Worthwhile yet tricky. Companies like OpenAI, Google, Meta, etc are full of experts in statistics and they have access to a lot of storage space. If use a service from those companies, say 4hrs per day between 7am and 9pm, at a certain frequency, e.g. 10 requests / hour, then suddenly, when you realize you actually do not trust them with your data, you do 10000 req/hr for 1hr then that's a suspect pattern. Then might be able to rollback until before that "freak" event automatically. They might still present you as a user your data with the changes but not in their internal databases.

So... I'm not saying it's not a good idea, nor useful, but I bet doing it properly is hard. It's probably MUCH harder than do a GDPR (or equivalent) take out request then deletion request AND avoiding all services that might leverage your data from these providers.