Chatting with a bot "trained" by digesting enormous amounts of text can be an interesting way to surface novel combinations of sources, identify robust and weaker patterns within those sources, and make new connections between superficially disconnected elements. As with all of the different generative AI types you encounter, it's important to provide the right information and instruction to the AI in your prompts.
Tools like these are usually called LLMs, "large language models," because they are made out of vast quantities of natural language textual strings; The other vital piece of an LLM is a complex set of rules for combing through and recombining the elements of those textual strings. If that sounds a lot like the way all databases work, you're not far from the truth!