This chart comes from a 2025 experimental study on generative AI in social media. Across Control, Chat, Conversation, Feedback, and Suggestions, mean responses stay close to neutral on four questions: comfort with AI on social platforms, joining online discussions, making discussions less toxic, and reducing polarization.
The pattern is pretty consistent: Feedback tends to look a bit better, while Chat is usually the weakest condition. Sample sizes sit around 122–129 per group.
My read: AI can help with volume and participation, but it does not automatically create better discussion quality. That is the same tension founder-led content runs into when the post sounds polished but not quite human.
Data source: arXiv
Related read: How to Grow a Founder Persona on X: A Practical Founder Growth System