I can accept that software has bugs. What I’m struggling to accept is spending a month trying to find out whether anyone is actually investigating one.
I had a working Custom GPT. During the move to plugins, a basic part of that workflow stopped working: reading the data packaged with it. Instead, it would sometimes search the internet for information it already had sitting in its own files. Through self-analysis, there is a problem on openai side.
That started a long troubleshooting exercise. Different versions, controlled tests, package checks, detailed reports. I was thorough. I used OpenAI’s own models to help investigate, including ChatGPT Astra. The analyses kept pointing toward an OpenAI-side problem, with the practical conclusion that I should report it and wait for a fix.
Fine. But what exactly am I waiting for? After almost a month, I still don’t know whether the case has reached anyone who can investigate or fix it.
Support has received technical reports, test results, package files, conversation links, and the history of what has already been checked. Yet replies keep circling back to generic questions about installation, connected accounts, permissions, and sharing—often things that do not apply to this particular plugin.
It feels like a support version of Groundhog Day. Someone acknowledges the details, and the next reply sends us back to the beginning.
Then there was the screen recording.
It showed my work desktop, so I shared it through SharePoint with access protection. Automated support kept saying it couldn’t open the link and asking me to provide it without a password. I refused to remove the protection. The response was essentially that they couldn’t help me.
After I challenged this twice, a human representative finally said I didn’t need to resend the recording or repeat the diagnostics. A perfectly reasonable answer—eventually.
To be clear, the model-assisted analyses are not proof of the backend cause. OpenAI could investigate and find a package issue or an undocumented limitation. I would welcome a concrete explanation. I’m not demanding that support agree with my diagnosis.
I’m asking whether there is an investigation at all?
Has someone reviewed the execution traces? Has engineering received the case? Is there a specific unanswered question? Is this a known limitation? Those would all be useful answers.
Instead, I’m left with polite emails and no visible technical progress.
The irony is hard to miss: OpenAI’s most capable models help me investigate the problem and tell me to contact OpenAI. Then contacting OpenAI becomes the part I can’t get through.
I don’t expect a miracle. After a month, I do expect to know whether somebody has picked up the problem. I am frustrated. I wantet to share it.
It feels like there’s nowhere to give meaningful feedback—every channel is a black hole that swallows the details and spits out another “thank you for reaching out.”