Most effective way to Chat-ify a website's FAQ section?

(a) Embed the FAQs, and use semantic search to match a user’s query to the best FAQ? Or, (b) ingest the FAQs into a GPT and let it answer questions directly?

I’m looking for the approach that is the most effective. Ease of setup is less important to me than getting the user to the right answer within one or two interactions. I already have the framework for doing embedding + semantic search on a custom document corpus. I’m already using embedding model and 3.5 for more general completions, so generalizing my existing system to handle FAQs is straightforward for me. If (a) and (b) are likely to be equally effective, then maintainability might be a secondary criterion.

Ultimately, I’ll be generalizing what I already have to route the user to the best page on the site; not sure if a GPT can do that? In particular, can it include links in its response? If not then being that I plan to build a more capable routing system anyway (and because that subsumes the requirements of the FAQ Q&A feature), I’d lean towards doing (a).

1 Like

I think your conclusion makes sense. Also bear in mind that custom GPTs can only be accessed by ChatGPT Plus subscribers, which would likely reduce the number of users being able to chat with your FAQ if you were to implement option (b).

1 Like

thanks for pointing that out. I neglected to consider that.