ChatGPT can carry facts between conversations, which removes the re-briefing tax and quietly introduces a confidentiality question.
Every rep who uses ChatGPT seriously has re-explained their product, territory and ICP hundreds of times. Memory removes that tax. It also means your commercial context now persists in a vendor system between sessions, which is a different risk profile from a throwaway chat.
Memory means ChatGPT retains specific facts you tell it — your product, your segment, how you like output formatted — and applies them in later conversations without being reminded.
OpenAI shipped this in stages, and the stages matter. The 2024 release stored discrete facts you or the model chose to save. The April 2025 update went further: ChatGPT began referencing your past conversations generally, not only saved snippets. That is a meaningful change in how much of your history is in play on any given answer.
The short version: it remembers you between sessions, so you stop re-explaining your job — and it now draws on far more of your history than the original version did.
OpenAI announced memory and associated user controls for ChatGPT on 13 February 2024, alongside temporary chat for sessions that leave no trace, memory for GPTs, and support for Team and Enterprise customers.
Initial release: 13 February 2024, a limited rollout rather than general availability.
Major update: 10 April 2025 — memory became comprehensive, referencing past conversations rather than only saved memories.
Free tier: 3 June 2025 — a lightweight version of memory improvements began reaching free users.
Controls: Users can turn off saved memories or chat-history referencing in settings; temporary chat bypasses memory entirely.
Price: Not stated — a plan capability, not a separate charge.
What it doesn't do: memory is not a CRM and is not shared across a team — it is per-user, and it leaves with the rep. It also does not distinguish between confidential and ordinary context on your behalf; that judgement stays with the person typing.
SDR — High-volume, high-repetition work benefits most. Teaching it your ICP and tone once, then drafting against that standing context, compounds across hundreds of touches.
AE — Useful but treat it carefully. Memory blends context across accounts, and a model that recalls one client's commercial detail while drafting for another is a confidentiality problem you created without meaning to.
Manager — Memory makes output personal to each rep, which cuts against consistency. If you want a shared standard, a documented shared instruction set beats each rep's privately accumulated memory.
RevOps — The governance item is retention and controls, not capability. Memory is user-toggleable and admin-manageable — decide the policy before reps have six months of commercial context stored under personal settings.
If you are going to give a model standing context about your accounts, make the context worth having. Build the account brief deliberately rather than letting it accumulate by accident.
Two minutes, once a week. What changed in AI, and what to run because of it.