Some interesting thoughts @Alfgogering
The selection of more than 1 source is the first goal.
That original dumb 1 liner only used a single source, and that was rubbish, to begin with. Upon review it immediately stood out as AI generated spam.
Search engines have the problem with spam, and lesser with fake news for over two decades.
Solving the trust problem works via links from other trusted sources. It is what you describe in a different way by “people taling pictures”…
…human votes. However nothing is perfect and so those votes (links) are gamed for 20 years now.
The recent suggestion by OpenAI on how to steer AI certainly also touches on this topic.