So You Want to Build a Chatbot?: A Systematic Case Study Comparing the Design and Development of Two Water Chatbots

Chatbots have become a popular method through which to deliver conversational-style information to users about a range of topics, including providing customer service, news and weather updates, educational content, and medical information. This article compares two chatbots created with different methods, including via custom architecture and custom GPT to determine the strengths and limitations of the development methods. The bots that our research team developed were built to deliver information about water and drought to Arizona residents. We compare the initial setup process, customization capabilities, the training process, prompt engineering requirements, file handling, costs, and outputs of each bot. The custom architecture bot offers the flexibility and control of answers, but it costs more than its comparator and takes more time. The custom GPT requires little experience with Large Language Models (LLMs) and no experience with coding, but offers less control. Because we recognize that public agencies often don't have the expertise or funding to build a fully-customized bot architecture, we conclude with suggestions about the contexts and purposes or which each type of bot should be developed.