I’ve just discovered a rather strange consequence of the way long ChatGPT conversations are handled now.
Until recently, when I wanted to keep a copy of one of my longer conversations, I could simply open the shared chat and use Print → Save as PDF. It wasn’t sophisticated, but it worked. I got the whole conversation in one file.
That no longer works.
Long conversations are now loaded in smaller sections. Only a few messages are rendered at a time, so printing or saving the page gives me only the currently visible part of the conversation.
I understand why this was done. Some of my chats are several months long, so keeping the whole thing rendered in the browser probably isn’t a brilliant idea.
The problem is what happens when I actually need the whole conversation.
I have one long technical chat containing months of PC diagnostics: crash dumps, WinDbg analysis, BIOS changes, RAM settings, tests, conclusions, wrong turns and eventually the changes that made the system stable. I now want to reuse that conversation as source material in another ChatGPT project.
ChatGPT itself cannot reliably recall every detail from such a huge old conversation. Memory obviously isn’t a replacement for the original data.
So I contacted OpenAI Support and asked what I’m supposed to do.
The official answer was essentially:
- Request a full account data export.
- Wait for it to be prepared — support says this can take up to 7 days.
- Download the export.
- Find the conversation I need in
conversations.jsonor the exported conversation files. - Extract it.
- Upload it back to ChatGPT.
I currently have more than 300 conversations.
So, unless I’m missing something, the official way to obtain one complete conversation created in ChatGPT is to export the data for 300+ conversations, search through it for the one I need, extract it, and then upload it back into ChatGPT.
That seems slightly excessive. 
I’m not asking for conversation virtualization to be removed. If it improves performance, great.
But virtualization really needs a corresponding Export conversation function that works independently of what happens to be rendered in the browser.
PDF would be useful for humans. Markdown, HTML or JSON would be even better for feeding an old conversation back into ChatGPT.
Something as simple as:
⋯ → Export conversation → Markdown / HTML / JSON / PDF
would solve the entire problem.
OpenAI Support has already recorded this as product feedback under case #15806671, but I’m posting it here because I’m curious whether other people with long-running ChatGPT projects have run into the same problem.