This article is an account of a 5-year-long process of creating an AI-based tool for use in our university library to support work at the information desk, especially when it is unmanned. Our first idea of creating a machine that could answer recurring simple questions at the library's information desk developed out of frustration. Information desk work was increasingly being reduced to questions such as "do you have this book?" or "how can I get a library card”. We started out 2019 by creating a chatbot that embraced Artificial Intelligence (AI) and natural language processing (NLP), which was connected to several database sources.In 2021, we purchased a "social robot head" that mimics human faces, and dialects, and can be set to imitate different accents.It was challenging converting from chatbot-typed questions to speaking to voice recognition. The problems of misunderstandings and misinterpretations are extensive, but at the same time the advantage of using natural language is significant for the user.During a student survey, we realized that the rules processed in the Microsoft Luis natural language system often interfered with the intentions received from the Microsoft QnA maker. This problem was solved in 2023 when we were able to integrate Chat GPT into our project. This move improved the performance of our social robot. It is now faster, more reliable, can respond much better to “social” questions”. It searches successfully for book titles, authors, subjects, can filter out digital or physical books according to preference and help students with library cards and related questions.During our study, we noted that students can be hesitant to ask questions out loud to a robot in the library. We therefore would recommend using a social robot in a closed environment, like a pod, while simultaneously offering students the opportunity to ask questions via a keyboard.