This post might get me tared and feathered, but I actually think automatic changeset comments by default could be useful.
Pretty much all examples of good comments, including those on the wiki page, have the following format (using some pseudo-syntax):
(added|updated|removed) (<object type>|<name>) [in <place>] [based on <source>]
All of this can actually be automatically generated; the type of change and object type (or name) based on the contents of the change, the place can be calculated based on admin boundaries or place= and source taken from the existing source= changeset tag.
I think that for an overwhelming majority of good changesets (that is, they are delimited in some way, i.e. only changing buildings in one town or updating speed limits in a country or similar) a comment like the above could be generated and would be as good or better (if the mapper themselves is bad at writing comments).
That has its limits though; and a good anti-example are the comments generated by some unnamed apps that look like “Added 3 shops, 2 playgrounds, 9 trees, 1 dog, 42 things, 98 addresses, 19 sandwiches” and so on. However, those cases might be possible to improve, the above could for example be rewritten as “Added shops and amenities in Stockholm based on walking around” or similar.
Another interesting case is StreetComplete, those changeset comments by themselves are alright, however since it by default splits it by quest type you can end up with a large number of small changesets, where it might have been more clear to have one with a comment such as “Updated details about streets in Stockholm”. Though that’s of course another discussion altogether.
Then again, for comments that can be automatically generated they could just as well be automatically generated when requested for reading, which would have the added benefit of adjusting the generation algorithm according to other aspects (if showing a list of changesets filtered by country we don’t need to show “in ”, since that’s sort of already expected).
So maybe another approach could be to discourage writing changeset comments completely, unless they add value that cannot be generated?
I actually believe that that would improve the average quality of changeset comments, though it would require some development work on the OSM website and other places displaying changeset comments. It might be interesting to try to generate a few days worth of changeset comments and compare them side-by-side with the comments provided by the mappers.