In a forum, it is the actual people who post who are responsible for sharing the recommendation.
In a chatbot, it is the owner (e.g. OpenAI).
But in neither case are they responsible for a random person who takes the recommendation to heart, who could have applied judgement and critical thinking. They had autonomy and chose not to use their brain.
It'd be different if one was signing up to an OpenAI Drug Advice Product, which advertised itself as an authority on drug advice. I think in this case the expectation is set differently up front, with a "ChatGPT can make mistakes" footer on every chat.
If I keep telling you I suck at math while getting smarter every few months, eventually you're just going to introduce me as the friend who is too unconfident but is super smart at math. For many people LLMs are smarter than any friend they know, especially at K-12 level.
You can make the warning more shrill but it'll only worsen this dynamic and be interpreted as routine corporate language. If you don't want people to listen to your math / medical / legal advice, then you've got to stop giving decent advice. You have to cut the incentive off at the roots.
This effect may force companies to simply ban chatbots from certain conversation.
I don't think that it's a good policy to forcibly muzzle their drug opinions just because of their good arithmetic skills. Absent professional licensing standards, the burden is on the listener to decide where a resource is strong and where it is weak.