Seventy percent of the fifty teenagers who sat down with the chatbot in Lima were already carrying more than minimal depressive symptoms. Sixty percent had anxiety symptoms above that same threshold. Ninety-two percent reported moderate or high stress. These were adolescents living with HIV, a group that researchers have long known struggles with mental health, and a group whose depression tends to show up later in the numbers that matter most for survival: whether they take their antiretroviral medication, whether the virus stays suppressed.
The chatbot is called EVA, short for Educación, Vinculación y Autoayuda, which translates roughly as education, linkage and self-help. Jerome Galea of the University of South Florida and colleagues in Peru and at Harvard Medical School built it with input from the very adolescents it was meant to serve. Its job is modest and specific: explain what depression and anxiety are, teach a few self-help skills, and point users toward real mental health services.
Between January and August of 2024, the team recruited fifty adolescents aged 11 to 19 in Lima. Each one filled out a baseline questionnaire, then spent twenty minutes alone with EVA, then filled out the questionnaire again. The main thing being measured was depression knowledge, scored on something called the Adolescent Depression Knowledge Questionnaire. Average scores went from 6.88 before to 8.12 after. The authors report that as a statistically significant jump with a large effect size.
Participants also rated EVA highly on whether it felt acceptable, appropriate and workable, and said they intended to use it again and recommend it to others. They liked the educational content, found it usable, and rated its self-help material well.
What twenty minutes can and cannot show
The honest reading of this study is narrow, and the authors do not pretend otherwise. This is a pilot, posted as a preprint on medRxiv, which means it has not yet been through peer review. Every participant got the chatbot; there was no comparison group sitting in a waiting room or reading a pamphlet. That matters because a quiz taken twice in one sitting can improve for reasons that have nothing to do with the intervention. People get better at a test the second time they see it.
The measured outcome is also knowledge, not health. Knowing more about depression is not the same as feeling less depressed, and it is not the same as walking into a clinic and asking for help. The team is explicit that future longitudinal work is needed to find out whether anything sticks, and whether more knowledge actually turns into help-seeking behavior. A twenty minute session tells you nothing about what a teenager remembers a month later.
And the researchers add a caution that reads as unusual in a paper reporting its own tool working: the benefits of mental health chatbots, they write, should be weighed against their potential mental health risks. They do not elaborate in the abstract on what those risks are. The inclusion of the sentence at all is worth noticing.
Why it matters
Mental health care is not well woven into HIV programs, and that gap is widest in low and middle income countries where most people with HIV actually live. A teenager in Lima who is depressed and living with HIV may have a clinic that manages their viral load beautifully and no one at all who asks how they are doing. Hiring enough psychologists to close that gap is expensive and slow. A chatbot is neither.
That is the argument the authors are making, and it is a reasonable one to test. Something that costs almost nothing to run, that a teenager can use privately on a phone without telling anyone, could reach people that a referral system never will. The stigma around both HIV and mental illness makes privacy genuinely valuable rather than merely convenient.
But the case rests on a small pilot with no control arm, and the honest verdict is that fifty adolescents in one city demonstrated that a co-designed chatbot is something they will use and something they can learn from in a single sitting. That is a real finding. It is a starting point rather than evidence that EVA improves anyone's mental health or their HIV outcomes.
What makes the study interesting is less the score change than the population it reached. Adolescents living with HIV are chronically under-served by mental health research, and the symptom numbers here (seven in ten with depressive symptoms above minimal) are a finding in their own right, independent of whether the chatbot worked. Fifty teenagers walked in and most of them were struggling. Whatever the eventual verdict on chatbots, that part of the picture does not depend on the intervention at all.