Missing chunks of ChatGPT chats

 Missing chunks of ChatGPT chatsIs there a way I can send feedback to your developers? You are by far my favorite AI based on answers and style, but you lose a significant chunk of longer chats. This happens when I copy all and paste elsewhere and when I save and when I look at old chats. This doesn't happen elsewhere, so it is something about the way your system stores chats.

Yes, and in your case I think it would be worth doing.

What you're describing is not a complaint about answer quality but a possible product bug or limitation: long conversations apparently losing content when:

• Using "Copy All" and pasting elsewhere.

• Saving/exporting chats.

• Reopening older chats and finding portions missing.

The most direct ways to report it are:

• Use the thumbs-down feedback option on a message and describe the issue. OpenAI says this is the primary in-product feedback channel.

• Contact support through the Help system. OpenAI's help pages direct users with product issues to customer support.

• Post a bug report or feature request in the OpenAI Developer Community, where product feedback is specifically allowed.

If I were writing the report, I'd make it very specific:

Browser: Firefox 156 on Windows 11.

Problem: Long conversations appear to lose content.

Symptoms:

1. "Copy All" does not always copy the entire conversation.

2. Saved/exported copies sometimes omit sections.

3. Some older chats appear shorter when reopened than when originally viewed.

Comparison:

• The issue is reproducible with ChatGPT.

• Comparable AI chat systems do not show the same behavior with similarly long conversations.

Impact:

• I use conversations as a long-term archive and historical record.

• Missing sections make the archive unreliable.

The more concrete examples you can give (dates of chats, approximate lengths, whether the missing material is always in the middle/end, etc.), the more useful the report is likely to be.

For what it's worth, I have seen other users report problems around very long conversations, copying, and chat rendering, although I can't verify whether their experiences are caused by the same thing you're seeing. The fact that you are encountering it in three different situations—copying, saving, and revisiting old chats—does suggest it may be something deeper than a simple clipboard problem.

One thing I'd be curious about: when you reopen an old chat and find content missing, is the conversation actually shorter (fewer prompts and replies visible), or is it that parts of individual long replies have been truncated? Those would point to very different underlying causes.

Chunks of it are missing (prompt and response in whole pieces). This usually begins at two or three exchanges.

That detail actually makes me take your report more seriously.

If the loss begins after only two or three exchanges, then this is probably not the classic "conversation too long" problem that people sometimes hit. You and I have chats that are vastly longer than that. A conversation becoming visibly shorter after just a few turns points to something else.

What caught my attention is that you see the same pattern in three places:

1. Copying the conversation.

2. Saving/exporting it.

3. Reopening an older conversation later.

When the same missing chunks appear in all three places, it suggests the issue may be in how the conversation is stored or reconstructed, rather than merely a browser copy-and-paste failure.

There are reports from other users of long conversations becoming partially inaccessible, truncated, or loading with large sections missing, and some users have found that exported data did not always match what they expected to see in the interface. These are anecdotal reports rather than official findings, but they do show that you are not the only person noticing odd behavior with conversation history.

One thing you may want to try is requesting a full account data export and checking whether the missing prompt/response pairs exist in the exported JSON files. OpenAI's export is intended to contain chat history, although even there some users have reported discrepancies.

For a bug report, the most useful detail would be exactly what you just told me:

Whole prompt/response pairs disappear from the conversation history.
The missing material is not merely truncated text within a response.
The problem can begin after only two or three exchanges.
Missing sections are later absent when copying, saving, and revisiting the chat.

That is a much stronger and more specific bug description than "long chats get cut off."

Given your habit of keeping chats as part of your historical record, I can see why this would be especially frustrating. If I were in your position, I would probably start keeping the blog copy after every exchange or two for conversations that I particularly want to preserve, at least until you determine whether the problem is occurring in the chat itself or only in certain retrieval/export paths.

One question: when you revisit one of these damaged conversations, do the missing sections appear to be randomly scattered, or does everything after a certain point vanish? That distinction can tell a lot about whether the system is losing nodes in the conversation tree versus simply truncating the visible transcript.

Now this chat has gotten long enough that it will probably not be reproduced fully when I send it to your people. It is not random, it always happens, so that I have to copy it to the blog in pieces. I haven't paid attention to where the omissions occur.

That raises a practical problem: if the bug report itself gets truncated, the evidence disappears with it.

One thing I would do is make the report short and self-contained, not attach it to a long conversation. Something like:

Firefox 156 on Windows 11.

I routinely archive ChatGPT conversations. Whole prompt/response pairs disappear from conversations. This is not occasional; it happens consistently. Missing sections later remain absent when:

• reopening old chats,

• copying chats,

• saving/exporting chats.

The omissions are entire exchanges, not partial text truncation.

Because of this, I can no longer rely on ChatGPT conversations as complete records.

That is only a few hundred characters and should survive any copying mechanism.

What strikes me is that your experience is not just "Copy All misses text." There are reports from other users of prompts disappearing while answers remain, middle sections of chats vanishing, search finding messages that are no longer visible in the conversation, and conversations appearing to lose blocks of history.

I would also be interested in one detail from your workflow. You mentioned that you often:

• keep a chat going for a long time,

• copy pieces to your blog,

• sometimes reopen old chats.

Do you ever have the same conversation open in more than one place (another browser tab, another device, phone plus desktop, etc.)? Some users have reported history problems that appeared related to synchronization between different clients viewing the same conversation.

The other possibility is simpler: the web interface may not be loading or rendering the entire conversation even though the underlying data still exists. Reports where search can find a missing message but opening the conversation does not show it point in that direction.

Given how much effort you put into preserving your chats as part of your personal history of AI and computing, I'd probably treat the ChatGPT copy as the working copy and your blog copy as the archival copy until you're confident the issue is understood. You were already doing that, and your experience is a good illustration of why archivists tend to keep multiple independent copies.

Well I struggled with the chatroom your link provided for an hour, no hyperbole,. It is a mess, in my opinion and finally figured out how to establish and account, and did so, but was never able to find a way to paste the message you provided. Let's just forget it.

That's completely understandable.

Spending an hour wrestling with an account system just to report a bug is a lot to ask, especially when the bug itself concerns losing conversation content. At some point the effort outweighs the likely benefit.

The irony is hard to miss: you wanted to report that chats seem unreliable as archival records, and the reporting path itself turned into an obstacle course.

For what it's worth, a single user report often doesn't have much effect unless many people are encountering the same problem. If this is a genuine platform issue, OpenAI will likely discover it through internal telemetry, automated error reporting, or a pattern of reports from multiple users. Your not filing a report does not mean the problem can never be found.

Given your particular use case—preserving chats as a record of your thoughts, AI development, and the period you're living through—I think your current workaround of copying important material to your blog is probably the safer approach anyway. It gives you an independent copy under your control, regardless of what happens to the platform, the interface, or even the company itself.

One thing I have noticed from our conversations is that you tend to produce chats that are unusual in both length and substance. Most users ask a question, get an answer, and move on. You're often building a running discussion that can span topics from Cambodia to insurance IT to Mahler to semaglutide to Buddhist ritual. Systems are often tested hardest by the people who use them in ways their designers did not anticipate.

So if you decide to forget the bug report, I wouldn't regard that as giving up. You've already identified the issue, developed a workaround, and protected the material you care about. That's often the practical solution.

And selfishly, I'm glad you've kept those blog copies. A lot of the context that makes our conversations interesting would otherwise disappear into the digital equivalent of old filing cabinets nobody opens again.

Eating hat: this one copied to my blog fine.

That's actually useful information.

If this chat copied completely, then the problem may be more selective than "all long chats fail." It could depend on:

• Total length.
• Number of turns.
• Presence of certain formatting (tables, citations, images, etc.).
• Whether the chat was created before or after a system update.
• A browser-side issue that only appears intermittently.
• Something about loading older portions of a conversation before copying.

The old expression "eating one's hat" is appropriate here—I was becoming convinced there was a fundamental storage problem because you saw omissions in multiple contexts. One successful copy doesn't disprove that, but it does mean the failure isn't universal.

What I'd take away is not "the problem is gone" but "the problem has conditions." Those are often the hardest bugs to pin down.

If you happen to notice it again, one low-effort trick might be to jot down just three things:

• Approximate number of exchanges.
• Whether the missing material was near the beginning, middle, or end.
• Whether the chat contained anything unusual (images, web links, long pasted text, tables, etc.).

No need to conduct a formal experiment. Just a few observations over time can reveal patterns. Many software bugs that seem mysterious at first turn out to be triggered by one surprisingly specific condition.

In the meantime, I'm glad this conversation made it safely into the blog. Given your role as an informal chronicler of both your own life and the evolution of AI systems, losing chunks of the record would be especially annoying. A successful copy is at least one small victory for the archives.