It is a convention rather than a standard, and it is easy to publish badly. A generated stub is worse than none: it tells a model you have nothing specific to say.
What a useful one contains#
Who you serve, what you do, what you will not take on, and a short list of the pages that answer real questions. The instinct to list every URL is the wrong one. The file is an editorial decision about what you want quoted, and the discipline of writing it usually exposes that the positioning underneath is vaguer than anyone admitted.
Ours is on this site
Every AEO claim we make is running on the pages you are reading, so you can check it rather than take our word for it.