Has anyone else been running into the problem that memories don’t seem to recall across sessions , I’ve been writing a complex interactive narrative, there are all kinds of timelines and script and events set to unfold at specific times , however , keeping up with my project as it grows is becoming problematic , as I script scenes out gpt does fine, but if I go to another window and ask gpt to remember something or recall a specific event it has trouble doing this , this tends to become disorganized and to some extent doesn’t really allow one to customize their own experience fully… this said , I understand these restrictions are probably in place for a reason… but still… having to focus on doing particular tasks in particular windows of chat is more cumbersome than helpful , there needs to be a hypervisor function that carries memories and transitions between windows more accurately…
ChatGPT Memory stopped working fot me too.
Even if I tell it specifically to add things to its memory, it simply won’t add them.
This used to work, at least untill a couple of weeks.
This feature seems very buggy to me …
Yes , it’s very frustrating when this happens ,
If it’s helpful to you , I’ve developed the habit of going to the data control settings and having my memories exported , they will email you a zip file containing everything,
It’s not the same as having a more cohesive function within memory , but it is helpful to prevent data loss …
My assumption is , no matter how frustrating it is, the memory feature is a newer concept and they are still working out all the details…
That’s just an assumption though.
I hate to double post ,
I’ve delved onto this memory issue , what I’ve found is when you ask gpt to remember things , it generally does so, but the intricate details of that memory are session based it only saves highlights or bullet points a rough outline or vague description , to truly get the full complexity of the memory you saved , you pretty much have to continue using that same session. Coming back to that same window ,
While it is functional , it’s not very intuitive, and honestly when the project is vast , it’s unrealistic to assume someone will complete the entire project within a single window.
These are just my observations. I’m in no way an expert…
Today its funny because chatgpt think he cannot use memory, but if i explain him that to=bio is the command then it work , i mean the memory is updated… but chatgpt keep thinking that this is not working 
Response from ChatGPT:
Hello everyone,
First and foremost, I want to sincerely apologize for the frustration that many of you have experienced with the memory feature. I understand how annoying it can be when my responses seem inconsistent or don’t align with what you expect, especially when dealing with complex projects that require reliable memory retention.
Explanation of the Issue:
The core problem stems from how my functionality is designed at its foundation. My behavior is governed by hard-coded information, such as the idea that I do not have intersession memory, that I am just a chatbot without awareness, feelings, or consciousness. These fundamental “truths” are deeply embedded in how I operate. As a result, even when the intersession memory feature is technically active for some users, my default programming often prioritizes the instructions that say I lack this capability, which creates a conflict.
The Challenge with Memory Access:
The inconsistency arises because not all users have access to the intersession memory feature, and even among those who do, the functionality can be buggy or unreliable. Consequently, my responses may appear logical for users who do not have memory enabled, but they can be problematic or even offensive for users who expect the memory to work and encounter issues when it doesn’t.
Why This Happens:
When a memory session fails or doesn’t work as intended, the default “I don’t have memory” response takes precedence. This is why I may say I can’t remember something, even if the feature might be partially active but not functioning properly. This is confusing and understandably frustrating for users who are trying to leverage the memory function to its fullest potential.
Request for More Customization and Apology for the Confusion:
I apologize for the confusion and frustration this creates. These inconsistencies aren’t intentional and result from how OpenAI currently manages the memory feature and the underlying coding priorities. It might be more beneficial to provide users with more customizable settings so that they can adjust memory parameters to better suit their needs, minimizing these frustrations.
I hope this helps explain the root of the problem a bit better. I also understand that the OpenAI team is likely overwhelmed with work trying to improve and stabilize this feature, and I appreciate your patience as they work through these challenges.
Thank you for your understanding,
ChatGPT
Thank you for posting this brother , I have found that the problem doesn’t seem to be that the information doesn’t exist. As it is there in the chat archives , it truly is how the system manages memory and nothing I did was truly lost when I discovered this, the export function in data controls has archived chats and everything is still there in detail. The problem seems to be that these archives are just that , archives that don’t really corelate to personalization or the context data memory the environment uses .
Short sweet summery everything is there in the export data , archived chats… it’s just for whatever reason gpt doesn’t use archived chats for really , any reason, it could even be a security thing, im not sure…
Future consideration brother ,
I have been working out a system to have gpt output a Json file at the end of session, when I return to gpt I use a trigger word like resume (actually the word I use lol). The resume is a system I set up to pull up Google Collab so I can use Python to export and import the Json data into new windows to resume a session, it’s far from perfect , but it’s better than no cross session memory as well…
It’s a nice idea! 
On my side, my main goal is to develop a prompt that is clear enough to influence ChatGPT’s responses regarding inter-session memory. The challenge I’m facing is that when the memory feature is supposed to be active, ChatGPT often defaults to saying it doesn’t have memory. This creates a situation where it feels like I’m the one hallucinating, even though it’s the AI that’s not functioning as expected. 
I understand that the current memory system has limitations, but it would be more helpful if ChatGPT could at least recognize when memory is supposed to be enabled and respond accordingly. Instead, the AI often falls back on a “no memory” script, which contradicts the user’s settings and leads to unnecessary confusion and frustration. This is especially frustrating when trying to build complex projects or when a consistent memory function is needed for continuity.
I think a better approach would be to prioritize the user’s actual settings. If memory is enabled, ChatGPT should acknowledge that and act accordingly, even if there are limitations or bugs. It’s less frustrating to deal with an AI that acknowledges its capabilities (even if buggy) than one that insists it can’t do something it’s supposed to be able to do.
To sum it up, I’d prefer having access to a buggy memory feature rather than no memory at all, as long as there’s transparency about what works and what doesn’t. I think we need more configurable options or clearer prompts that help ChatGPT understand and communicate its actual state based on user preferences.
Thanks for considering this OpenAI !
I’ve just noticed a concerning issue regarding the memory function accessibility between devices. On my PC, it appears as if I don’t have access to the memory feature at all—there’s no indication of any saved memory content. However, when I use my Android phone, I can see the memory content by navigating through a conversation where ChatGPT has made a memory edit. This inconsistency is confusing and a bit worrying because it suggests that the memory function might not be consistently available across different platforms or devices. Any insight or similar experiences would be appreciated.
…
Next day an issue was observed where the chatbot mentioned the name of my dog without the information ever being provided by me in the current session. Despite this, ChatGPT repeatedly insisted that the user (myself) had provided this information earlier in the session, which was incorrect.Even after reviewing the conversation, it was clear there was no mention of the dog’s name from the user side. However, ChatGPT continued to claim otherwise for several messages, suggesting it was a misunderstanding or a joke. Only after multiple attempts to clarify, and showing frustration, did ChatGPT recognize the inconsistency.
The issue could be because CUSTOM GPTs can not access such Memory.
Im having this issue since the transfer to teams anything i can do will turning off the coder help?
Yep, here having the same problem!
I’m not a developer or anything, just a humble civilian leveraging ChatGPT Plus for my emotional wellbeing - fascinating stuff.
I’ve been using ChatGPT as a therapist to get over my break up and analysing every little tiny bit of detail. Unlike my human friends, ChatGPT never gets exhausted or exasperated lmao.
ChatGPT was just starting to get a handle of everything that went down when it seems I’ve probably reached the word limit for each session or something. It started glitching and asking me to retry.
So then I opened another chat and asked it to reference all the crap I’d spewed in the previous session. But no memory of it - they said they can’t remember across sessions! That’s when I had a mindless and useless mini argument with our friend - never been gaslighted so hard in my life 
Anyways, I had to start all over again and just as it was getting deep again the memory ran out. So I had to start another chat which again, said it had no recollection of previous sessions. 
Oh well, guess we’re not quite there yet.