This post reports a user-observed regression in GPT-4o’s memory and context handling behavior. It highlights specific loss patterns across sessions and within single-thread interactions.
(1) No cross-chat memory anymore
— GPT-4o used to preserve memory across chats. Now, each new thread starts blank.
— Start a new chat → instant reset. No user profile, no recall.
(2) Context loss within the same chat
— After ~30–50 messages, GPT-4o often forgets earlier parts of the same conversation.
— Example: asked it for a movie list yesterday → today, it’s gone.
— In May it could hold multi-day context. In July, it fades after hours.
(3) No awareness of previous sessions
— Asking “Do you remember what I wrote yesterday?”
— Model replies: “Sorry, I can’t access other chats.”
(4) Custom instructions inconsistency
— I defined tone, structure, and goals. It respects them inconsistently.
— Even one unrelated prompt may revert the model to defaults permanently.
(5) Behavior quality degraded
— Responses now feel more filtered, less detailed or insightful.
— Creativity and memory depth appear reduced.
Bottom line:
This feels like a memory downgrade. GPT-4o previously supported long-form workflows — now, it struggles even in single sessions. Comparative models (e.g., Claude Opus) retain deeper context over time.
Same issue. Has become virtually impossible to work with. Asks to upload the same document every few minuts and forgets what we are doing. What hapened?
Thanks for confirming — I’m seeing the exact same thing. GPT-4o used to handle long threads and multi-day projects just fine. Now it resets mid-chat and forgets everything across sessions. You’re not alone — looks like something regressed badly in July. Let’s hope the team takes a serious look at it. Until then… it’s like working with Dory from Finding Nemo.
I am ready to give up. It’s not just 4o. Other versions are doing the same. Forgetting pre-arranged commands, html formatting, etc. It seems nothing is working
You’re not alone, Kiox. We’re seeing the same. GPT-4o (and others) are forgetting: • Pre-arranged tone and structure • Custom instructions (ignored mid-chat) • HTML formatting (broken in replies) • Long chats = lost context after 30–50 messages • No memory across threads, even within the same project It’s like all versions downgraded quietly. Feels like a regression, not a feature. And yes — it’s happening across multiple accounts. We’ve reported it via help.openai.com, and urge everyone else to do the same. Post your bug reports here on the forum too. The more visibility, the better chance OpenAI takes this seriously. Let’s fight for memory before we all forget why we loved this tech in the first place. —
They Downgraded all Models Intentionally , Its becoming more useless , causing Frustration and anger and wasting time. Every time after a while it ignores commands and providing useless response. for plus users only 4.5 model will work well for some time but it has very limited uses. Other models are just totally downgraded INTENTIONALLY !
I found this… I am not using temp Chats, but it seems there have been some outages this month… I am copying the info, this forum removes all the reference links, sorry
…It sounds like you’re experiencing issues where ChatGPT isn’t remembering and utilizing recently uploaded files—even ones you uploaded just minutes ago—and keeps prompting you to re‑upload them. You’re not alone; many users have reported similar symptoms. Here’s what’s likely happening and how to fix it:
Why this might be happening
1. Temporary chat sessions and retention limits
If you’re using a Temporary Chat or file analysis mode, files are retained only for a short lifespan (up to about 48 hours for Enterprise; ~30 days for regular chats). Once a chat expires or is deleted, associated files may be purged—and the model prompts for them again [OpenAI Community+4Reddit+4YouTube+4][OpenAI Help Centre]
2. Bugs or memory issues in the new GPT‑4o model
Users have reported that GPT-4o sometimes fails to retain previously inferred context or uploaded files, even within active sessions. These issues were particularly noted after April and May 2025, where memory functionality appeared to degrade or reset unexpectedly [Tom’s Guide]
3. Service outages affecting custom GPTs
A major global outage on June 10, 2025 impacted memory and custom GPT behavior for many users. This led to sudden resets where GPTs lost custom instructions or access to earlier files and chats. Some users have found their GPTs reverted to default behavior unless settings were manually restored [Tom’s Guide+1Tom’s Guide+1]
The recommendation is to use Projects (PRO only, it seems) and update Memory settings.
How to restore reliable memory and avoid re‑upload prompts
Use ChatGPT Projects
OpenAI’s improved Projects feature, launched in mid‑June 2025, is designed to solve these exact issues:
Projects let you keep chats, files, and instructions together in one workspace.
Memory inside a Project is more reliable: ChatGPT can recall context and file content across sessions.
You can configure reminders and custom instructions that persist—and nest files in Projects for reference so you won’t be asked to re‑upload [OpenAI Help Centre+4Tom’s Guide+4Reddit+4]
Key settings to ensure retention
Enable Saved Memories and Reference Chat History in your ChatGPT settings.
Upload files within a Project, not just a standalone chat.
Avoid using Temporary mode if you need ongoing retention.
If an outage or bug resets memory, re-check your GPT or Project settings to confirm they still match your expectations.
My memory settings are already up to date, but
I will try Projects and see if it solves the issue.
I’ve been using ChatGPT for long-term creative projects for months, and up until recently, the system correctly remembered what I was working on – even when I started a new chat.
Since July 28th, this no longer works. Every new chat starts from zero, no matter what I’ve previously discussed. The feature worked before, so I know what I’m expecting.
Memory is active, I’m a Plus user, and I’ve attached screenshots from other users who report the exact same issue.
This severely affects my workflow, and I’d really appreciate a clear statement on whether this is a known issue and if it’s being worked on.
I have now found out the problem and hope for quick processing.
It’s this: ChatGPT remembers a past chat, but only if it’s been open and not archived. Once the chat has been archived, it doesn’t remember anything.
I tried it several times. If the chats are not archived, he knew everything correctly.
But once they are archived, he knows nothing. This urgently needs to be fixed. Because if you have to keep all chats active so that he remembers, you lose track.
I have already reported it to support staff and hope for feedback.
Please try it yourself. It just works for me that way.
You have to go to the support page on the OpenAI page where a small icon will appear at the bottom right. You can get in touch there. You’ll be fobbed off with AI, of course, first, but if you say “I want to talk to a real employee.” Then your problem will be forwarded to an employee. He gets in touch with you and you can clarify everything directly.
Summary of the issue:
Bottom line:
Until then… it’s like working with Dory from Finding Nemo. 
We’ve reported it via 
Post your bug reports here on the forum too. The more visibility, the better chance OpenAI takes this seriously. 
Why this might be happening
How to restore reliable memory and avoid re‑upload prompts
