Watched words sounds like it might be a good idea, although of course it’s a blunt instrument.
I may choose to document a process of periodic checking of the use of user-to-user messages. But that works for me because my users barely use the facility.
/admin/reports/user_to_user_private_messages
If there was an uptick in these messages it might indicate elevated risk, and I’d need a response to that. I’m not sure what there is between the extremes of
- users will let admins know if messages are problematic
- admins will read some (or all) of the messages to check they are not problematic
Personally I don’t want automated analysis which would have false negatives as well as false positives.
I would (will) concentrate on simple statements which position my forum as very low risk on all axes. Proportionate checking is therefore to do nothing, or if something, then something very minimal and easy. Relying on flags is simple and easy.