Following this phase of familiarization with the topic, we explored the questions of how chatbots work, what user data is important, and what requirements must be met to be able to train and program a chatbot in a hands-on interactive setting. We chose a simplified "Wizard-of-Oz" approach to interactively allow the participants to take a playful approach to developing simple dialogs for a recipe chatbot. For this, we divided the group into two: one group took on the role of the chatbot, and the other the role of the user. Each group tried to anticipate what the other actor could bring into the conversation. The chatbot group wrote down potential answers to hypothetical questions that the user could ask; the user group wrote down potential questions that they would expect a recipe chatbot to be able to answer. This initial phase of "training" the chatbot ended with a first test run: the user group would select an input card from their stack of cards which the chatbot group tried to respond to meaningfully by selecting a card from their stack that they believed could fit the user input. When the chatbot group realized, in this first round, that they had not anticipated the user's input that they were now confronted with, they reacted creatively: The users had asked for a recipe for vegetarian lasagna. The chatbot group only had a card with "meat" and with "vegetables" and decided to use these two cards in combination as the ingredient "meat-alternative," creating a recipe that includes "vegetable meat."