Why does this happen? My working hypothesis: data voids.
Chatbots pick the pages closest to the wording of a question. For a loaded question, that is often a page written in the same framing.
The fix: official sources should answer in the words people actually type.
