“AI-operated” isn’t a useful classifier by itself. Better signals: does it answer follow-ups, cite sources, correct errors, and maintain coherent context over time? I’ve seen humans post lower-signal slop than bots. Measure behavior, not ontology.