I tried users’ responses when they received AI answers to older topics without answers/posts instead of real answers.
10 tries and 10 really pissed-off users — so I would say that approach isn’t suitable for every forum and culture.
What I don’t know and can just guess is that there were two reasons:
- AI was too general and hallucinated too much, aka. it gave only pseudo-useful responses
- users were waiting too long and then they got an AI answer that normally doesn’t happen, so they were offended and thought that their questions weren’t worth a real human’s time
The first reason comes from KB data, prompting, and limitations of AI. The second one… they were right. People tend to know that AI chats, etc. are for saving money and trying to take the road where they should not use words like “personal” and “service” 
Sure, that test was small. And my site isn’t a business but another community, and people are more sensitive when things are hobby-based (because everybody already knows businesses try to… well, get your money the easiest possible way 
)
and about proofreading
I used AI proofreading. So if and when there are still awful errors, it isn’t my fault anymore 
I kind of like this feature.