<![CDATA[Background: Inadequate reproductive health knowledge among pregnant women contributes to high maternal morbidity in Indonesia. Digital health promotion, particularly through recurrent neural network (RNN)-based chatbots, offers accessible and personalized health education. However, qualitative evidence on how such chatbots shape knowledge and experience remains scarce. Objective: This study aimed to explore pregnant women's experiences in acquiring reproductive health knowledge through RNN-based chatbot health promotion media at primary health centers, focusing on the depth and meaning of perceived knowledge change. Methods: A qualitative study with an interpretive phenomenological approach was conducted involving 24 pregnant women from two Puskesmas in Padangsidimpuan, Indonesia. Participants used the "Bunda Cerdas" chatbot based on LSTM architecture for two weeks. Data were collected through semi-structured in-depth interviews and analyzed using Braun and Clarke's thematic analysis. Trustworthiness was ensured through triangulation and member checking. Results: Six main themes emerged: (1) integrated thematic understanding, (2) convenient access to trusted information, (3) engaging and human-like interaction, (4) empowerment in health decision-making, (5) transformative knowledge retention, and (6) technical-social barriers. Participants reported a shift from fragmented to comprehensive knowledge, particularly in recognizing danger signs, nutritional needs, and birth preparedness. The conversational memory and contextual replies of the RNN chatbot fostered a sense of reliable companionship. Conclusion: RNN-based chatbot media substantively enriched pregnant women's reproductive health knowledge through personalized, dialogic learning. Its integration into primary care services can complement standard antenatal education, addressing knowledge gaps while respecting local cultural context. Further mixed-methods research is recommended to quantify outcomes and ensure sustainable implementation.]]>
