BACKGROUND-AIM:
Retrieval-Augmented Generation (RAG) Artificial Intelligence is a method used to increase the accuracy of Large Language Models (LLMs) and reduce hallucinations. RAG AI is built by fine-tuning foundational LLMs, using trusted data sources rather than relying solely on the data used to train the LLM. This paper aims to describe the development of a RAG AI Notebook designed to help youth gain knowledge and skills in identifying tobacco industry tactics and understanding health communication methods in tobacco control.

METHODS:
Using Google NotebookLM, we developed a RAG AI system to create a chatbot, along with visual and audio materials, for participants of a workshop dedicated to tobacco industry interference and youth advocacy. This workshop was organized as a youth-focused event during the European Conference on Tobacco Control 2025 (ENSP ECTC 2025). This paper describes the development and implementation of the RAG AI.

RESULTS:
Two in-person workshops were designed for young advocates and researchers at ENSP ECTC 2025. The first workshop, led by the Institute of Economic Sciences from Belgrade, covered tobacco industry tactics to interfere with tobacco taxation. The second workshop, led by the youth group of the European Network for Smoking and Tobacco Prevention (ENSPNext) and the National Initiative of Non-Smokers from Serbia (NINS), covered best practices in tobacco control advocacy. The AI Notebook was developed through a four-step process. First, digital resources available for fine-tuning the LLM were selected based on following criteria: quality, relevance, and copyright regulations. Resources included: Framework Convention on Tobacco Control, Policy Papers, Health communication guidelines, etc. Second, the RAG AI Chatbot was tested by simulating user scenarios of increasing complexity, such as asking for information not present in the training resources, requesting simple facts present in training resources, and requesting complex information synthesized from multiple training resources. Third, the chatbot was personalized to user needs regarding language, answer length, and structure. The „learning guide“ template by NotebookLM was chosen. Finally, explainer videos, podcasts, interactive mind maps, and quizzes were generated based on the selected digital resources to guide users in utilizing the AI chatbot for detailed study.

CONCLUSIONS:
The integration of RAG AI technology into tobacco control training has great potential for empowering youth advocates and researchers in this field. By fine-tuning LLMs using verified legal and scientific documents, we successfully created a reliable, interactive tool that lowers the risk of hallucinations by AI and allows for AI generated materials to be tailored specifically to the provided training resources. Future research and implementation should focus on upscaling such RAG AI systems for application in tobacco control educational campaigns and providing tailored messages for smoking cessation based on scientific evidence.