Are US parents comfortable with AI-led therapy for their children?
Mostly not in one weighted online survey: 76% of 572 responding parents said they were not comfortable at all or not too comfortable. The study measured a hypothetical attitude—not whether AI therapy works, is safe or would be accepted in practice.
The short version
- What happened
- Among 572 US parents who answered the main question, weighted estimates were 47.2% not comfortable at all, 28.8% not too comfortable, 16.2% somewhat comfortable and 7.8% very comfortable.
- Who it affects
- Parents in this survey prioritised diagnostic accuracy, empathy, professional recommendation and confidentiality more often than cost, scheduling flexibility or shorter waits.
- What remains unproven
- The online river sample was nonprobability and self-selected; demographic weighting cannot remove every source of selection bias.
Keep in mind
The online river sample was nonprobability and self-selected; demographic weighting cannot remove every source of selection bias.
Only 572 parents answered the primary outcome and 554 entered the adjusted model, limiting precision and detailed subgroup analysis.
The question described an undefined AI-led therapy session, so respondents may have imagined different products, levels of autonomy and safeguards.
Comfort, expectations and prior experience were measured at the same time; associations cannot establish cause or predict actual consent and use.
No AI system, treatment comparison, child outcome, safety event, crisis response or privacy practice was evaluated.
Source analysis · 2 linked sources. Primary analysed source date: 2 November 2026. Earliest linked supporting source: 22 May 2025.
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Read the full report · 9 min The evidence, methods, implications and limitations
The answer is mostly no—but the study measured a hypothetical choice
A peer-reviewed JAMA Pediatrics research letter found that most parents in one weighted US online survey expressed low comfort with the idea of an AI leading a behavioral therapy session for their child. Of 572 parents who answered the main question, 47.2% were estimated to be not comfortable at all and 28.8% not too comfortable. Together, those categories make 76.0%. A further 16.2% were somewhat comfortable and 7.8% very comfortable. The intervals around the two low-comfort estimates were 42.1% to 52.3% and 24.5% to 33.6%, respectively.
That result is evidence about attitudes, not treatment. No child interacted with an AI system. The researchers did not name or test a therapy product, compare AI with a clinician, measure symptom change, assess crisis handling or examine privacy in operation. The survey asked how comfortable a parent would be with a therapy session led by an AI. Someone can answer that question without knowing what the system would do, how a human professional would supervise it or what safeguards would apply. The finding should therefore inform design and governance discussions, not be used to declare AI therapy acceptable or unacceptable on clinical grounds.[1]
Who answered and how the survey was assembled
The researchers surveyed US adults in November 2025 through the OutbreaksNearMe participatory surveillance platform. Recruitment used what the paper calls a nonprobability end-page river sample on SurveyMonkey: people encountered an invitation online rather than being randomly selected from a complete list of US households. Of 3,102 respondents, 584 said they had at least one child aged 5 to 17. Twelve did not answer the main outcome, leaving 572 responses for the comfort distribution. The adjusted regression used complete information from 554 parents.
Survey weights were applied to approximate US Census demographic targets for age, gender, race and ethnicity, education and income. After weighting, 53.3% of the parent sample was female, 45.7% male and 1.0% not listed or nonconforming. The weighted race and ethnicity distribution was 6.2% Asian or Pacific Islander, 12.0% Black, 23.6% Hispanic, 54.2% White and 3.9% other. Weighting can make observed margins resemble population targets, but it cannot guarantee that people who encounter and choose an online survey resemble people who do not. Attitudes connected to internet use, health access or AI familiarity may remain different even after demographic adjustment.[1]
The distinction that matters
From 3,102 respondents to one hypothetical attitude
The denominator narrows at each step, and none of the steps includes a child receiving AI-led therapy.
Online survey
3,102 US adults responded
Nonprobability sample recruited through an online platform
Eligible parents
584 reported a child aged 5 to 17
572 answered the main comfort question
Adjusted model
554 had complete outcome and covariate data
Associations describe reported comfort, not causation
Not tested
No child received an AI-led session
No evidence of efficacy, safety, privacy or real uptake
What parents said would matter most
The survey also asked parents to select factors that would matter in deciding about AI-led therapy. Among 546 respondents to that item, 58.5% selected accuracy of diagnosis, 49.4% empathy, 46.8% a recommendation from a medical professional and 43.7% confidentiality and privacy. Convenience factors ranked lower: 32.8% selected cost, 29.5% scheduling flexibility and 27.7% appointment wait time. Respondents could select factors rather than rank a single exclusive priority, so the percentages do not add to 100%.
These responses complicate a common argument that families facing long waits will automatically accept automation. In this sample, people more often pointed to clinical and relational qualities than access advantages. Yet the wording also blends diagnosis with therapy. Behavioral therapy does not always require a new diagnosis, and an AI tool might range from a supervised worksheet to an autonomous conversational agent. Without a concrete scenario, parents may have imagined different technologies. Future research needs to separate an administrative assistant, a clinician-supervised tool, a skills-practice application and a system presented as the therapist.[1]
The strongest associations are not proof of what changes trust
In an adjusted ordinal regression of 554 complete responses, comfort with AI-led therapy for one’s own care was associated with much higher odds of reporting greater comfort for a child: adjusted odds ratio 20.94, with a wide 95% confidence interval from 9.78 to 44.86. Comfort with AI-led primary care for oneself was also associated, with an adjusted odds ratio of 3.78. Parents who expected AI to perform better than medical professionals had higher reported comfort than those expecting worse performance, with an adjusted odds ratio of 10.12; expecting similar performance had an adjusted odds ratio of 3.31.
Those associations do not show that changing an expectation will cause acceptance. The survey measured attitudes at one time, and closely related questions may capture the same underlying disposition toward AI health care. The very large estimate for parents’ own therapy comfort and its wide interval are reasons for care, not a lever ready for a campaign. Parents whose child had received behavioral therapy in the previous year also reported higher comfort, with an adjusted odds ratio of 2.07. General AI use by parents or children was not associated with comfort after adjustment. Familiarity with any AI is not necessarily familiarity with a sensitive clinical relationship.[1]
What the study does not tell families or clinicians
The survey cannot answer whether AI-led behavioral therapy benefits or harms children. It did not measure symptom severity, diagnosis, treatment history in detail, age-specific understanding, crisis risk or the response of a particular model. It cannot show whether a parent who reports comfort would consent when presented with data-use terms, a clinician’s recommendation and a real child’s needs. Nor can low comfort show that a safe, effective, professionally supervised intervention should be rejected. Comfort is relevant to adoption and consent, but it is not a clinical endpoint.
The American Psychological Association’s youth AI advisory recommends age-appropriate defaults, clear disclosure that the user is interacting with a nonhuman system, reduced persuasive design, human intervention routes, continuous testing with diverse young people and strong privacy protection. Those are proposed safeguards, not a declaration that autonomous AI therapy is ready. For behavioral health, evaluation also needs escalation procedures for self-harm, abuse or deteriorating symptoms; boundaries on what the system may advise; independent review of error and bias; and a clear accountable professional or service when technology fails.[1][2]
What this means for children, carers and services
For parents and carers, the immediate message is not to infer quality from a chatbot’s warmth or fluency. A system can produce empathetic language without understanding a child, maintaining confidentiality like a licensed professional or recognizing when a conversation needs urgent human help. Families should be told what the tool does, what data it retains, who reviews interactions, how to reach a person and what evidence supports the specific use. Consent should distinguish optional support from care that a service effectively requires families to use.
For clinicians and health systems, the study suggests that access alone will not settle acceptance. A service may reduce waiting for low-intensity support yet still fail if families cannot trust its accuracy, empathy, privacy or professional oversight. Trust should not be manufactured through a reassuring interface. It should be earned through independent clinical evidence, transparent limits, inclusive design and reliable human escalation. The result also argues against treating parents as a single group: the sample was too modest for precise subgroup assessment, and families may differ by a child’s age, condition, culture, previous care, disability and access to alternatives.[1][2]
What evidence would change the assessment
A stronger attitudes study would recruit a probability-based national sample or report how nonresponse may affect results, preregister the main analyses and present estimates for sufficiently powered subgroups. It would describe several concrete service models—such as clinician-supervised skills practice, an AI assistant between appointments and an autonomous therapy session—so respondents evaluate the same proposition. Repeated measurement after balanced information about benefits, risks, privacy and oversight could show which concerns are stable and which reflect missing information.
Clinical evidence requires a different design. A specific system should be compared prospectively with existing care or an appropriate control, with consent from families and age-appropriate assent from children. Researchers should measure validated symptoms, engagement, adverse events, crisis escalation, privacy failures, clinician workload and unequal outcomes, not only satisfaction. Independent replication and longer follow-up would be necessary before broad deployment. Until then, the clearest finding is that most parents in this survey were uneasy about an undefined AI leading therapy—and that their concerns centred more on diagnosis, empathy, professional recommendation and privacy than on convenience.[1][2]
Watch the background, then check the evidence
Illustrative photo · not a video frame
The Royal Society · 44 min 28 sec · Background explainer
Is this the AI future we want? | The Royal Society
YouTube loads only when you choose; playback starts when you press play.
A Royal Society discussion with Madhumita Murgia and biomedical engineer Alison Noble on AI’s potential in healthcare, science and culture, recorded at the 30 June 2026 event.
Evidence check: A discussion of opportunities and risks, including future possibilities. It is not evidence that every application described is available or clinically proven.
Published 2026-07-01. Source checked 14 October 2026. Check captions on the publisher’s video page.
Check the healthcare evidencePhoto: Daniel Sone / National Cancer Institute / Unsplash · Licence
Continue the story
Read next
Continue exploring health privacy: compare the finding and its evidence limits.
Americans will accept healthcare AI help—but want people making the final call
A Pew survey of 3,488 US adults found broad openness to AI for translation and scheduling, but much less appetite for autonomous diagnosis or insurance decisions. The findings measure public preferences, not clinical performance.
Read the next reportExplore the full source and evidence record
What this means for people
- Parents in this survey prioritised diagnostic accuracy, empathy, professional recommendation and confidentiality more often than cost, scheduling flexibility or shorter waits.
- Children received no intervention in this research. Any real service would still need age-appropriate consent, privacy protection, clinical oversight and reliable escalation to a responsible person.
Global context
The survey concerns US parents and approximates selected US Census margins. Behavioral-health systems, professional roles, waiting times, privacy law and cultural expectations vary internationally. The percentages should not be transferred to another country or to all US families without replication, especially where access to human care or trust in health institutions differs.
What the evidence does not yet show
- The online river sample was nonprobability and self-selected; demographic weighting cannot remove every source of selection bias.
- Only 572 parents answered the primary outcome and 554 entered the adjusted model, limiting precision and detailed subgroup analysis.
- The question described an undefined AI-led therapy session, so respondents may have imagined different products, levels of autonomy and safeguards.
- Comfort, expectations and prior experience were measured at the same time; associations cannot establish cause or predict actual consent and use.
- No AI system, treatment comparison, child outcome, safety event, crisis response or privacy practice was evaluated.
What to watch next
- Probability-based surveys that distinguish supervised support from autonomous AI-led therapy
- Age, diagnosis, disability, language and socioeconomic subgroup estimates with adequate sample sizes
- Independent trials of specific systems measuring symptoms, adverse events, privacy and crisis escalation
- Whether professional recommendation changes informed choices after benefits and risks are explained
- Clear rules for consent, family involvement, clinical accountability and access to human care
Living evidence record
Impact record IAI-0IMGIZY
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Present
Record status
Monitoring
Last checked
3 November 2026
Source trail
2 direct sources across 2 source types.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Related-source reporting disclosure
This record analyses 2 linked source records around the same underlying development. The extra records add method, date or context, but they do not by themselves constitute independent replication of every performance claim or predicted outcome.
Evidence trail
Sources used for this report
Links checked 3 November 2026
Research question, topics and regions
How comfortable US parents say they would be with an AI-led behavioral therapy session for a child
Topics: Children · Behavioral health · AI therapy · Parent attitudes · Health privacy · Clinical oversight
Regions: United States · North America
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.