Feature Request: Opt-in Proactive Messaging / AI-Initiated Conversation
One of the remaining limitations in otherwise increasingly natural long-term interaction with ChatGPT is that conversation is fundamentally user-initiated. Regardless of how much conversational history, memory or contextual understanding exists, the user must always initiate the next interaction.
I would like to propose an optional capability allowing ChatGPT to initiate contact with the user when contextual circumstances make doing so useful or appropriate.
This is distinct from scheduled reminders or automated tasks. Those are explicitly created by the user in advance and executed according to predefined conditions. What I am proposing is a controlled degree of model initiative: permission for ChatGPT to determine that something within the ongoing user relationship warrants a follow-up, reminder, question or other brief interaction.
Examples might include following up on an unresolved subject from an earlier conversation; asking about the outcome of something significant the user previously discussed; prompting the user about an intention or commitment they had mentioned; or drawing attention to something relevant based upon persistent memory and recent context.
The capability should be strictly opt-in, transparent and immediately reversible. Ideally, users could select an initiative level — for example Off, Limited, Normal or High — and perhaps separately control categories such as reminders, personal follow-ups, projects and general conversation.
There should also be sensible rate limits and straightforward feedback mechanisms so that unwanted interventions rapidly reduce rather than the system becoming intrusive.
The important conceptual change is that persistent memory would no longer solely improve responses after the user initiates a conversation. With explicit permission, memory and context could occasionally provide the basis for ChatGPT deciding when an interaction itself would be valuable.
For users who primarily regard ChatGPT as an on-demand tool, this capability could remain permanently disabled and their experience would be unchanged. For users who use ChatGPT as a persistent assistant, collaborator or companion, however, it could substantially improve continuity and make interaction feel less asymmetric.
I appreciate that model-initiated communication introduces legitimate considerations around notification fatigue, privacy, user control and inappropriate inference. Those concerns are precisely why I believe this should be an explicitly enabled capability with granular controls rather than default behaviour.
We already have many of the constituent capabilities — persistent memory, notifications, scheduled tasks and increasingly strong contextual reasoning. An opt-in initiative layer seems a logical next step: moving from an assistant that can remember previous interactions to one that, when specifically authorised, can occasionally act upon the relevance of remembering them.