In chat,ChatGPT will save a memory with the “Memory updated” message.
When you hover over the message, it will correctly display the memory it has created. If you click on the message to open the memory, there’s nothing there.
In addition to that, if you go to the actual memory menu (Account → Settings → Personalization → Manage Memory, it will show that every time it tried to record a memory, it is just duplicating the same one over and over.
So to sum it up, it is creating “a memory” record every time but instead of saving the right information, it’s saving some random previous one duplicated.
Issue has been ongoing for 3 days now.
Here is what is being added, dredging some old part of its memory where I mentioned playing Magic the Gathering in highschool. What is really odd is that it is hallucinating on that, changing the content every time ever so slightly, and inventing details like “Arena” or “AI opponents” which are not part of the original memory or ever discussed.
+1 to this bug report. Same issue of confabulated memories when ChatGPT tries to edit or store new memories. Started for me since yesterday and confirmed to still happen today.
I have an extensive memory storage and the model makes outlandish, detailed and confabulated memories. I was worried at first that they may have crossed over from another account, but I think it’s just confabulating from existing memories (in sometimes very disturbing ways). I’ve turned off memory for now and it still seems to try updating memories.
[Edit: The issue with “Memory off” being ignored seemed to only happen with the immediate conversation after I toggled the option in the mobile app. A new conversation is adhering to memory off. I would like to use memory again but will avoid until this is fixed.]
[Edit Feb. 4th: Seems to be fixed for me now. I can edit and update memories with ChatGPT’s help. I noticed that a new memory that ChatGPT made ended in “…” instead of going on very long. So, maybe the issue was caused by very long memories (at least for me). I’m back to normal use now and don’t see any more confabulations in my memory storage.]
I’ve kept testing and it keeps just re-hashing the same “stuck” memory varying on the terms. I am wondering if we are not hitting a size issue because a few days ago, the memory was supposedly 94% full and now the warning is gone.
Even worse, in my case, those anomalous entries are not being seen by the conversation:
Say the weird memory entries are #112, #113. In the conversation where the storage request is made, if you ask ChatGPT, what’s entry #112:
I have encountered this issue as well and observed that the last correctly stored memory was created on Jan 25th. I have been doing some testing of the scope of this issue and have used ChatGPT to help me synthesize the findings and create the following report.
Summary of Memory Duplication & Storage Failure Issue
We have observed a consistent and reproducible issue with ChatGPT’s memory function, where attempts to store new information result in the unintended duplication of existing memory entries, while the new data fails to save. Instead of updating or replacing stored knowledge, the system reinforces old entries, sometimes creating multiple copies of the same stored memory.
1. Steps to Reproduce
A new piece of information is explicitly stored using a direct command (e.g., “Store in memory: [specific knowledge]”).
The user checks visual memory to confirm whether the update was saved.
Instead of the new information appearing, an existing stored memory is duplicated (sometimes multiple times), reinforcing an older entry rather than saving the new one.
Recall testing in a new session confirms that the newly stored knowledge is completely absent, while the duplicated memory persists.
2. Observed Behavior
New memory updates are being discarded.
Previously stored memory entries are being duplicated instead of updated or replaced.
The more new information is attempted to be stored, the more old entries multiply. (For example, a previously stored memory started with 4 copies but increased to 11 after multiple new storage attempts.)
Deleting an existing memory does not guarantee successful storage of new information—the old memory may still return.
3. Expected Behavior
New memory updates should persist and replace outdated knowledge when necessary.
Existing memories should not duplicate uncontrollably when a new memory is stored.
Deleted memories should not return or multiply upon further interactions.
4. Additional Context
Troubleshooting Steps Taken:
Attempted different phrasing and structuring of the memory update.
Tried storing smaller portions of information incrementally instead of a full block.
Used direct user commands to store information rather than indirect updates.
Deleted an existing memory to see if it allowed a new memory to be stored.
Tested recall in a fresh session to confirm whether memory loss was occurring.
Findings:
Memory system seems to be reinforcing certain prioritized memories (e.g., character-related ones) while rejecting others (e.g., abstract concepts). == NOTE FROM USER == I am uncertain if this is correct, but I am leaving it in because it might be.
New memory persistence is entirely failing, even with repeated attempts.
The memory system is stuck in a loop of reinforcing past stored data instead of evolving.
Next Steps
This issue suggests a bug or unintended behavior in the memory prioritization system, potentially related to:
A limit on new stored memories, causing them to be blocked or ignored.
A prioritization error where only certain types of memories persist while others are discarded.
A systemic issue with memory replacement, where old memories reinforce rather than evolve.
Since this behavior fundamentally disrupts how memory is expected to function, further investigation is necessary. If any diagnostic tools or additional tests would be helpful, please advise.
Looks like there might be a similar thread with more traction, /memory-not-updating-in-chatgpt-4o-issue-persists-across-conversations/1105721. Found this after noticing new memories seemingly not committing even after I explicitly tell it to (and it notes that it supposedly did).
And me. Although mine isn’t storing anything. It’s saying updating memory and saved but when I look in the memory it’s not there — and on another chat it (obviously) can’t recall the information I asked it to store in memory. This has been going on for 4 days for me.
I couldn’t identify any such message in my memory unfortunately. I have however, after seeing the bug, noticed 4o trying to save memories in the format describe (e.g. multiple line breaks).
I’ve been experiencing similar memory storage issues and have conducted several tests to better understand the behavior. My findings align with what has been described in this thread, particularly regarding failed memory updates, unintended reinforcement of old memories, and storage limitations.
Key Observations
New memories are not always stored separately
When attempting to store a new piece of information, it does not appear as a separate memory.
Instead, an existing related memory is modified to include the new information.
Previously stored memories are sometimes edited instead of being replaced
Expanding on an existing topic results in the original memory being updated rather than a new entry being created.
Deleting memories does not guarantee new ones will store
Removing a memory and then attempting to store a new one does not always work.
Instead, an older memory is sometimes reinforced rather than replaced.
Potential prioritization issue
Some types of information seem to be prioritized for storage while others fail to persist.
Possible Explanations
Storage quota or rate-limiting: There may be a limit on how frequently new memories can be added.
Memory reinforcement instead of replacement: The system may be designed to reinforce existing knowledge rather than create new entries.
A bug in memory prioritization: Some types of data may be more likely to persist while others are ignored.
Further Testing Suggestions
Try storing unrelated information to see if it saves separately.
Experiment with different phrasing to test whether certain structures are more likely to be stored.
Observe if older memories duplicate when adding new ones.
This behavior prevents users from reliably updating stored knowledge, making it difficult to manage evolving information. If OpenAI could provide insights into how memory updates are processed—or confirm whether this is intentional—it would be extremely helpful.
I have the same issue, noticed it yesterday. (Before that, my memory has been at 100% for weeks). It just keeps saving the entry from my information about myself, and just now saved someone else’s cat instead of my request to use less bold font. Weird.
In my case, it seems like ChatGPT gets stuck on an older memory (which was about me playing MTG in high school) and just keeps rephrasing it.
Before that it was doing the same with another memory entry related to medication until I deleted the entry, just repeating that medication one changing a few words at a time.
Unfortunately, according to other users, clearing the memory doesn’t fix the issue at the moment
Adding to this thread. It looks like the previous ticket/thread was closed due to “discussion is about the ChatGPT app, not OpenAI developer platform”. I cleared my entire memory and painstakingly re-entered everything I could and it worked again for about 5 days until it stopped remembering things again. If this is not where we need to be reporting this issue, where then?
It gets worse. From the desktop app, the long term looks completely empty. ChatGPT keeps “saving” details from the story we’re working on, details he had in his long term for months. They go to nowhere.
The web page and the Android app show memory being filled, with most later entries repeating the same “Online name is Dani” (my “about me” info) and “Favourite color is blue” (hallucination). Also assigned me a second non-existent cat. X_X
The issue is getting worse and more concerning as now instead of remixing existing elements (MTG, which was discussed with ChatGPT before and committed to memory), ChatGPT is saving random knowledge that … we have never talked about:
Experiencing the same issue. Had no problems for months but noticed the memories weren’t being saved properly about a week ago. Now I can’t get it to keep more than one memory at a time. Been trying to resolve the issue for the last hour to no avail. Glad I’m not the only one experiencing it but really hope it gets solved soon.
