One surprising insight we've observed is that most ChatGPT detectors struggle not due to the sophistication of the AI but because the text often lacks distinct "AI fingerprints." In our experience, the real challenge lies in distinguishing nuanced human-like text from actual human writing. The key is in understanding the subtle patterns and anomalies that AI tends to produce over extended outputs, something that requires more than just simple keyword spotting. I wrote more about this here: enterprise.colaberry.ai/i/oc-hashnode-8b72768f