Maternal AI chatbot shows zero postpartum mental health gain
Access to Rosie, an AI chatbot for maternal health, showed no statistically significant anxiety or depression benefit at 12 months compared with a monthly children’s book program in a 400-person randomized trial.
The study, published Nov. 4 in JAMA Network Open, tested access to health advice among pregnant and postpartum participants with generally low starting symptom scores.
Rosie was designed to answer maternal and infant health questions, offer personalized guidance, and provide reminders and educational support. A clinical social worker reviewed distress alerts, contacted participants, and provided resources, so the intervention included human support alongside the chatbot.
Researchers assigned 200 participants to each group. The US study enrolled English-speaking people aged 14 or older who were pregnant or parenting an infant younger than six months and identified as members of racial or ethnic minoritized groups.
Enrollment ran from October 2023 through December 2024, with follow-up completed in December 2025.
The primary measures were self-reported anxiety symptoms on the GAD-7 questionnaire and depression symptoms on the PHQ-9 questionnaire. The 12-month analyses included 361 participants for anxiety and 362 for depression.
The adjusted difference between the chatbot and control groups was 0.10 points for anxiety, with a 95% confidence interval of -0.77 to 0.96. For depression, it was -0.23 points, with an interval of -1.05 to 0.59.
Both intervals include zero, leaving uncertainty about the size and direction of any underlying effect. The trial established neither a clear benefit nor equivalence between the approaches.
Read the two primary outcomes
Read the two primary outcomesAnxiety · GAD-7Depression · PHQ-9
Anxiety: adjusted mean difference 0.10 points. The 95% confidence interval crosses zero.
Anxiety: adjusted difference and 95% confidence intervalReported adjusted mean difference 0.1; 95% confidence interval -0.77 to 0.96. The interval crosses zero.+0.10 points95% CI: -0.77 to 0.96-10+1Chatbot minus control, questionnaire pointsZero = no between-group differenceLower scores mean fewer symptoms.Line: confidence interval. Dot: estimate.
Anxiety: adjusted difference and 95% confidence intervalReported adjusted mean difference 0.1; 95% confidence interval -0.77 to 0.96. The interval crosses zero.+0.10 points95% CI: -0.77 to 0.96-10+1Chatbot minus control, questionnaire pointsZero = no between-group differenceLower scores mean fewer symptoms.Line: confidence interval. Dot: estimate.
Depression: adjusted mean difference -0.23 points. The 95% confidence interval crosses zero.
Depression: adjusted difference and 95% confidence intervalReported adjusted mean difference -0.23; 95% confidence interval -1.05 to 0.59. The interval crosses zero.-0.23 points95% CI: -1.05 to 0.59-10+1Chatbot minus control, questionnaire pointsZero = no between-group differenceLower scores mean fewer symptoms.Line: confidence interval. Dot: estimate.
Depression: adjusted difference and 95% confidence intervalReported adjusted mean difference -0.23; 95% confidence interval -1.05 to 0.59. The interval crosses zero.-0.23 points95% CI: -1.05 to 0.59-10+1Chatbot minus control, questionnaire pointsZero = no between-group differenceLower scores mean fewer symptoms.Line: confidence interval. Dot: estimate.
Both confidence intervals cross zero. The trial establishes neither a clear benefit nor equivalence.
US, English-speaking pregnant/postpartum participants from racial or ethnic minoritized groups, with generally low baseline symptoms. 400 randomized; endpoint samples: anxiety 361, depression 362. Outcomes were self-reported. Rosie included incentives and human support; both groups could contact staff.
Human support and low baseline symptoms in AI chatbot study
Participants received small payments for questions and book rewards for the most active users. The control group received one children’s board book per month for a year, and both groups could communicate with study staff.
Those conditions limit how directly the findings translate to routine chatbot use.
Participants generally began with low symptom scores. Among those with available baseline responses, the median anxiety score was 2, and the median depression score was 3. About 11.1% and 9.5% of the participants, respectively, met the study’s moderate-to-severe threshold.
Self-reported health outcomes are another limitation, and the authors urged caution about secondary and exploratory findings.
Rosie, delivered with incentives and human backup to this population, showed no statistically significant 12-month advantage on the two primary mental-health measures. Whether other AI interventions or different clinical populations would yield different results remains open.
