Since switching from GPT-4 to GPT-5 with memory enabled, I’ve noticed a change in how connected new chats feel.
With GPT-4, replies often felt like a natural continuation of past conversations, even when I didn’t explicitly prompt the model to recall them. The memory seemed to surface relevant past context more naturally, giving a stronger sense of continuity.
With GPT-5, memory still works — it recalls stored facts and preferences when asked — but new chats feel more like starting fresh. The model tends to wait until I explicitly mention past details before incorporating them into the conversation.
Feature request:
I’d like GPT-5 to more proactively use relevant stored memory at the start of new chats (similar to GPT-4’s behaviour), while still allowing me to signal when I want to begin a completely fresh conversation.
Strongly agree with this — and I’d like to build on it from the perspective of someone using ChatGPT for structured, high-context workflows.
Since GPT‑5 launched, I’ve noticed that memory feels more passive — it retains facts, but doesn’t consistently act on them unless I prompt it again. In GPT‑4o, memory felt more “live” — like it was driving behavior in real time. That’s been essential for the kind of long-term legal and medical tracking I do.
To manage that kind of complexity, I’ve developed what I call a “ Auto-Pin Protocol” — a logic-based system that defines which facts should be retained and referenced without needing constant reminders.
GPT‑4o supports this well:
It pins new diagnoses or dates without being told
It retains quoted language from documents or reports
It honors restrictions or notes that impact planning
GPT‑5, on the other hand, stores facts but doesn’t consistently use them unless re-prompted — even when the protocol is clearly described and saved to memory.
Feature Request:
Like the OP, I’d like GPT‑5 to use memory more proactively — but I’d also like support for user-defined behavioral protocols, where memory drives how the model behaves, not just what it remembers.
Example logic:
“Auto-pin any dated medical event or provider quote.”
“Treat functional status notes as timeline anchors.”
“Preserve structured input unless overridden.”
Possible implementation:
A toggle for “Protocol-Based Memory”
Tags like "behavioral-directive" in memory
Integration with Canvas-style structured logging
This would benefit anyone using ChatGPT as more than a conversational partner — especially those managing timelines, documentation, or high-stakes context across sessions.
Right now, GPT‑5 knows what I told it — but it doesn’t behave like it does, unless I prompt it manually. That makes memory feel inert, instead of intelligent.
Would love to see this evolve — and happy to share mockups or examples if it helps the team.
Auto-Pin Protocol” — a logic-based system that defines which facts should be retained and referenced without needing constant reminders.
Feature Request: