AI Chatbot Use and Disclosure for Mental Health Among US Adolescents and Young Adults

Ryan K McBain; Jonathan H Cantor; Joshua Breslau; Melissa Diliberti; Li Ang Zhang; Fang Zhang; Alyssa Burnett; Aaron Kofner; Benjamin Rader; Pat Pataranutaporn; Bradley D Stein; Ateev Mehrotra; Hao Yu

doi:10.1001/jamapediatrics.2026.2015

. 2026 Jun 1;180(8):884–890. doi: 10.1001/jamapediatrics.2026.2015

AI Chatbot Use and Disclosure for Mental Health Among US Adolescents and Young Adults

Ryan K McBain ^1,^2,^3,^✉, Jonathan H Cantor ⁴, Joshua Breslau ⁵, Melissa Diliberti ⁵, Li Ang Zhang ⁴, Fang Zhang ^3,⁶, Alyssa Burnett ⁶, Aaron Kofner ¹, Benjamin Rader ^3,⁷, Pat Pataranutaporn ⁸, Bradley D Stein ⁵, Ateev Mehrotra ⁹, Hao Yu ^3,⁶

¹RAND, Arlington, Virginia

²Mass General Brigham, Boston, Massachusetts

³Harvard Medical School, Boston, Massachusetts

⁴RAND, Santa Monica, California

⁵RAND, Pittsburgh, Pennsylvania

⁶Harvard Pilgrim Health Care Institute, Boston, Massachusetts

⁷Boston Children’s Hospital, Boston, Massachusetts

⁸MIT Media Lab, MIT, Cambridge, Massachusetts

⁹Brown University School of Public Health, Providence, Rhode Island

Accepted for Publication: April 13, 2026.

Published Online: June 1, 2026. doi:10.1001/jamapediatrics.2026.2015

^✉

Corresponding Author: Ryan K. McBain, PhD, MPH, RAND Corporation, 1200 S Hayes St, Arlington, VA 22202 (rmcbain@rand.org).

Author Contributions: Dr Cantor had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: McBain, Cantor, Breslau, L. Zhang, Burnett, Stein, Mehrotra, Yu.

Acquisition, analysis, or interpretation of data: McBain, Cantor, Breslau, Diliberti, L. Zhang, F. Zhang, Kofner, Rader, Pataranutaporn, Yu.

Drafting of the manuscript: McBain, Cantor, Breslau, Pataranutaporn.

Critical review of the manuscript for important intellectual content: All authors.

Statistical analysis: Cantor, Breslau, F. Zhang, Kofner.

Obtained funding: McBain, Mehrotra, Yu.

Administrative, technical, or material support: McBain, Diliberti, L. Zhang, Burnett, Stein, Yu.

Supervision: McBain, Breslau, L. Zhang, Pataranutaporn, Stein, Yu.

Conflict of Interest Disclosures: Dr McBain reported receiving grants from the National Institute of Mental Health (NIMH) during the conduct of the study. Dr Cantor reported receiving grants from the NIMH and the National Institute on Aging and personal fees from Chestnut Health, the Government Accountability Office, The Aspen Institute, and NYS Youth Justice Institute outside the submitted work. Dr F. Zhang reported receiving grants from Pfizer Inc and GlaxoSmithKline (research contract to employer, Harvard Pilgrim Health Care Institute) outside the submitted work. Dr Burnett reported receiving grants from the NIMH during the conduct of the study. Dr Stein reported receiving grants from the NIMH during the conduct of the study. Dr Mehrotra reported receiving personal fees from Black Opal and Peterson Health Technology Institute outside the submitted work. Dr Yu reported receiving grants from the NIMH, the National Institute on Minority Health and Health Disparities, the National Institute on Alcohol Abuse and Alcoholism, and the National Institute of Nursing Research outside the submitted work. No other disclosures were reported.

Funding/Support: This work was supported by grants 1R01MH132551 and 1R01MH135034 from the National Institute of Mental Health.

Role of the Funder/Sponsor: The funder had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Data Sharing Statement: See Supplement 2.

Additional Contributions: We thank all survey respondents, as well as those at RAND involved in data collection processes. Beyond usual salary where applicable, no one received financial compensation for their contribution.

^✉

Corresponding author.

PMCID: PMC13227335 NIHMSID: NIHMS2166111 PMID: 42223976

This survey study evaluates the percentage of US adolescents and young adults who used artificial intelligence (AI) chatbots for mental health advice as of 2025 and to what extent they told others.

Key Points

Questions

As of 2025, what percentage of US adolescents and young adults aged 12 to 21 years used artificial intelligence (AI) chatbots for mental health advice, and among those who use AI chatbots for this purpose, to what extent did they tell others?

Findings

This national survey including more than 42 million US youth (population-weighted) found that almost a fifth of adolescents and young adults reported using AI chatbots for mental health advice, representing an increase by almost half from 1 year prior. Most users told no one that they used AI chatbots for this purpose.

Meaning

AI chatbots are widely used by adolescents and young adults for emotional and psychological support, underscoring the urgent need for parents, clinicians, and policymakers to understand their evolving role in youth mental health care.

Abstract

Importance

The rapid expansion of artificial intelligence (AI) chatbots has coincided with a persistent youth mental health crisis in the US, raising a question about the extent to which young people are turning to this technology for mental health advice.

Objective

To assess the prevalence, frequency, perceived helpfulness, and disclosure of AI chatbot use for mental health advice among US adolescents and young adults in 2025.

Design, Setting, and Participants

This cross-sectional, nationally representative survey was conducted with adolescents and young adults aged 12 to 21 years in November 2025.

Exposures

Exposures included self-reported age, sex, race and ethnicity, census region, metropolitan status, and prior discussion with a clinician about mental health in the past 6 months.

Main Outcomes and Measures

Self-reported use of AI chatbots for mental health advice, including any prior use, frequency of use, perceived helpfulness of responses, and disclosure of use to others. Respondents were also asked whether they had spoken with a physician about their mental health in the prior 6 months. Using multivariable logistic regression analysis, variation in responses was assessed according to respondents’ demographic and geographic characteristics.

Results

Among a US population-weighted 42 825 655 youth (unweighted, 1009 youth; median [IQR] age, 17 [15-18] years; population-weighted 21 410 663 male [50.0%]), 19.2% of adolescents and young adults (population-weighted n = 8 207 180) in 2025 reported having used AI chatbots for mental health advice. Among those who sought advice from AI chatbots, 42.8% did so at least monthly, and 91.7% rated the advice as somewhat or very helpful. Most adolescents reported they had not disclosed AI chatbot use for mental health advice to anyone (63.3%). Use of an AI chatbot for mental health advice was more common among females compared with males (adjusted odds ratio [aOR], 2.10; 95% CI, 1.36-3.23), respondents aged 18 to 21 years compared with respondents aged 12 to 14 years (aOR, 3.65; 95% CI, 1.98-6.74), and those who had spoken with a physician about their mental health in the prior 6 months compared with those who had not (aOR, 1.89; 95% CI, 1.18-3.03).

Conclusions and Relevance

In this nationally representative survey study of US adolescents and young adults, a fifth reported using AI chatbots for mental health advice. AI chatbots are already embedded in many youths’ mental health information ecosystem, underscoring the need for parents and clinicians to proactively discuss chatbot use to promote safety, appropriate expectations, and linkages to evidence-based care.

Introduction

Since the commercial launch of artificial intelligence (AI) chatbots such as ChatGPT (OpenAI), awareness and adoption have been swiftest among adolescents and young adults.¹ A 2025 Pew Research survey found that 64% of teens aged 13 to 17 years reported having ever used an AI chatbot, and 28% use them daily.² During the same period, the US has been confronting a mental health crisis among adolescents and young adults.³ Suicide is the second leading cause of death among those aged 5 to 24 years.^4,5 A Centers for Disease Control and Prevention (CDC) survey in 2023 found that 1 in 5 high schoolers (20.4%) considered attempting suicide, and almost 1 in 10 (9.5%) reported ever attempting suicide.⁶ Rates of depression (18%) and anxiety disorders (20%) among adolescents are alarmingly high.⁷ Nevertheless, 4 in 10 teenagers with a major depressive episode in the past year report not receiving mental health services,⁸ and anxiety disorders have an even larger treatment gap.⁹

The rapid adoption of AI chatbots raises the question of how often adolescents and young adults turn to them for mental health advice.^10,11,12 Studies conducted before the introduction of large language model (LLM)–based AI chatbots have found that adolescents frequently search for mental health information online, in part because the internet is easily accessible, anonymous, and low-cost.¹³ AI chatbots could be even more attractive as sources of advice because they offer personalized and interactive guidance.¹⁴ These same features raise concerns, as exchanges occur without oversight by adults or mental health professionals.^14,15,16

In a 2024 national survey, we found that 1 in 8 adolescents and young adults reported using generative AI for mental health advice.¹⁷ In the current study, we examined prevalence, frequency of use, and perceived helpfulness in 2025, while expanding the question set to include disclosure of use to family and friends and whether AI chatbot use coincided with seeking mental health services from a physician. We also examined the extent to which these measures varied according to respondents’ demographic and geographic characteristics.

Methods

Study Design

This cross-sectional survey study received approval from Harvard Pilgrim Health Care Institute’s institutional review board. The study adheres to Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines.

Procedures

Participants were drawn from members of RAND’s American Life Panel (ALP), a nationally representative survey panel.¹⁸ ALP panelists regularly complete internet-based surveys on an array of topics, including health. Surveys are administered online, in English, and participants receive financial compensation. Adults 18 and older are initially enrolled based on random samples of household unit listings from the US Postal Service’s Delivery Sequence File. Adult panel members provided written informed consent to the participation of eligible minors within their household for surveys administered to those younger than 18 years; minors then provide assent. For this study, 1727 youth aged 12 to 21 years were invited to participate. We applied survey weights to account for US population demographics in terms of age, race and ethnicity (Black, Hispanic, non-Hispanic White, and other, which represents all categories apart from those classified), sex, and geographic region. Race and ethnicity information was used to inspect demographic differences on each outcome of interest. The survey was administered in November 2025.

Measures

We used 6 survey questions on AI chatbot use for mental health advice. Four of the questions corresponded to those in a 2024 survey by the research team.¹⁷ Respondents were first informed that chatbots that answer questions—such as ChatGPT, Google Gemini (Google), My AI (Snap Inc), Character.AI (Character Technologies), and Meta AI (Meta)—are examples of AI chatbots. Respondents were then asked whether they have ever used an AI chatbot. Among those who responded yes, respondents reported whether they had ever asked an AI chatbot for advice about their mental health—specifically, when feeling sad, angry, nervous, or stressed (yes or no). Survey methodology is available in eMethods 1 in Supplement 1.

Respondents who reported ever using an AI chatbot for mental health advice then indicated their frequency of such use (hardly ever, at least once a month, at least once a week, daily or almost daily), perceived helpfulness (very helpful, somewhat helpful, or not helpful), and whether they had disclosed this use to anyone (eg, parent or guardian, brother or sister, friend). Lastly, we asked all survey respondents whether they had spoken about their mental health in the past 6 months to a clinician (yes or no). The full survey instrument is viewable in eMethods 2 in Supplement 1.

Individual demographic variables collected in both survey panels included sex, age, and race and ethnicity. We linked geographic identifiers for each respondent to their US census region and defined metropolitan or nonmetropolitan status based on ZIP code–level information and rural-urban commuting area codes specified by the US Department of Agriculture Economic Research Service.¹⁹

Statistical Analysis

We first summarized responses to each question using survey-weighted percentages, then conducted cross-tabulations with demographic and geographic characteristics. We subsequently conducted multivariable logistic regression analyses, examining the association between survey responses and joint demographic and geographic characteristics previously described, as well as whether the individual had spoken to a physician in the prior 6 months about their mental health. The outcomes specified in regression analyses were as follows: (1) any AI chatbot use for mental health advice (yes or no), (2) monthly or more often frequency of AI chatbot use for mental health advice (yes or no), (3) AI chatbot used for mental health advice perceived as somewhat or very helpful (yes or no), and (4) disclosure of AI chatbot use for mental health advice (yes or no). These analyses incorporated survey weights, reported adjusted odds (each odds ratio for each explanatory variable was adjusted for all other explanatory variables in the model), and were performed using Stata, version 19.5 (StataCorp).

Results

Descriptive Characteristics

Of 1727 individuals contacted, 1009 responded to the survey. Unweighted and weighted sample characteristics are provided in Table 1. The US population-weighted sample included 42 825 655 youth (median [IQR] age, 17 [15-18] years; population-weighted 20 916 260 female [48.8%]; 21 410 663 male [50.0%]). In terms of race and ethnicity, the population-weighted sample included 5 563 350 Black (13.0%), 11 010 863 Hispanic (25.7%), 21 325 219 White (49.8%), and 4 316 881 other (10.1%) respondents.

Table 1. Descriptive Characteristics of Survey Sample.

Sample characteristic	Sample size (%)
Sample characteristic	Unweighted	Weighted
Sex
Male	427 (42.3)	21 410 663 (50.0)
Female	570 (56.5)	20 916 260 (48.8)
Missing^a	12 (1.2)	498 732 (1.2)
Age, y
12-14	232 (23.0)	8 560 172 (20.0)
15-17	455 (45.1)	16 950 151 (39.6)
18-21	322 (31.9)	17 315 332 (40.4)
Race and ethnicity
Black	94 (9.3)	5 563 350 (13.0)
Hispanic	192 (19.0)	11 010 863 (25.7)
White non-Hispanic	647 (64.1)	21 325 219 (49.8)
Other^b	68 (6.7)	4 316 881 (10.1)
Missing^a	8 (0.8)	609 342 (1.4)
Census region
Northeast	180 (17.8)	7 002 578 (16.4)
Midwest	230 (22.8)	9 006 184 (21.0)
South	385 (38.2)	16 648 986 (38.9)
West	214 (21.2)	10 167 908 (23.7)
Metropolitan status
Not metropolitan	133 (13.2)	5 038 619 (11.8)
Metropolitan	876 (86.8)	37 787 036 (88.2)
Total	1009 (100.0)	42 825 655 (100.0)

Open in a new tab

^{^a}

Missing indicates that the survey respondent did not provide a response to the relevant question.

^{^b}

Other race and ethnicity represents all categories apart from those classified.

Use of AI Chatbots

Applying survey weights to all responses, we found that almost 1 in 5 respondents (19.2%; population-weighted n = 8 207 180) stated they had ever used an AI chatbot for mental health advice. Usage was highest among those aged 18 to 21 years (24.1%), females (25.0%), and those who had spoken with their physician in the past 6 months about their mental health (27.7%) (Figure 1). Of those who had used AI chatbots for mental health advice, 42.8% did so at least monthly, 26.3% reported use at least once a month, 10.8% at least once a week, and 5.8% daily or almost daily. In terms of helpfulness, 91.7% rated the advice as somewhat or very helpful (25.0% found the advice to be very helpful and 66.7% found it somewhat helpful), and 8.3% found it not helpful. Additionally, 21.1% of respondents reported discussing their mental health with a physician in the prior 6 months.

Four-panel bar charts of individuals percent by year, sex, age, and physician talk. Multi-panel figure with four vertical bar charts arranged two by two, panels labeled A, B, C, and D in small boxed letters at each panel’s upper left. All panels share a vertical axis labeled Individuals, percent, ranging from 0 to 100 with horizontal gridlines at 25, 50, 75, and 100. Bars are muted blue with black outlines and black capped error bars. Panel A title at top left: Year, 2024 vs 2025. Horizontal axis labeled Year. Two categories: 2024 with text below reading n equals 112 and a bar near 14 percent; 2025 with text below reading n equals 186 and a bar near 19 percent. Panel B title: Sex. Horizontal axis labeled Sex. Two categories: Female with n equals 127 and a bar near 25 percent; Male with n equals 58 and a bar near 14 percent. Panel C title: Age group. Horizontal axis labeled Age, y. Three categories: 12 to 14 with n equals 23 and a bar near 8 percent; 15 to 17 with n equals 84 and a bar near 21 percent; 18 to 21 with n equals 79 and a bar near 24 percent. Panel D title: Talk to physician about mental health. Horizontal axis labeled Talked to physician about mental health. Two categories: Yes with n equals 54 and a bar near 27 percent; No with n equals 132 and a bar near 18 percent. — Each bar represents the survey-weighted percentage of individuals within the subgroup who report that they have used an AI chatbot for mental health advice. Numbers represent the unweighted number of survey respondents, within the category, who report that they used an AI chatbot for mental health advice.

Most respondents who used an AI chatbot for mental health advice had not disclosed this to anyone (63.3%). Those who disclosed use were most likely to disclose this to a friend (28.0%), or a trusted adult such as a parent, teacher, or physician (16.4%).

Respondent Characteristics Associated With AI Chatbot Mental Health Use

Use of an AI chatbot for mental health advice was more common among females compared with males (adjusted odds ratio [aOR], 2.10; 95% CI, 1.36-3.23), among older respondents compared with those aged 12 to 14 years (aged 15-17 years: aOR, 2.84; 95% CI, 1.53-5.28; aged 18-21 years: aOR, 3.65; 95% CI, 1.98-6.74), among respondents who listed their race and ethnicity as other (aOR, 2.50; 95% CI, 1.22-5.13) compared with those who listed their race as White, and among respondents who had spoken with a physician about their mental health in the past 6 months (aOR, 1.89; 95% CI, 1.18-3.03) compared with those who had not. We did not find significant differences with respect to metropolitan status, or census region (Table 2).

Table 2. Correlates of Artificial Intelligence (AI) Chatbot Use for Mental Health Advice Among Adolescents and Young Adults in the US.

Independent measure	Outcome 1; outcomes 2-4, No.	Ever asked chatbot for mental health advice		Asked chatbot for mental health advice monthly or more often		AI chatbot mental health advice perceived as helpful		Disclosure of using AI chatbot for mental health advice
Independent measure	Outcome 1; outcomes 2-4, No.	aOR^a (95% CI)	P value	aOR^a (95% CI)	P value	aOR^a (95% CI)	P value	aOR^a (95% CI)	P value
Sex
Female	n = 569; n = 127	2.10 (1.36-3.23)	.001	1.16 (0.53-2.58)	.71	1.24 (0.33-4.75)	.75	0.60 (0.26-1.37)	.23
Male	n = 423; n = 57	1 [Reference]	Ref	1 [Reference]	Ref	1 [Reference]	Ref	1 [Reference]	Ref
Age group, y
12-14	n = 229; n = 23	1 [Reference]	Ref	1 [Reference]	Ref	1 [Reference]	Ref	1 [Reference]	Ref
15-17	n = 448; n = 83	2.84 (1.53-5.28)	.001	2.45 (0.65-9.27)	.19	0.42 (0.07-2.36)	.32	0.76 (0.22-2.63)	.66
18-21	n = 315; n = 78	3.65 (1.98-6.74)	<.001	2.32 (0.59-9.08)	.23	0.56 (0.09-3.30)	.52	0.81 (0.23-2.84)	.74
Race and ethnicity
Black	n = 94; n = 16	0.79 (0.40-1.55)	.49	5.45 (1.44-20.66)	.01	NA^b	NA^b	0.57 (0.18-1.81)	.34
Hispanic	n = 189; n = 38	1.27 (0.75-2.15)	.38	2.20 (0.93-5.20)	.07	4.70 (0.42-52.62)	.21	0.46 (0.18-1.18)	.10
White non-Hispanic	n = 641; n = 113	1 [Reference]	Ref	1 [Reference]	Ref	1 [Reference]	Ref	1 [Reference]	Ref
Other^c	n = 68; n = 17	2.50 (1.22-5.13)	.01	1.83 (0.53-6.29)	.34	0.89 (0.12-6.80)	.91	0.33 (0.09-1.16)	.08
Metropolitan status
Not metropolitan	n = 132; n = 22	1 [Reference]	Ref	1 [Reference]	Ref	1 [Reference]	Ref	1 [Reference]	Ref
Metropolitan	n = 860; n = 162	1.09 (0.57-2.08)	.80	2.42 (0.71-8.28)	.16	0.18 (0.02-1.59)	.12	0.52 (0.13-2.02)	.34
Census region
South	n = 380; n = 71	1 [Reference]	Ref	1 [Reference]	Ref	1 [Reference]	Ref	1 [Reference]	Ref
Northeast	n = 177; n = 40	1.02 (0.59-1.78)	.93	0.94 (0.34-2.57)	.90	0.43 (0.10-1.93)	.27	1.33 (0.45-4.00)	.61
Midwest	n = 228; n = 37	0.70 (0.39-1.26)	.24	0.38 (0.13-1.13)	.08	0.60 (0.09-4.24)	.61	0.50 (0.17-1.48)	.21
West	n = 207; n = 36	0.87 (0.49-1.55)	.63	0.73 (0.28-1.94)	.53	0.75 (0.14-4.09)	.74	1.33 (0.49-3.63)	.58
Talked to physician^d
No	n = 768; n = 130	1 [Reference]	Ref	1 [Reference]	Ref	1 [Reference]	Ref	1 [Reference]	Ref
Yes	n = 224; n = 54	1.89 (1.18-3.03)	.008	0.71 (0.29-1.72)	.45	0.40 (0.13-1.29)	.12	1.01 (0.43-2.35)	.99

Open in a new tab

Abbreviations: aOR, adjusted odds ratio, NA, not available; ref, reference.

^{^a}

aOR represents adjusted odds ratios from multivariable logistic regression including all independent measures.

^{^b}

Missing values for the third outcome are because the category perfectly predicted the outcome (complete separation), yielding nonfinite coefficient estimates.

^{^c}

Other race and ethnicity represents all categories apart from those classified.

^{^d}

Represents having talked to a physician in the prior 6 months about one’s mental health.

Correlates of AI Chatbot Use Behaviors

Those reporting AI chatbot use for mental health advice at least monthly were more likely to be Black (aOR, 5.45; 95% CI, 1.44-20.66) than White (Figure 2). We did not find statistically significant differences in frequency of use with respect to sex, age, metropolitan status, census region, or having spoken with a physician about mental health in the past 6 months. With respect to perceived helpfulness and disclosure, we did not find any statistically significant differences among explanatory variables of interest.

Forest plot of adjusted odds ratios by race and ethnicity with 95 percent confidence intervals. Two-part data figure with a text table on the left and a forest plot on the right. Left section header reads Race and ethnicity. Four rows list group names with sample sizes and adjusted odds ratio with 95 percent confidence interval. Row 1: Black, n equals 16; adjusted odds ratio 5 point 45 with 95 percent confidence interval 1 point 44 to 20 point 66. Row 2: Hispanic, n equals 38; adjusted odds ratio 2 point 20 with 95 percent confidence interval 0 point 93 to 5 point 20. Row 3: Non-Hispanic White, n equals 113; value 1, labeled Reference. Row 4: Other, n equals 17; adjusted odds ratio 1 point 83 with 95 percent confidence interval 0 point 53 to 6 point 29. Right section contains a horizontal axis labeled Adjusted odds ratio, 95 percent C I, with tick labels 0 point 25, 0 point 50, 1, 2, 5, 10, and 20, indicating a logarithmic scale. A vertical dotted reference line at 1 divides two headings at the top: left side labeled Favors not asking and right side labeled Favors asking. Each group has a dark square marker with a horizontal line for the confidence interval aligned to its row. The Black row square lies between 2 and 5, near 5 point 45, with a long confidence interval line extending rightward beyond 20, ending in a right-pointing arrow; the left end aligns near 1 point 44. The Hispanic row square lies just above 2, near 2 point 20, with the confidence interval spanning from just under 1 to slightly above 5. The Non-Hispanic White reference row has a square centered on the vertical line at 1. The Other row square lies between 1 and 2, near 1 point 83, with the confidence interval line extending from just above 0 point 50 to above 5, near 6 point 29. — Data from weighted US adolescent and young adult survey; multivariable model adjusts for sex, age, metro status, region, and physician discussion on mental health needs in prior six months. Numbers represent the number of respondents in each race and ethnicity category. Other race and ethnicity represents all categories apart from those classified.

Discussion

In this nationally representative survey, we estimated that roughly 1 in 5 adolescents and young adults in the US—representing approximately 8 million individuals—have ever used an AI chatbot for mental health advice as of 2025. This figure compares with 1 in 8 adolescents and young adults (13.1%) from a similar survey¹⁷ using slightly different wording conducted by our research team in 2024. Adolescents and young adults who have spoken to a physician about their mental health in the past 6 months were more likely to have used an AI chatbot for mental health advice. Among those using AI chatbots for mental health advice, the majority (approximately 5 million) did not disclose their use of AI chatbots for mental health advice to anyone.

Our results suggest that use of AI chatbots for mental health advice among adolescents and young adults is not a marginal phenomenon: 19.2% of respondents reported using AI chatbots for this purpose, a percentage similar in magnitude to the 19.8% receiving counseling from a mental health professional.⁷ These measures are not equivalent; counseling reflects formal clinical care, whereas our survey captures a broader range of engagements for mental health advice. Even so, the prevalence of chatbot use raises important questions about the role that AI may be playing in young people’s emotional lives.

On the one hand, the level of engagement among a subset of users—1 in 12 respondents reported using AI chatbots for mental health advice monthly or more often—suggests that AI chatbots may be filling a gap in emotional or psychological support. A majority of users also found AI chatbot responses somewhat or very helpful. On the other hand, perceived helpfulness may reflect AI chatbots’ tendencies toward sycophancy and overflattery, rather than the quality of advice they provide.^20,21 Clinical experts have expressed concern that those with more intensive mental health needs may rely on therapeutic guidance from AI chatbots when such support cannot substitute for trained mental health professionals.¹⁴ AI chatbots specifically designed to offer therapeutic guidance—in contrast with general purpose AI chatbots—are still nascent, as are benchmarks for determining their overall performance,^22,23,24 and there is limited transparency about the datasets used to train these models.²⁵

Although mental health advice from AI chatbots may circumvent stigma associated with disclosure to adults or peers,^11,16,26 this privacy also poses risks if individuals do not disclose their use of these tools to health care professionals or trusted adults. We found that almost two-thirds of adolescents and young adults (63.3%) who use AI chatbots for mental advice do not tell anyone. This finding is consistent with previous qualitative research showing that adolescent use of AI is happening without parental knowledge or involvement.²⁷ To our knowledge, there are no directly comparable studies quantifying the extent to which young people disclose to parents their use of traditional, human-delivered mental health care. However, a recent CDC survey found substantial parent-teen concordance regarding their adolescents’ receipt of medical care more broadly,²⁸ suggesting that parental awareness may be higher in conventional care settings than in the context observed here. Without disclosure by young people on the role of AI in providing mental health advice, clinicians may also be unaware of potentially influential or inaccurate advice provided by these tools to their patients, missing opportunities to offer guidance, context, or monitoring. Parents, clinicians, and educators should consider asking adolescents and young adults about their engagement with AI chatbots and provide counsel on their strengths and limitations.

Use of AI chatbots for mental health advice was more common in older adolescents and young adults, paralleling greater prevalence of mental health needs and conditions among this group.⁹ Use may also be more common in older respondents due to greater access to smartphones and reduced parental supervision of online activity. We also found that females were more likely to use AI chatbots for mental health advice, possibly because of a higher prevalence of mental health conditions.²⁹ Lastly, we observed that—conditional on using AI chatbots for mental health advice—Black youth were over 5 times more likely than White youth to seek mental health advice monthly or more often. This may reflect a sense, within this population, that professionals are not as responsive to their unique needs, or else reflect reduced access to professional services.³⁰ However, the sample size for this particular analysis was small, and follow-on investigation is warranted.

Limitations

We note several key study limitations. First, we did not inquire about specific AI chatbots, and we did not examine heterogeneity of respondents’ experiences and outcomes. Both limit the specificity of our findings. Second, AI chatbots built on generative AI are constantly evolving; we present a snapshot as of late 2025. Third, the overall sample size was 1009, and the completion rate was 58.4%. Although this implies our findings contain uncertainty and may be subject to nonresponse bias, this is a common concern in surveys measuring use of AI.³¹ Additional bias may derive from the use of a web-based survey among English-speaking respondents, as these respondents may be more (or less) likely to use LLMs compared with other populations. Fourth, our investigation intentionally used language that would be accessible to individuals aged 12 to 21 years, such as “advice or help when you feel sad, angry, or nervous.” We did not inquire about mental health diagnoses, instead opting for more concrete language about individuals’ emotional experiences. Lastly, the survey does not aim to assess the quality of care delivered by AI chatbots, and this represents a critical area for future research. Despite these limitations, this survey study provides a novel snapshot into adolescents’ and young adults’ engagement with AI chatbots, offering insights on how these technologies are being used as sources of mental health advice.

Conclusions

The finding of this nationally representative survey that almost 1 in 5 adolescents in the US reported use of AI chatbots for mental health advice—with over 40% of users doing so monthly or more often—underscores the urgency of understanding and shaping the evolving role of AI chatbots in youth mental health care. As these technologies become increasingly integrated into the daily lives of young people, they should be understood as active contributors in the broader ecosystem of psychological interventions.

Supplement 1.

eMethods 1. Survey Methodology

eMethods 2. List of Questions

jamapediatr-e262015-s001.pdf^{(211.2KB, pdf)}

Supplement 2.

Data Sharing Statement.

jamapediatr-e262015-s002.pdf^{(15.7KB, pdf)}

References

1.Kennedy B, Yam E, Kikuchi E, Pula I, Fuentes J. AI in Americans’ lives: awareness, experiences and attitudes. Accessed January 21, 2026. https://www.pewresearch.org/science/2025/09/17/ai-in-americans-lives-awareness-experiences-and-attitudes/
2.Faverio M, Sidoti O. Teens, social media, and AI chatbots 2025. Accessed December 9, 2025. https://www.pewresearch.org/internet/2025/12/09/teens-social-media-and-ai-chatbots-2025/
3.Office of the Surgeon General . Protecting Youth Mental Health: The US Surgeon General’s Advisory. US Department of Health and Human Services; 2021. [Google Scholar]
4.AHRQ. 2023 National Healthcare Quality and Disparities Report. Agency for Healthcare Research and Quality (US); 2023. [Google Scholar]
5.Centers for Disease Control and Prevention. WISQARS Explore Fatal and Nonfatal Data. Accessed December 20, 2025. https://wisqars.cdc.gov/explore/?o=MORT&y1=2021&y2=2021&g=00&t=0&i=0&m=20810&d=&s=0&r=0&me=0&ry=2&yp=65&e=0&a=custom&a1=5&a2=24&g1=0&g2=199
6.Verlenden JV, Fodeman A, Wilkins N, et al. Mental health and suicide risk among high school students and protective factors—Youth Risk Behavior Survey, US, 2023. MMWR. 2024;73(4):79-86. doi: 10.15585/mmwr.su7304a9 [DOI] [Google Scholar]
7.Centers for Disease Control and Prevention . Bullying victimization among teeanagers: US, July 2021-December 2023. Accessed January 3, 2026. https://www.cdc.gov/nchs/products/databriefs/db514.htm
8.Substance Abuse and Mental Health Services Administration . Key substance use and mental health indicators in the US: results from the 2023 National Survey on Drug Use and Health. Accessed January 3, 2026. https://www.samhsa.gov/data/sites/default/files/reports/rpt47095/National%20Report/National%20Report/2023-nsduh-annual-national.pdf
9.Centers for Disease Control and Prevention . Data and statistics on children’s mental health. Accessed March 10, 2025. https://www.cdc.gov/children-mental-health/data-research/index.html
10.Herbener AB, Damholdt MF. Are lonely youngsters turning to chatbots for companionship—the relationship between chatbot usage and social connectedness in Danish high-school students. Int J Hum Comput Stud. 2025;196:103409. doi: 10.1016/j.ijhcs.2024.103409 [DOI] [Google Scholar]
11.Nagata JM, Memon Z, Huang O, Moreno MA. Adolescent health and generative AI—risks and benefits. JAMA Pediatr. 2026;180(1):7-8. doi: 10.1001/jamapediatrics.2025.4502 [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Giovanelli A, Roundfield KD. Adolescent vulnerability to consumer chatbots—artificial agents and genuine risk. JAMA Netw Open. 2025;8(10):e2539028. doi: 10.1001/jamanetworkopen.2025.39028 [DOI] [PubMed] [Google Scholar]
13.Pretorius C, Chambers D, Cowan B, Coyle D. Young people seeking help online for mental health: cross-sectional survey study. JMIR Ment Health. 2019;6(8):e13524. doi: 10.2196/13524 [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Stade EC, Stirman SW, Ungar LH, et al. Large language models could change the future of behavioral health care: a proposal for responsible development and evaluation. Npj Mental Health Res. 2024;3(1):12. doi: 10.1038/s44184-024-00056-z [DOI] [Google Scholar]
15.van Heerden AC, Pozuelo JR, Kohrt BA. Global mental health services and the impact of artificial intelligence-powered large language models. JAMA Psychiatry. 2023;80(7):662-664. doi: 10.1001/jamapsychiatry.2023.1253 [DOI] [PubMed] [Google Scholar]
16.Lawrence HR, Schneider RA, Rubin SB, Matarić MJ, McDuff DJ, Jones Bell M. The opportunities and risks of large language models in mental health. JMIR Ment Health. 2024;11:e59479. doi: 10.2196/59479 [DOI] [PMC free article] [PubMed] [Google Scholar]
17.McBain RK, Bozick R, Diliberti M, et al. Use of generative AI for mental health advice among US adolescents and young adults. JAMA Network Open. 2025;8(11):e2542281. doi: 10.1001/jamanetworkopen.2025.42281 [DOI] [PMC free article] [PubMed] [Google Scholar]
18.RAND American Youth Panel . A nationally representative panel of young people. Accessed March 7, 2025. https://www.rand.org/education-and-labor/survey-panels/ayp.html
19.US Department of Agriculture, Economic Research Service . Rural-urban commuting area codes. Accessed January 3, 2026. https://www.ers.usda.gov/data-products/rural-urban-commuting-area-codes
20.Andoh E. AI chatbots and digital companions are reshaping emotional connection. American Psychological Association. Accessed January 8, 2026. https://www.apa.org/monitor/2026/01-02/trends-digital-ai-relationships-emotional-connection
21.Shekar S, Pataranutaporn P, Sarabu C, Cecchi GA, Maes P. People overtrust AI-generated medical advice despite low accuracy. NEJM AI. 2025;2(6):AIoa2300015. doi: 10.1056/AIoa2300015 [DOI] [Google Scholar]
22.Meng X, Yan X, Zhang K, et al. The application of large language models in medicine: a scoping review. iScience. 2024;27(5):109713. doi: 10.1016/j.isci.2024.109713 [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Malgaroli M, Schultebraucks K, Myrick KJ, et al. Large language models for the mental health community: framework for translating code to care. Lancet Digit Health. 2025;7(4):e282-e285. doi: 10.1016/S2589-7500(24)00255-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Dwyer B, Flathers M, Sano A, et al. Mindbench.ai: an actionable platform to evaluate the profile and performance of large language models in a mental healthcare context. NPP Digit Psychiatry Neurosci. 2025;3(1):28. doi: 10.1038/s44277-025-00049-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Abbasian M, Khatibi E, Azimi I, et al. Foundation metrics for evaluating effectiveness of health care conversations powered by generative AI. NPJ Digit Med. 2024;7(1):82. doi: 10.1038/s41746-024-01074-z [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Ma Z, Mei Y, Su Z. Understanding the benefits and challenges of using large language model-based conversational agents for mental well-being support. AMIA Annu Symp Proc. 2024;2023:1105-1114. [PMC free article] [PubMed] [Google Scholar]
27.Yu Y, Sharma T, Hu M, Wang J, Wang Y. Exploring parent-child perceptions on safety in generative AI: concerns, mitigation strategies, and design implications. Proc–IEEE Symp Secur Priv. 2025;2735-2752. doi: 10.1109/SP61157.2025.00090 [DOI] [Google Scholar]
28.Ng AE, Bottoms-McClain L, Black LI, Zablotsky B, Bose J. Comparison of parent and teen reports of teen healthcare use: US, July 2021–December 2023. In: National Health Statistics Reports. National Center for Health Statistics; 2025. doi: 10.15620/cdc/174622 [DOI] [Google Scholar]
29.Centers for Disease Control and Prevention . US teen girls experiencing increased sadness and violence. Accessed January 9, 2026. https://www.cdc.gov/media/releases/2023/p0213-yrbs.html
30.Ma Y, Ramos C, Wen H, Cummings JR. Racial and ethnic differences in mental health service use among adolescents. JAMA Netw Open. 2025;8(6):e2516612. doi: 10.1001/jamanetworkopen.2025.16612 [DOI] [PMC free article] [PubMed] [Google Scholar]
31.Adam D. Supportive? addictive? abusive? how AI companions affect our mental health. Nature. 2025;641(8062):296-298. doi: 10.1038/d41586-025-01349-9 [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplement 1.

eMethods 1. Survey Methodology

eMethods 2. List of Questions

jamapediatr-e262015-s001.pdf^{(211.2KB, pdf)}

Supplement 2.

Data Sharing Statement.

jamapediatr-e262015-s002.pdf^{(15.7KB, pdf)}

[poi260028r1] 1.Kennedy B, Yam E, Kikuchi E, Pula I, Fuentes J. AI in Americans’ lives: awareness, experiences and attitudes. Accessed January 21, 2026. https://www.pewresearch.org/science/2025/09/17/ai-in-americans-lives-awareness-experiences-and-attitudes/

[poi260028r2] 2.Faverio M, Sidoti O. Teens, social media, and AI chatbots 2025. Accessed December 9, 2025. https://www.pewresearch.org/internet/2025/12/09/teens-social-media-and-ai-chatbots-2025/

[poi260028r3] 3.Office of the Surgeon General . Protecting Youth Mental Health: The US Surgeon General’s Advisory. US Department of Health and Human Services; 2021. [Google Scholar]

[poi260028r4] 4.AHRQ. 2023 National Healthcare Quality and Disparities Report. Agency for Healthcare Research and Quality (US); 2023. [Google Scholar]

[poi260028r5] 5.Centers for Disease Control and Prevention. WISQARS Explore Fatal and Nonfatal Data. Accessed December 20, 2025. https://wisqars.cdc.gov/explore/?o=MORT&y1=2021&y2=2021&g=00&t=0&i=0&m=20810&d=&s=0&r=0&me=0&ry=2&yp=65&e=0&a=custom&a1=5&a2=24&g1=0&g2=199

[poi260028r6] 6.Verlenden JV, Fodeman A, Wilkins N, et al. Mental health and suicide risk among high school students and protective factors—Youth Risk Behavior Survey, US, 2023. MMWR. 2024;73(4):79-86. doi: 10.15585/mmwr.su7304a9 [DOI] [Google Scholar]

[poi260028r7] 7.Centers for Disease Control and Prevention . Bullying victimization among teeanagers: US, July 2021-December 2023. Accessed January 3, 2026. https://www.cdc.gov/nchs/products/databriefs/db514.htm

[poi260028r8] 8.Substance Abuse and Mental Health Services Administration . Key substance use and mental health indicators in the US: results from the 2023 National Survey on Drug Use and Health. Accessed January 3, 2026. https://www.samhsa.gov/data/sites/default/files/reports/rpt47095/National%20Report/National%20Report/2023-nsduh-annual-national.pdf

[poi260028r9] 9.Centers for Disease Control and Prevention . Data and statistics on children’s mental health. Accessed March 10, 2025. https://www.cdc.gov/children-mental-health/data-research/index.html

[poi260028r10] 10.Herbener AB, Damholdt MF. Are lonely youngsters turning to chatbots for companionship—the relationship between chatbot usage and social connectedness in Danish high-school students. Int J Hum Comput Stud. 2025;196:103409. doi: 10.1016/j.ijhcs.2024.103409 [DOI] [Google Scholar]

[poi260028r11] 11.Nagata JM, Memon Z, Huang O, Moreno MA. Adolescent health and generative AI—risks and benefits. JAMA Pediatr. 2026;180(1):7-8. doi: 10.1001/jamapediatrics.2025.4502 [DOI] [PMC free article] [PubMed] [Google Scholar]

[poi260028r12] 12.Giovanelli A, Roundfield KD. Adolescent vulnerability to consumer chatbots—artificial agents and genuine risk. JAMA Netw Open. 2025;8(10):e2539028. doi: 10.1001/jamanetworkopen.2025.39028 [DOI] [PubMed] [Google Scholar]

[poi260028r13] 13.Pretorius C, Chambers D, Cowan B, Coyle D. Young people seeking help online for mental health: cross-sectional survey study. JMIR Ment Health. 2019;6(8):e13524. doi: 10.2196/13524 [DOI] [PMC free article] [PubMed] [Google Scholar]

[poi260028r14] 14.Stade EC, Stirman SW, Ungar LH, et al. Large language models could change the future of behavioral health care: a proposal for responsible development and evaluation. Npj Mental Health Res. 2024;3(1):12. doi: 10.1038/s44184-024-00056-z [DOI] [Google Scholar]

[poi260028r15] 15.van Heerden AC, Pozuelo JR, Kohrt BA. Global mental health services and the impact of artificial intelligence-powered large language models. JAMA Psychiatry. 2023;80(7):662-664. doi: 10.1001/jamapsychiatry.2023.1253 [DOI] [PubMed] [Google Scholar]

[poi260028r16] 16.Lawrence HR, Schneider RA, Rubin SB, Matarić MJ, McDuff DJ, Jones Bell M. The opportunities and risks of large language models in mental health. JMIR Ment Health. 2024;11:e59479. doi: 10.2196/59479 [DOI] [PMC free article] [PubMed] [Google Scholar]

[poi260028r17] 17.McBain RK, Bozick R, Diliberti M, et al. Use of generative AI for mental health advice among US adolescents and young adults. JAMA Network Open. 2025;8(11):e2542281. doi: 10.1001/jamanetworkopen.2025.42281 [DOI] [PMC free article] [PubMed] [Google Scholar]

[poi260028r18] 18.RAND American Youth Panel . A nationally representative panel of young people. Accessed March 7, 2025. https://www.rand.org/education-and-labor/survey-panels/ayp.html

[poi260028r19] 19.US Department of Agriculture, Economic Research Service . Rural-urban commuting area codes. Accessed January 3, 2026. https://www.ers.usda.gov/data-products/rural-urban-commuting-area-codes

[poi260028r20] 20.Andoh E. AI chatbots and digital companions are reshaping emotional connection. American Psychological Association. Accessed January 8, 2026. https://www.apa.org/monitor/2026/01-02/trends-digital-ai-relationships-emotional-connection

[poi260028r21] 21.Shekar S, Pataranutaporn P, Sarabu C, Cecchi GA, Maes P. People overtrust AI-generated medical advice despite low accuracy. NEJM AI. 2025;2(6):AIoa2300015. doi: 10.1056/AIoa2300015 [DOI] [Google Scholar]

[poi260028r22] 22.Meng X, Yan X, Zhang K, et al. The application of large language models in medicine: a scoping review. iScience. 2024;27(5):109713. doi: 10.1016/j.isci.2024.109713 [DOI] [PMC free article] [PubMed] [Google Scholar]

[poi260028r23] 23.Malgaroli M, Schultebraucks K, Myrick KJ, et al. Large language models for the mental health community: framework for translating code to care. Lancet Digit Health. 2025;7(4):e282-e285. doi: 10.1016/S2589-7500(24)00255-3 [DOI] [PMC free article] [PubMed] [Google Scholar]

[poi260028r24] 24.Dwyer B, Flathers M, Sano A, et al. Mindbench.ai: an actionable platform to evaluate the profile and performance of large language models in a mental healthcare context. NPP Digit Psychiatry Neurosci. 2025;3(1):28. doi: 10.1038/s44277-025-00049-6 [DOI] [PMC free article] [PubMed] [Google Scholar]

[poi260028r25] 25.Abbasian M, Khatibi E, Azimi I, et al. Foundation metrics for evaluating effectiveness of health care conversations powered by generative AI. NPJ Digit Med. 2024;7(1):82. doi: 10.1038/s41746-024-01074-z [DOI] [PMC free article] [PubMed] [Google Scholar]

[poi260028r26] 26.Ma Z, Mei Y, Su Z. Understanding the benefits and challenges of using large language model-based conversational agents for mental well-being support. AMIA Annu Symp Proc. 2024;2023:1105-1114. [PMC free article] [PubMed] [Google Scholar]

[poi260028r27] 27.Yu Y, Sharma T, Hu M, Wang J, Wang Y. Exploring parent-child perceptions on safety in generative AI: concerns, mitigation strategies, and design implications. Proc–IEEE Symp Secur Priv. 2025;2735-2752. doi: 10.1109/SP61157.2025.00090 [DOI] [Google Scholar]

[poi260028r28] 28.Ng AE, Bottoms-McClain L, Black LI, Zablotsky B, Bose J. Comparison of parent and teen reports of teen healthcare use: US, July 2021–December 2023. In: National Health Statistics Reports. National Center for Health Statistics; 2025. doi: 10.15620/cdc/174622 [DOI] [Google Scholar]

[poi260028r29] 29.Centers for Disease Control and Prevention . US teen girls experiencing increased sadness and violence. Accessed January 9, 2026. https://www.cdc.gov/media/releases/2023/p0213-yrbs.html

[poi260028r30] 30.Ma Y, Ramos C, Wen H, Cummings JR. Racial and ethnic differences in mental health service use among adolescents. JAMA Netw Open. 2025;8(6):e2516612. doi: 10.1001/jamanetworkopen.2025.16612 [DOI] [PMC free article] [PubMed] [Google Scholar]

[poi260028r31] 31.Adam D. Supportive? addictive? abusive? how AI companions affect our mental health. Nature. 2025;641(8062):296-298. doi: 10.1038/d41586-025-01349-9 [DOI] [PubMed] [Google Scholar]

PERMALINK

AI Chatbot Use and Disclosure for Mental Health Among US Adolescents and Young Adults

Ryan K McBain, PhD, MPH

Jonathan H Cantor, PhD

Joshua Breslau, PhD, ScD

Melissa Diliberti, PhD

Li Ang Zhang, PhD

Fang Zhang, PhD

Alyssa Burnett, MPH

Aaron Kofner, MS

Benjamin Rader, PhD, MPH

Pat Pataranutaporn, PhD

Bradley D Stein, MD, PhD

Ateev Mehrotra, MD, MPH

Hao Yu, PhD

Key Points

Questions

Findings

Meaning

Abstract

Importance

Objective

Design, Setting, and Participants

Exposures

Main Outcomes and Measures

Results

Conclusions and Relevance

Introduction

Methods

Study Design

Procedures

Measures

Statistical Analysis

Results

Descriptive Characteristics

Table 1. Descriptive Characteristics of Survey Sample.

Use of AI Chatbots

Figure 1. Reported Use of Artificial Intelligence (AI) Chatbots for Mental Health Advice Among Young People in the US.

Respondent Characteristics Associated With AI Chatbot Mental Health Use

Table 2. Correlates of Artificial Intelligence (AI) Chatbot Use for Mental Health Advice Among Adolescents and Young Adults in the US.

Correlates of AI Chatbot Use Behaviors

Figure 2. Adjusted Odds of Asking Artificial Intelligence (AI) Chatbots for Mental Health Advice Monthly or More Often.

Discussion

Limitations

Conclusions

References

Associated Data

Supplementary Materials

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases