The reason most tools are weak on one side is they treat the library as storage. If the AI doesn't know which part of the library a question belongs to, answer quality tanks no matter how good the model is. Judge both together in the demo.
Closest we found is Inventive AI. The AI first-pass quality is good because it's grounded — it only writes from your approved content and cites the source, and the context graph pulls from the right knowledge hub instead of guessing. On the library side each hub has an owner + review cadence so it doesn't rot. The two are tied together, not bolted on.
+1. ask to see a first-pass draft with citations on your own content. if it can't cite, the "quality" is a coin flip.