Guide
Using web search with AI without losing track of sources
A language model knows what it was trained on. Anything newer, and anything specific enough that it was never well covered, has to come from a search. The question is less whether to use web search and more how to keep its results organised.
When to turn search on
Search is worth using when the answer depends on something that changes: prices, product features, release notes, regulations, recent events. It is also the right call when you need a source you can point to, because an answer with a link can be checked and an answer from memory cannot.
It is not always better. For reasoning over material you have already provided, such as your own documents, search adds noise. In Bonscape search is a toggle on the prompt, so you decide each time.
Why search results belong in branches
Search results are bulky. A few queries can add a lot of text to a conversation, and in a single thread all of it stays in the history for every later question. Putting each research question in its own branch keeps the findings for one topic away from the others, and keeps the main line short.
It also makes the sources easier to audit. When each branch answers one question, the links under it are the evidence for that question and nothing else.
A habit worth building
Treat a cited answer as a starting point. Open the sources for any claim you plan to rely on, highlight the passages that matter so they are saved to your list, and only then merge the branches into a summary. The merged answer is as good as the branches feeding it.
If you are doing this for vendors or competitors, the competitive research workflow shows the full pattern.