https://mitchellh.com/writing/prompt-engineering-transactional-prompting > **[[Interactive prompting]]** is ChatGPT-style prompting where you're having more of a _discussion_ with the language model. If the model returns an unclear response, you can use the pre-existing context to make clarifications and guide the model towards the correct response. Further, interactive prompting is driven primarily by a human being. > **[[Transactional prompting]]** treats the language model more like a function in a programing language: you give it some input and desire some output. This isn't limited to a single prompt, you may chain prompts or perform "agent"-like behaviors here. The point is that you're treating the language model transactionally for a very specific problem, usually at high volume, and driven by software.