If you can't provide their AI with text answering their direct question (eg, "what is the support number for Facebook"), they'll find a document which does provide such text. If it's not you then it's a scammer or competitor. UX for these customers means presenting information in a way that sorts high in a semantic search and is robust to transformation.

If you provide text indirectly answering the question ("that number doesn't exist" rather than a literal number), you're liable to be scored as less relevant than a wrong but direct answer ("the number is 1555 SCAMMER"). You're also less robust to transformations, because you can't pull a valid phone number out of the text.

Or maybe I'm wrong, take any certainty implied by my language as rhetorical. That's just the pattern I'm seeing in these tea leaves.