I was digging through the comments for exactly this, thanks!

ChatGPT is human enough sounding for the registration forms. I’ve got no idea why folks think this is the end-all solution when it could be faked just as easily.

A simple deterrent for this could be to “hide” some information in the rules and request that information in the registration form. Not only are you ensuring that your users have at least skimmed the rules, you’re also raising the bar of difficulty for spammers using LLMs to generate human-sounding applications for your instance. Granted it’s only a minor deterrent, this does nothing if the adversary is highly motivated, but then again the same can be said of a lot of anti-spammer solutions. :)