Effect of a Breastfeeding Chatbot on Mothers' Breastfeeding Attitudes, Behaviors, and Self-Efficacy

Study Design and Objectives

This study is designed as a two-phase research project involving the development and evaluation of an artificial intelligence (AI)-based breastfeeding counseling chatbot. The first phase involves chatbot development, expert validation, technical testing, and pilot evaluation. The second phase involves a two-arm, parallel-group, single-blind randomized controlled trial to evaluate the effectiveness of the chatbot in supporting breastfeeding outcomes among primiparous mothers.

The study aims to evaluate the effects of the chatbot on exclusive breastfeeding, breastfeeding continuation and duration, breastfeeding self-efficacy, breastfeeding attitudes, and breastfeeding behaviors.

Development of the AI-Based Chatbot

The chatbot will be developed using natural language processing (NLP) and Retrieval-Augmented Generation (RAG) technology. Its knowledge base will contain evidence-based breastfeeding guidelines, frequently asked questions, structured educational materials, and expert-reviewed breastfeeding counseling content.

The system will use the nomic-embed-text embedding model for information retrieval and the qwen3.5 language model for response generation. Retrieved content will be used to generate contextually relevant responses grounded in the approved knowledge base.

The chatbot will provide breastfeeding education, answers to frequently asked questions, guidance on breastfeeding techniques, and access to relevant educational resources. It will not provide medical diagnoses, prescribe medications, or replace healthcare professionals. Situations requiring clinical assessment will be directed to appropriate healthcare services.

Before implementation, the system will undergo expert review, technical performance assessment, and pilot testing with 10 mothers who will not be included in the randomized controlled trial.

Participants and Randomization

The randomized controlled trial will be conducted at Adiyaman University Training and Research Hospital in Türkiye.

The study is currently planned to include 92 primiparous mothers, with 46 participants allocated to each group. Eligible participants will be aged 18 years or older, have delivered a healthy term singleton infant, intend to breastfeed, and have access to a smartphone and the internet.

Participants will be randomly allocated to the intervention or control group in a 1:1 ratio using computer-generated permuted block randomization. The randomization sequence will be generated by an independent researcher. Group allocation will occur after eligibility assessment, written informed consent, and baseline data collection.

Due to the nature of the digital intervention, participants cannot be blinded to group allocation. However, group identities will be masked during statistical analysis to reduce analytical bias.

Intervention and Control Groups

Participants in the intervention group will receive standard breastfeeding counseling provided by healthcare professionals, in addition to access to the AI-based breastfeeding counseling chatbot for six months postpartum.

The chatbot will provide evidence-based information and educational support related to breastfeeding difficulties. User interactions and system usage data will be recorded for evaluation.

Participants in the control group will receive routine breastfeeding education and counseling provided by healthcare professionals. No additional digital breastfeeding intervention will be provided as part of the study.

Data Collection and Outcome Assessment

Data will be collected at four time points:

T0: Within the first 24 hours postpartum.

T1: At one month postpartum.

T2: At three months postpartum.

T3: At six months postpartum.

Data collection instruments will include a Participant Information Form, the Iowa Infant Feeding Attitude Scale (IIFAS), the Breastfeeding Self-Efficacy Scale-Short Form (BSES-SF), and a Breastfeeding Follow-up Form.

The study will evaluate exclusive breastfeeding status at six months postpartum, breastfeeding continuation and duration, breastfeeding self-efficacy, breastfeeding attitudes, and breastfeeding behaviors.

Technical Performance and Safety Evaluation The technical performance of the chatbot will be evaluated using information retrieval metrics, including Precision@5 and Recall@5, expert assessment of response accuracy, consistency with source information, response relevance, hallucination rates, appropriate referral performance, response time, and system error rates.

User satisfaction, usability, and interaction patterns will also be evaluated. The chatbot will operate within predefined safety boundaries and will provide educational and supportive information rather than diagnostic or therapeutic recommendations.

Expected Contribution This study is expected to provide evidence regarding the effectiveness, safety, usability, and feasibility of an AI-supported breastfeeding counseling system. The findings may contribute to the development of evidence-based digital interventions supporting breastfeeding practices and postpartum maternal care.