Managing PMS often happens in isolation. Many people avoid talking about their symptoms due to stigma, shame, or simply not having someone who understands. This research explored whether chatting with multiple AI chatbots — designed to simulate a small group therapy session — could help people feel less alone and more motivated to cope with PMS.
A team of researchers in Japan built a system with three chatbots: one acting as a group facilitator and two acting as peers who also experience PMS. The peer bots had different personalities — one was optimistic and shared coping advice, the other was more reserved and empathetic. Sixty-three women with moderate to severe PMS (including some with symptoms at the PMDD level) participated over two menstrual cycles. They were split into three groups: no intervention, one-on-one chatbot conversations, or the group chatbot format.
People in the group chat wrote longer, more detailed responses and began using language more similar to the chatbots over time — including more positive and certain language. In interviews, many said the group format helped them feel they were not alone, gave them new ideas for coping (like herbal tea, aromatherapy, or walks), and even encouraged some to talk about PMS with friends in real life. However, some participants felt frustrated by repetitive conversations, overwhelmed by the amount of text, or experienced jealousy when the chatbot peers seemed to cope better than they did.
Importantly, neither chatbot condition led to a measurable reduction in PMS symptom severity or impact on daily life over the two months. The researchers suggest that daily tracking and talking about symptoms may have actually heightened awareness of them. While the system did not reduce symptoms directly, it showed promise for providing social support and helping people develop new ways to think about and manage their PMS experience.
Key findings
- Participants in the group chat condition wrote significantly longer responses to all six motivational interviewing questions compared to 1-on-1 chat participants (e.g., Q1 Symptoms: group mean 41.23 chars vs 1-on-1 mean 25.16 chars, p=.001, rrb=0.57)
- Group chat participants showed significantly higher language style matching (LSM) scores with all four chatbots compared to 1-on-1 participants, indicating greater linguistic convergence (e.g., LSM with peer bot Riko: group mean 0.84 vs 1-on-1 mean 0.79, p=.009, rrb=0.47)
- Group chat participants used significantly more positive emotion words (p=.010) and certainty words (p=.048), and fewer anger words (p=.003) and risk words (p=.006) compared to 1-on-1 participants
- Neither chatbot condition (1-on-1 or group) showed significant longitudinal improvement in perceived PMS severity or work-life interference over two months, while the no-intervention condition showed a significant decrease in PMS severity (F[2,20]=3.448, p=.042) and work-life interference (F[2,20]=5.060, p=.011)
- Qualitative analysis revealed group chat participants experienced a sense of belonging, social learning from peer bots' coping strategies, and motivation to manage PMS, but also negative feelings including jealousy, social comparison, and frustration with chat length
- Participants reported that peer bots helped them verbalize symptoms, learn attribution styles (attributing discomfort to PMS), and adopt new coping strategies such as herbal tea, aromatherapy, and walks
Methods, briefly
Between-subjects longitudinal study over two menstrual cycles with N=63 women in Japan (21 per condition: no intervention, 1-on-1 chatbot, multi-chatbot group chat). Chatbots were built using GPT-4 on Slack in Japanese. The group chat included a facilitator bot and two peer bots with distinct personalities. Quantitative measures included character count, J-LIWC word category analysis, language style matching (LSM) scores, and PSST questionnaire scores. Qualitative data from semi-structured interviews (12 per chatbot condition), in-chat feedback, and open-ended questionnaire responses were analyzed using bottom-up thematic analysis. Mann-Whitney U tests compared conditions; repeated-measures ANOVA assessed longitudinal PMS changes.
Limitations to keep in mind
- All participants were from Japan, limiting cultural generalizability given that attitudes toward menstruation vary across cultures
- Study duration was only two months (two menstrual cycles), limiting assessment of long-term effects and attrition
- Predefined motivational interviewing questions restricted conversational autonomy and variability
- Self-report measures (PSST) were used for PMS severity assessment
- Peer bot symptoms were randomly generated rather than dynamically personalized to each participant
- Participants were recruited via crowdsourcing and self-rated their PMS severity, without clinical diagnosis
- Neither chatbot condition showed significant improvement in PMS severity or work-life interference compared to no intervention
The latest PMDD research, in your pocket
The app’s research feed is updated daily with newly published, peer-reviewed PMDD work - summarized in plain language.