Perhaps the answers were indeed accurate, or you were able to verify the output to ensure it was true/reliable/safe.
My experience of using AI, and becoming warier of their chaotic nature day after day, show me that they are way less trustable I previously thought. And every single day I carefully use AI to get information or answers, I notice mistakes, unreliable or simply false information. The more prompts you send (the longer the conversation is), the more it happens. But even single prompts often returns untrustworthy info.
Of course, it also heavily depends on what I’m looking for. I agree that ask.discourse or general AI can return very accurate info about Discourse, other software, command-line tools, etc.
I ask it to return a FFmpeg command, it’ll do what I ask most of the time.
I ask a developer AI agent to make a full Discourse plugin, I’ll check security features twice more than once!
I ask for accurate information about animals, biology stuff, diseases, medicine, whatever has information scattered among many websites and research papers, it’ll very quickly hallucinate information, quotes, will miss-source, or even lie, if I can use this word. To the point I can absolutely not rely on the output. The answers should be considered nothing more than hints, pointers, that we can use to look by ourself for actual data.
Now, I’ve, like many, answered “try asking ask.discourse.com” to some users here on meta. 
. What can I say! That robot just works[1]! 
well, most of the time
↩︎