I don’t think this is a good long term solution. LLMs can do easy language substitutions and you can even force them to add errors. So relying on that alone won’t work as people intentionally make things look more “human.”