Guides

Is the assistant answering from the right material? “Preparing your content” covers what to crawl, what to upload, what to leave out, and how the structure of a page changes what the assistant can find in it. Most wrong answers are created here, long before a model is involved, and most of them can be fixed by rewriting a page rather than changing a setting.

Where should the line be? “Choosing a retrieval threshold, by measurement” explains the single setting that decides whether the assistant answers or declines, and how to find the right value for your own content instead of copying someone else's. It includes the measurements we made, and the ones that did not transfer from one store to another.

Which platform should we choose? “Evaluating AI chat platforms” is the list of questions that separate vendors during a trial, including the ones we would rather you did not ask us. It is written to be useful even if you choose someone else.

Every figure in a guide comes from a measurement we can reproduce, and every recommendation says when it does not apply. Where our own product has a gap relevant to the guide, the guide names it. We would rather you trusted the guide than finished it with the impression that there is only one possible answer.

Store owners setting up the assistant for the first time, developers putting it on a custom site, and agencies deciding how to run it for several clients. Each guide says at the top who it is for and how long it takes to read, so you can tell in a few seconds whether it is the one you need. None of them needs a technical background to follow, and where a step needs a developer, the guide says so.

If there is a question you have had to answer for yourself that deserves a guide, tell us. The most useful guides here started as support conversations.