerns about AI, but did not clarify or define “concerns.”16
