It seems to me that over time, some of the GPTs I set up for the students started to give poor answers, and failed to refer to the correct files. I reuploaded a file that should have been referenced, and it fixed the issue. I am wondering if GPTs need reindexing once in a while, maybe after model updates or something?
Has anyone else experienced performance degradation over time even though instructions did not change?
After updating something changed:
Even some GPTs cannot create images some of them lost their files. And some GPTs capibilities were enable but now disable:
You can contact with help center. Because this forum is OpenAI Developer Forum, they don’t reply any topic about ChatGPT or custom GPTs.
For example these two custom GPTs of OpenAI don’t create images for three days:
Yes, this is a known issue that tends to surface after major model updates, like the shift to GPT-4o. Custom GPTs can lose their internal ability to reference uploaded files, even though the files still appear to be present in the interface.
This is likely due to changes in backend file indexing, embedding schemas, or capability toggles that OpenAI implements silently during system upgrades.
In short, re-uploading the file triggers a fresh indexing cycle. That usually resolves the issue because the new upload syncs correctly with the current model’s logic.
Some observed reasons this happens:
• File embeddings become outdated or disconnected from internal memory routing.
• Capability flags (like image generation or file referencing) are disabled or reset.
• The session state tied to the GPT’s configuration becomes stale or corrupted.
Recommended fixes:
Re-upload the files.
Refresh and re-save the system instructions
Keep a clear record of all needed files
Test: Create a log book for your gpt instructions as well as try adding within your instructions to preserve file integrity through system updates
There’s currently no official documentation addressing this, but multiple developers are seeing the same thing. Reindexing after a large update seems to be necessary until a solution is found to include debugging for this or more extensive separation when updates occur and how they effect the existing files and data
I hope this helps!
Just to clarify—this question was correctly posted here. Custom GPT file referencing issues, especially after major model updates like GPT-4o, fall under developer concerns. The Developer Forum is the appropriate space to discuss behavioral changes, indexing bugs, and emerging GPT logic issues.
Also, OpenAI Support typically doesn’t handle these types of technical matters. They focus on account-level support (billing, access, etc.), and for Custom GPT behavior, they’ll usually redirect you back here or offer general guidance.
So the original question was valid, and important to track—especially if more developers are noticing similar file referencing failures or capability losses after updates.
