Why not use a bot to automatically post brand new chatgpt-generated fake comments in between edits with fake stuff? Or automate making edits that insert/delete/replace one word with a random word?

There's many different things to do that could trash the data and make it worthless for training AI. With the right tools, this sort of automation can also be made very difficult to detect (randomize the time interval between edits and the time of day on which they occur [normally distributed during prime time]).

The data is already there.

People will get bored with manipulations and they can just ignore current timespan.

Even currently, creative individuals is 0.01% of users which is easy to spot with simple heuristics especially given such a small percentage and the fact that false positives don't hurt.

Detecting automated / fake content and silently marking it is probably 90% of what coders do there even without the drama. I think they are pretty decent at it.

Even if they couldn't distinguish fake content, there's a good chance neither could its buyers so nobody cares.

I just think it's a bad fight to choose.

Costs more money to develop and run a ChatGPT bot than it is to say

    UPDATE comments WITH SKIPLOCK 
    SET _soft_deleted_time = null
    WHERE _soft_deleted_time > '2023-05-01'

Why not shut down the whole site then? They already have an archive of everything from before then. The question is what you do about future content? If it can no longer be trusted then it’s worthless as a dataset for training.

That is not how the GDPR works. They have to delete all information upon request or termination of the service.