Welcome to the forefront of conversational AI as we explore the fascinating world of AI chatbots in our dedicated blog series. Discover the latest advancements, applications, and strategies that propel the evolution of chatbot technology. From enhancing customer interactions to streamlining business processes, these articles delve into the innovative ways artificial intelligence is shaping the landscape of automated conversational agents. Whether you’re a business owner, developer, or simply intrigued by the future of interactive technology, join us on this journey to unravel the transformative power and endless possibilities of AI chatbots.
Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
npj Digital Public Health has APC waivers available that can be allocated upon acceptance on an ad-hoc basis. For additional information, contact the Journal Publisher, Tony Chen.
npj Digital Public Health volume 1, Article number: 21 (2026)
Reports suggest 3–70% of AI users utilize AI for mental health. A more precise estimate is needed to guide public health policy, research, and to understand the limitations of producing an estimate. We approximate that 27% of AI users utilize AI for mental health support. However, we acknowledge that heterogeneous data sources and a lack of nosology standardization confound estimates. Consensus building will be key before better prevalence estimates are possible.
The recent surge in interest in artificial intelligence (AI) for mental health is unmistakable. Widely cited popular press reports note that activities like emotional support, companionship, and therapy may be the single most popular use of AI, and numerous academic editorials have noted a paradigm shift in action1. Yet while there certainly is increased use and interest in AI, it is unclear exactly how many people are using it and for what. While there are well-known cases of overuse, overreliance, and addiction that have been reported in association with what is often reported as AI-psychosis and even deaths by suicide, these more extreme cases are hopefully rare and not representative of use patterns by the broader population. For the broader population, actual use patterns and harms are not well established2. By some estimates, only 3% of AI users may be using AI for mental health-related support, and yet by others it may be as high as 70%3,4.
The rapid expansion of AI, along with the impressive updates in clinical potential, presents a challenge in assessing use patterns. Simply put, what AI can do today is very different from what it could offer 12 months ago (and presumably what it may be able to offer 12 months from now), and so the use cases around therapeutic support will also be different. A further challenge to quantifying use patterns is the blurred lines between a therapeutic conversation with AI and therapy itself5. While formal definitions of therapy exist, they are less actionable with AI today as some chatbots may already be delivering what is defined as therapy and yet denying such to avoid regulatory scrutiny6. Others may be claiming to offer therapy and yet not delivery evidence based. All chatbots will change and evolve so that even if we could probe if they did or did not offer what is defined as therapy, such an assessment may not reflect what the chatbot does tomorrow. Against this challenge backdrop, there are reports that many AI users are turning to AI for companionship, psychoeducation, and various degrees of mental health support ranging from skills practice to perhaps even therapy-like interactions7,8. Assessing the actual interest and uptake is important to assess how AI users are utilizing these tools, the extent of their reach, and the scope of their potential health impacts. While the rapidly changing nature of AI and the lack of definitions around mental health support delivered by AI preclude a systematic review, this review offers a feasible and transparent methodology to provide an initial estimate while outlining the barriers to solve for achieving more precise estimates in the future.
Assessing the prevalence and use cases for mental health AI requires non-traditional methods as there are no well-established definitions of mental health AI and most reports and results are presented outside of the peer-reviewed literature, with large companies like OpenAI and Google sharing their AI research/results as blog posts. We thus used a snowball approach by searching for the terms ‘AI for mental health’, ‘Survey of AI therapy use’, and ‘AI therapy’ in Google, Google Scholar, and PubMed on August 25, 2025. We followed the references and citations of relevant papers in an attempt to assess sources of AI use for mental health across both academic and non-academic literature. Due to AI’s rapid development and increasing adoption, an updated search was conducted on April 20, 2026, using the same snowball approach and search terms. See Table 1 for specific Inclusion/Exclusion criteria. Of note, we excluded surveys conducted before January 1, 2024, given that changes in AI in the last 24 months make those prior results less relevant for understanding how AI may be used today in early 2026. Unfortunately, many AI use surveys do not follow scientific standards or basic reporting methodology, and as such are often difficult to interpret and preclude a search consistent with standard systematic review methods.
Any results reported in the popular press (i.e., non-peer-reviewed surveys of use), or data from scientific research studies found via these search strategies and which fit our inclusion criteria were included in this paper. The initial search conducted in August 2025 led to the screening of 61 papers from an original total of 27,000 records. After screening, only 13 records fit our inclusion criteria and were included in the final review. After the updated search in April 2026, an additional 6 records were found to fit the inclusion criteria and have been included.
Given the already discussed limitation of a lack of standardized nosology or even means to verify what aspects (if any) of therapy or well-being support are being offered, we found that many of the surveys created and followed their own definitions of AI use for mental health. Therefore, as part of our methodology, we collected a list of terms that have been used by the surveys identified in this paper and categorized them into groups (see Fig. 1). We developed these groups based on the work of Siddals et al. and the groupings characterized by Anthropic3,9. Siddals et al. categorized AI uses into four groups under the larger umbrella of ‘mental health’: Joy of Connection, Emotional Sanctuary, The AI Therapist, and Insightful Guidance. In addition, Anthropic created groupings which included the following as ‘mental health’: Companionship, Psychotherapy or Counseling, Interpersonal Advice, and Coaching. Given that ‘companionship’ was often defined by the surveys as a form of emotional sanctuary, “Joy of Connection” or “Companionship” has been merged with ‘Emotional Support’ under our groupings, Therapy, Emotional Support, and Wellness Support9. We categorized papers/surveys based on this approach using these two sources as models and also considered harm (if reported).
Therapy, Emotional Support, and Wellness Support. All categories were considered to fall under the umbrella of Mental Health Support. Figure created using Canva.
Additionally, due to the variety of survey methods, population types sampled, sample size and specification of mental health uses, we present broad summaries of the data in a standardized format in Table 3a–d. Unfortunately, three papers/surveys varied in their methods to such an extent that we could not code them or include them in the table. These papers/surveys have been described separately in the results.
In addition to specific use cases, we examined willingness to use AI for mental health as many surveys and results shared data on this topic. This information can supplement the findings from actual use surveys, allowing us to potentially elucidate a percentage range of AI users currently using AI for mental health support and who may want to. The inclusion/exclusion criteria used for the primary search were also applied to the willingness search.
Based on our inclusion criteria, we have listed the surveys included from our search in Table 2.
The results of this study are presented in Table 3a, d. Mean use percentages were calculated for individual categories. For surveys in which questions regarding AI use case were free response, the largest value that fits as a mental health usage has been used in calculations.
Additionally, we present the weighted average for the estimated percent population using AI for mental health, as shown in Fig. 2 below.
Estimated percent of population using AI for mental health by participant type and sample size.
In addition to using percentages, 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 did not list time of survey conduction have been dated based on time of publication and are listed with an (*).
Due to the variation in methods of measuring AI use for mental health, three studies could not be coded for in Table 2, thus we present a discussion of those two studies here.
Anthropic, the parent company of Claude chatbot, ran a 2025 study using their collection of 4.5 million conversations users have had with Claude3. Researchers first identified and flagged phrases and words that they determined would count as an “affective conversation.” Of 4.5 million, they found 131,484 affective conversations with Claude. From this, they identified that 2.9% of all Claude conversations were affective.
OpenAI in collaboration with MIT, conducted a 2024 survey investigating users’ experiences with ChatGPT10. The authors collected 4,076 responses, of which 2333 were power (frequent) users, and 1743 were control (average) users. The authors then created classifiers (loneliness, vulnerability, self-esteem, problematic use, and potentially dependent), for which they would be able to identify which users perceived their conversations with ChatGPT to be emotional. The results show that of power users, ≥18% reported using affectionate language with ChatGPT, <4.5% expressed desire for feelings, and ≥5% 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 20251. The author of this study was contacted for more information regarding the methodology of this study, and additional details of the study were provided by personal communication. This study used Reddit Forums to track and collect uses of AI for mental health. In this most recent publication, the researchers noted that in 2025, mental health support is the most common use case of AI.
Among the surveys that measured or asked respondents about harmful incidents with an AI chatbot, the most common concern and harm listed was data security and privacy. Albikawi et al. conducted a survey in February of 2025 and measured harm using a participant perception scale, which found that data storage and data privacy were listed as concerns by participants. Additionally, a study by Rousmaniere et al. noted that 9% of respondents reported harm or inappropriate responses from an AI chatbot and that in less than 1% of those cases did AI encourage harmful behavior. A more recent survey conducted in the UK reported that of individuals who use AI for MH support, 11% experienced triggering or worsening of psychosis symptoms, 11% were given harmful information about suicide, 11% felt more anxious or depressed after AI use, and 9% experienced triggering of self-harm or suicidal thoughts11. An additional study also identified that 41.2% of respondents reported that a chatbot gave them wrong advice12. Three surveys also note concerns about privacy and security, with the number of respondents reporting concerns ranging from 37–77%13,14,15. Lastly, an additional survey mentioned that 27% of respondents expressed concern about AI, but did not clarify or define “concern” 16.
While an exact estimate remains impossible, our review of the popular press and academic literature on AI use for mental health suggests that ~27% of AI users may already be turning to AI for support (weighted total, 30% unweighted), and perhaps 40% are interested in and open to using it. Despite limitations in arriving at such an estimate, our results indicate that there is actual interest and use behind the current discussion about the impact of AI on the mental health space. Overall, use of AI for mental health appears to remain between 20–40% in sample sizes ≤30,000. Additionally, our results demonstrate that of individuals who use AI for mental health, 24.5% choose friendship/connection, 27.7% list advice/guidance, and 23.3% pick general emotional or mental support as the primary reason for using AI for (see Table 3c).
Our results are derived from studies conducted across the world and across a range of age demographics. Interest and use of AI appear high with results from Australia, Canada, India, Indonesia, Jordan, the UK, and the US suggesting the global nature of AI for mental health. We found that 21% of youth are using AI for mental health. This is consistent with previous research indicating that individuals of Generation Z are the largest population of AI users16,17. While there is the strongest interest among younger generations, there is a notable lack of data on older adults and people with diagnosed mental illness. Extrapolating results of college students and teenagers to these other populations is tenuous and highlights the need for more targeted research exploring the interest and use of AI for older adults and people with mental illness. Although other generations are not clearly specified in most of the surveys, our research does, however, indicate that ~36% of adults using AI have, or are using it for mental health support (see Table 3b).
Beyond results suggesting rates of AI use for mental health, our paper highlights the fragmentation and decentralized nature of research on AI uses for mental health. Without consensus on what mental health AI is and what use entails, it will remain challenging to understand actual uptake and engagement, as well as regulate and improve it. The lack of consensus on basic metrics, for example, what engagement is, has significantly harmed the credibility of the mental health app space and has raised issues for both payers and users18. The mental health AI field has an opportunity to avoid similar mistakes by either unifying clear definitions of AI wellness vs AI mental health support and what constitutes meaningful engagement19.
As noted above, our results are limited by challenges around the available data. Most surveys define terms like “Mental Health”, “wellness”, “companionship” differently or not at all. For example, Anthropic defined “affective conversations” as ranging from psychotherapeutic conversations, emotional support seeking, practicing conversational skills, wellness, and well-being, while another survey did not specify a definition of “mental health”16. A related issue for defining AI use for mental health is that even users cannot agree on clear roles, with one study conducted by OpenAI indicating that heavy AI users may be more likely than control users to see ChatGPT as a ‘friend.’ These data are in line with other research which shows that generative AI use and its meaning vary by cohort and generation 20.
Survey methods also create barriers to interpreting our results. A lack of consensus among definitions and sample demographics makes it impossible to cross-check or compare the findings from one survey to those of another. Most studies we reviewed were results of online samples or social media-based polls, which do not verify human responders. This is notable as recent research suggests that 30-50% of responses in such online surveys may be AI bots, which raises issues around the veracity of these sources17. In addition to the novel challenges posed by a lack of consensus around definitions of AI, classical research issues like selection bias with many surveys recruiting on social media and smaller sample sizes that may not generalize make it difficult to develop a single, representative number of AI usage among the public.
Because of the limitations of the data on mental health AI, as discussed above, a systematic review was not possible. Additionally, given the rapidly evolving nature of AI, there are likely papers that have been published since the writing of this work. For example, OpenAI recently released a paper that states that only 1.9% of ChatGPT conversations were about “relationships and personal reflection,” while Anthropic also published a report in September, citing “personal relationship advice and life guidance” are overrepresented in the United States by 1.34x18,19. As data continue to accumulate in this area, it is very likely that prevalence estimates will change, as well as the data summarized here (use, willingness, and harm).
As discussed already, a core limitation of this study was the lack of consensus of what ‘mental health support’ means or looks like in practice. Although there are broad definitions of mental health support published by international health organizations such as the World Health Organization, there is no one definition that is universally understood or used in the AI mental health realm20. Because of this, it is impossible to identify what definition of ‘mental health support’ survey authors intended when writing survey questions and what definition responders intended when answering questions. Thus, this review is limited in scope and ability to compare survey responses, further highlighting the difficulty in deducing a single value for how many AI users are using AI for mental health support. These issues exemplify the pressing need for a standardized meaning of mental health support and scientific research on AI chatbot use cases.
Finally, it is important to note that given the fast-moving nature of AI development, it is possible that new survey data have been published since our search was conducted.
The numerous issues discussed in this review, such as bot responses, no standardized polling method, and lack of consensus on an operationalized definition of mental health support, notwithstanding, a current estimate based on the available data suggests that ~27% of individuals who use AI have used it for mental health support. While our data is complex to interpret because of the aforementioned challenges, the expanding interest in AI uses for mental healthcare indicates its potentially broad reach and use.
Data collected and used as presented in Tables 2 and 3 a-d are accessible via the provided DOIs listed in the references.
Zao-Sanders, M. How People are Really Using Generative AI Now. Harvard Business Review. (2025).
Kusche, I. Possible harms of artificial intelligence and the EU AI Act: fundamental rights and risk. J. Risk Res. 1–14 (2024).
McCain, M. et al. How People Use Claude for Support, Advice, and Companionship. June 26, 2025. https://www.anthropic.com/news/how-people-use-claude-for-support-advice-and-companionship.
Robb, M. B. & Mann, S. Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions. Common Sense Media. (2025).
Lieberman, H. I’m a Therapist. ChatGPT is eerily effective. The New York Times. August 1, 2025. https://www.nytimes.com/2025/08/01/opinion/chatgpt-therapist-journal-ai.html.
Merriam-Webster. “Therapy”. https://www.merriam-webster.com/dictionary/therapy. Accessed December 22, 2025.
Heritage, S. ‘I felt pure, unconditional love’: the people who marry their AI chatbots. The Gaurdian. July 12, 2025. https://www.theguardian.com/tv-and-radio/2025/jul/12/i-felt-pure-unconditional-love-the-people-who-marry-their-ai-chatbots.
Quiroz-Gutierrez, M. Gen Z is increasingly turning to ChatGPT for affordable on-demand therapy, but licensed therapists say there are more dangers many aren’t considering. Fortune. June 1, 2025. https://fortune.com/2025/06/01/ai-therapy-chatgpt-characterai-psychology-psychiatry/.
Siddals, S., Torous, J. & Coxon, A. It happened to be the perfect thing: experiences of generative AI chatbots for mental health. npj Ment. Health Res 3, 48 (2024).
Article PubMed PubMed Central Google Scholar
Top Generative AI Statistics for 2025. Salesforce. Initial report published in 2023, updated report in February 2025.
George Mason University. New survey explores the promise and peril of using AI for managing stress, anxiety, and other mental health needs. December 16, 2025. https://publichealth.gmu.edu/news/2025-12/new-survey-explores-promise-and-peril-using-ai-managing-stress-anxiety-and-other.
Kwon, A., Draper, D., Sandalow, M., & Lewis, J. Survey Shows Widespread Use of Apps and Chatbots for Mental Health Support. Bipartisan Policy Center. April 16, 2026. https://bipartisanpolicy.org/article/survey-shows-widespread-use-of-apps-and-chatbots-for-mental-health-support/.
Wigmore, S. The rise of AI as a source of emotional support. Kantar. July 17, 2025.
How can young people create economic value and reap emotional value through AI? Let’s take a look at this “2025 Generation Z AI Usage Report.” Just So Soul Research Institute. March 31, 2025. https://mp.weixin.qq.com/s/4eqv2HntKFex3IcHvbMpDQ Smith, Wendy. How Americans Use AI by Generation. SurveyMonkey. February 18, 2025. https://www.surveymonkey.com/curiosity/ai-trends-by-generations/.
McBain R. K. et al. Use of Generative AI for Mental Health Advice Among US Adolescents and Young Adults. JAMA Netw Open. 2025. https://doi.org/10.1001/jamanetworkopen.2025.42281.
Montero, A. et al. KFF Tracking Poll on Health Information and Trust: Use of AI for Health Information and Advice. KFF. March 25, 2026. https://www.kff.org/public-opinion/kff-tracking-poll-on-health-information-and-trust-use-of-ai-for-health-information-and-advice/ (2026).
Zhang, S., Xu, J. & Alvero, A. Generative AI Meets Open-Ended Survey Responses: Research Participant Use of AI and Homogenization. Sociological Methods Res. 54, 1197–1242 (2025).
Article Google Scholar
Chatterji, A. et al. How people use ChatGPT. NBER Working Paper No. 34255. (2025).
Appel, R. et al. The Anthropic Economic Index report: Uneven geographic and enterprise AI adoption. September 15, 2025.
Mental Health. World Health Organization. October 8, 2025. https://www.who.int/news-room/fact-sheets/detail/mental-health-strengthening-our-response.
Albikawi, Z., Abuadas, M. & Rayani, A. M. Nursing Students’ Perceptions of AI-Driven Mental Health Support and Its Relationship with Anxiety, Depression, and Seeking Professional Psychological Help: Transitioning from Traditional Counseling to Digital Support. Healthcare 13, 1089 (2025).
Article PubMed PubMed Central Google Scholar
Cross, S. et al. Use of AI in mental health care: community and mental health professionals survey. JMIR Ment. Health 11, e60589 (2024).
Article PubMed PubMed Central Google Scholar
Rackoff, G. N., Zhang, Z. Z. & Newman, M. G. Chatbot-delivered mental health support: Attitudes and utilization in a sample of U.S. college students. Digit. Health 11, 20552076241313401 (2025).
Article PubMed PubMed Central Google Scholar
Rousmaniere, T., Zhang, Y., Li, X. & Shah, S. Large language models as mental health resources: Patterns of use in the United States. Practice Innovations. Advance online publication. https://doi.org/10.1037/pri0000292 (2025).
Me, myself, and AI: understanding and safeguarding children’s use of AI Chatbots. Internet Matters.org. July 2025.
Jeelani, G. 88% of Students turn to AI during stress, survey finds; ChatGPT the most popular tool. Mint. August 20, 2025. https://www.livemint.com/technology/tech-news/88-students-turn-to-ai-during-stress-survey-finds-chatgpt-the-most-popular-tool-11755658354656.html.
2025 Norton Cyber Safety Insights Report. Norton. July 2025.
In AI, We Trust: Part 1. Snapcart. April 23, 2025. https://snapcart.global/in-ai-we-trust-part-1/.
Mental Health UK. Over one in three using AI chatbots for mental health support, as charity calls for urgent safeguards. November 18, 2025. https://mentalhealth-uk.org/news-and-insights/over-one-in-three-using-ai-chatbots-for-mental-health-support-as-charity-calls-for-urgent-safeguards/ Mental Health Research Canada. Understanding the mental health of Canadians. March 31, 2026. https://www.mhrc-rsmc.ca/sondage-national-25.
Bansal, B. UK trials AI therapy: Young adults most open to talking to AI-powered chatbots about mental health. YouGov.com. May 20, 2024. Mental Health Awareness Month: Key Findings on U.S. Attitudes and Barrier to Care. The Harris Poll. May 7, 2025. Oliver Wyman Forum. How Generative AI Is Transforming Business And Society. 2024.
Phang, et al. Investigating Affective Use and Emotional Well-being on ChatGPT. OpenAI. March 2025.
Sidoti, O., & McClain, C. 34% of U.S. adults have used ChatGPT, about double the share in 2023. Pew Research Center. June 25, 2025.
Linardon, J. et al. Association between user engagement and clinical outcomes in smartphone apps for depression and anxiety: A systematic review and meta-analysis. Psychiatry Res. 25, 116864 (2025).
Google Scholar
Torous, J., Firth, J. & Goldberg, S. B. Digital mental health’s unstable dichotomy: Wellness and health. JAMA Psychiatry 81, 539–540 (2024).
Article PubMed PubMed Central Google Scholar
Download references
This work is supported by a charitable grant from the Argosy Foundation.
Beth Israel Deaconess Medical Center & Harvard University, Boston, MA, USA
Rebekah Bodner, Katherine Lim, Steven Siddals & John Torous
University of Wisconsin, Madison, WI, USA
Simon Goldberg
PubMed Google Scholar
PubMed Google Scholar
PubMed Google Scholar
PubMed Google Scholar
PubMed Google Scholar
J.T. and R.B. conceived the study. KL and RB ran the initial search. K.L and R.B. and J.T. wrote the first draft. SS created the figures. S.G. revised the paper and rewrote sections. All authors then revised and edited the paper and approved the final version.
Correspondence to John Torous.
The authors declare no competing interests.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
Reprints and permissions
Bodner, R., Lim, K., Siddals, S. et al. Barriers to understanding how many people use AI for mental health support: an estimate and narrative review. npj Digit. Public Health 1, 21 (2026). https://doi.org/10.1038/s44482-026-00025-7
Download citation
Received:
Accepted:
Published:
Version of record:
DOI: https://doi.org/10.1038/s44482-026-00025-7
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.
Provided by the Springer Nature SharedIt content-sharing initiative
Advertisement
npj Digital Public Health (npj Digit. Public Health)
ISSN 3091-2784 (online)
© 2026 Springer Nature Limited
Sign up for the Nature Briefing: AI and Robotics newsletter — what matters in AI and robotics research, free to your inbox weekly.