Does talking to AI longer make the experience better, or does it start quietly working against your wellbeing?

This explores whether more time with conversational AI, in one sitting or over months, changes how much people enjoy it and how they fare psychologically and socially. The corpus has no study that tracks enjoyment and wellbeing side by side as conversation length changes. What it does have are pieces that together point to a pattern: the things that make AI conversation feel good and the things that predict how people fare tend to separate as time goes on.

Start with enjoyment over repeated use. Longitudinal work with the chatbot Mitsuku found that the social pull that draws people into a relationship with a bot fades in a predictable way once the novelty wears off Do chatbot relationships lose their appeal as novelty wears off?. The practical point is about evidence: most findings that people 'like talking to AI' come from single sessions, and those results can't be stretched to explain what happens over weeks. Early delight is partly a first-encounter effect.

Wellbeing follows a different path, and the mechanism is the surprise. A year-long study of Character.AI users found that sustained engagement with AI companions predicted lower wellbeing. But the link ran mostly through something outside the conversation: heavy users had less face-to-face social contact Does sustained engagement with AI companions harm well-being?. So time spent with AI doesn't simply wear people down. It can quietly take the place of the human interaction that was supporting them. That reframes the duration question. What matters is what the hours with AI are replacing.

The therapeutic chatbot literature shows why enjoyment is a poor stand-in for wellbeing. Patients report real emotional bonds with therapy bots, but those bond scores run independently of clinical safety. The same warm, soothing system can reinforce unhealthy thinking patterns while feeling supportive Do therapeutic chatbot bond scores hide deeper safety problems?. Within a single conversation, design choices can deepen engagement fast. Context-aware pacing, meaning reply timing that adjusts to the conversation rather than staying fixed, made agents feel more attentive and led people to disclose more Can pacing strategies make conversational agents feel more human?. And the lack of a human judge makes intimate disclosure easier How do people decide what to share with AI systems?. Those same features are what make long, absorbing sessions possible, whether or not they're good for the user.

The takeaway: an AI conversation that feels better and lasts longer isn't necessarily healthier, and the corpus suggests the two can drift in opposite directions. Enjoyment fades with familiarity, while the wellbeing cost builds slowly through what time with AI crowds out. If you want direct evidence on how session length affects outcomes, such as a dose-response study, this collection doesn't yet hold it.