I really appreciate these concrete tips. I've recently been reading about the design of intelligent tutor systems, which include explanations as a core component. I've been struck by how "localized" models for good explanations (and hints, and feedback, etc.) are: we don't seem to have good general models, because educational contexts vary so wildly.

The literature does seem to agree with your second two tips, but I've found it frustrating how little guidance there is like your first tip, on word choice and explanation structure. Once again, I think that's because it feels so specific: a middle-school classroom, a chatbot tutor, and a nursing home all call for different methods.

If you're curious, this seems to be a well-respected and still-influential taxonomy in tutor system design: Kurt VanLehn. The Behavior of Tutoring Systems. Int. J. Artif. Intell. Ed. 16, 3 (August 2006), 227–265.

Thanks for writing this!