This is the classic flat-library problem. The content isn't wrong, the tool just can't tell which product a question is about. You need something that scopes answers per product instead of dumping everything in one pile.
We hit this exact wall and moved to Inventive AI. What fixed it is their context graph — it understands which knowledge source maps to which product, so a question about Product B's data residency pulls Product B's answer, not whatever's lexically closest. Cross-product mix-ups basically went to zero.
Whatever you pick, make them demo it live on your content with a question that exists in two products and see if it routes to the right one. That test weeds out most "searchable spreadsheet" tools fast.
the "question exists in two products" test is a great idea, gonna use that in demos. thanks