As long as the search part of search + prompt is good, the prompt part will emit accurate results or will say it couldn't find the answer. You can also cite the

As long as the search part of search + prompt is good, the prompt part will emit accurate results or will say it couldn't find the answer. You can also cite the sources this way. It seems pretty expensive because you may be pasting a lot of context into each query. If you allow the user to have follow up queries and you want to retain context of the conversation that seems expensive too. But it does seem like it should give the best results for Q&A as long as question is directly answered in your data somewhere.