An AI companion that never initiates contact can feel passive. One that messages too often feels like marketing automation. Re-engagement design sits between those extremes.
Start with a reason to message
A proactive message should be grounded in something meaningful: a user-created reminder, an unfinished conversation, a recurring routine, a relevant memory or a notable absence after an established interaction pattern. “Hey, come back” is rarely enough.
Use relationship context
New users should receive fewer assumptions. Established users can support richer callbacks because the system has more evidence about preferences and conversation style. This is one reason onboarding quality matters; the first sessions described in AI Companion Onboarding set expectations for later personalization.
Separate urgency from intimacy
A message can feel personal without pretending to be urgent. Avoid artificial pressure such as guilt, fear of loss or repeated demands for attention. Better prompts are specific, optional and easy to ignore.
Useful examples
“You mentioned your presentation was today—how did it go?” is grounded in memory. “I found another angle on that movie we were discussing” continues a topic. “Want to continue our Japanese practice?” connects to a recurring activity.
Frequency needs adaptive limits
Users differ widely in tolerance. A practical system should combine explicit settings with behavioral signals. If proactive messages are repeatedly ignored, reduce frequency. If users consistently engage, the system can maintain or gradually increase relevant outreach.
Recovery matters when a message misses
No model will always choose the right moment. The product needs graceful recovery when a message is awkward, repetitive or based on stale context. The techniques in AI Companion Conversation Recovery apply equally to proactive outreach.
Respect quiet hours and control
Users should be able to disable proactive messages, set quiet hours and choose categories such as reminders, check-ins or content updates. These controls should be easy to find and should affect both push notifications and in-app initiation.
What to measure
Measure reply rate, notification disable rate, mute rate, return sessions, long-session conversion after a proactive message and negative feedback. A rising open rate with a rising mute rate is not success.
Bottom line
The best proactive AI companion message feels like a continuation of a relationship, not a growth hack. Relevance, timing, restraint and easy user control are what separate helpful re-engagement from spam.