The issue with any attempt to identify generated text is that anything generated could have been written by a human. There’s also the issue with model diversity. Even if a specific model has common patterns that can be identified, there are hundreds of other models that will have different patterns.