Loading
Because they start customer-facing. The widget goes live on the website before anyone has checked what it says, and the first wrong answer is seen by a customer rather than a colleague.
Internal first, always. Let staff ask it questions for a few weeks — they will find the wrong answers, and finding them costs nothing.
Then agent-assist, where a person checks every draft before it leaves. Only then customer-facing, and only for the topics your articles genuinely cover.
Long enough to see the wrong answers — a few weeks of real use.
Documentation and support, not checkout or pricing.
Yes, per channel and per page, immediately.
Half an hour with your own data usually saves reading three of these. The guides will still be here afterwards.