I participate on several Stack Exchange sites that have scope that's a little more subjective than the norm.  These topics tend to also be more "accessible" to people; you pretty much have to be a programmer to answer questions on Stack Overflow, but anybody who once knew somebody who had a cat might feel qualified to answer questions on Pets.  Ditto topics like the workplace, health, writing, parenting, and lots of others -- lots of people have some *experience* with the topic (or maybe just some speculative theories), but answers work best when they come from a position of relevant *expertise*.  But, to add a wrinkle, sometimes expertise *does* come from experience; it's just that not all experience leads to expertise.

We would like to raise our overall answer quality, which I think means a combination of attracting more users who can contribute high-quality answers and discouraging the less-informed answers.  In this question I want to focus on the latter.

We'd prefer to help the users who are posting the speculative, anecdotal, answers to do better; I'd rather they improve their contributions than go away.  People already leave constructive comments and downvote (not enough of the latter, but some) on individual answers; I'm looking for things we can do at a broader level.  Ideally I'd like to intercept these answers *before they are posted* -- that is, find a way to help users think about what's really needed and then *do that*.

What have other communities done to address this?  I know of one Stack Exchange site that has a flowchart for questions, but I haven't seen anything like that for answers.  Some sites have rules about citing sources, but that isn't a complete solution -- sometimes an experience-based answer *is* appropriate.  (I imagine that DIY would be very different if that weren't permitted.)  What has worked for others?

While this question arises out of Stack Exchange, this isn't a technical question.  I'm really looking for a behavioral solution, and there are other communities besides SE that may have addressed this issue.