← sources

Paid ChatGPT Pro 100 user dealing with repeated file failures, inaccessible work and support loops. Has anyone actually got this investigated and fixed?

source: reddit (reddit:ChatGPT)published 2026-11-11 23:43ingested 2026-11-12 02:37

extract

I need help from someone who has dealt with this and found an actual resolution. I pay for ChatGPT Pro 100 and rely on it heavily. Over the last several days, basic file workflows have repeatedly failed across different conversations, including new ones. Uploaded files appear to be there, but ChatGPT cannot read them. Generated files appear to be ready, but I cannot download them. This involves documents, spreadsheets and images. It is not just one CSV, one oversized document or one conversation. Meanwhile, I have spent days going back and forth with support, supplying information, explaining that the problem is continuing, asking for an investigation and receiving replies that often do not answer what I am asking. I am exhausted. Something that was supposed to save time has become a job in itself. I am leaving personal information, account details, private project names and support identifiers out of this post. I want help fixing the problem, not people contacting individual support staff. 1. The basic workflow I need is failing I am not asking ChatGPT to do something wildly unusual. I need to upload documents, have it work with them, and download the result. Sometimes that means comparing two documents. Sometimes it means reconciling spreadsheet information against screenshots. Sometimes it means getting an image it generated. If I cannot reliably get the finished work out, the fact that ChatGPT can describe what it supposedly did is not much use. There are at least two different failures here: • Files that appear to have uploaded successfully, but cannot actually be read by the assistant in that conversation. • Files that ChatGPT generates and presents for download, but that I cannot download. Those may have different technical causes. I cannot diagnose that from the interface. But both break the work I am trying to do. 2. Seeing a filename does not mean ChatGPT can use the file One of the most frustrating parts is that a file can appear to exist. ChatGPT can refer to its name. That makes it look as though we are ready to work. Then it turns out it cannot access the contents. In one document workflow, the explanation was that the file record existed, but the actual file contents had not reached the environment where the document needed to be opened and edited. The failure happened before it could actually open or parse the Word document. That is different from a document being corrupt. It is also different from the file simply not being there. As a user, though, I am left trying to figure out why the interface says one thing and the assistant can do another. I need a clear indication of whether a file is actually available for use. I should not have to spend a conversation discovering that a filename is visible but the document behind it cannot be opened. 3. A small CSV also failed This was not confined to a large document. I uploaded a CSV of roughly 6 KB along with five screenshots for a reconciliation task. The file could be identified, but the assistant could not read it in the conversation where I needed the work done. Support later said it found an upload record matching the conversation, size and timestamp I supplied. It also said the analysis tool logged file-read errors in that conversation, although those errors did not identify a filename. That distinction matters: support found a matching upload and file-read errors. I am not claiming those logs prove every detail of the underlying cause. The practical result was that I could not get the task done there. I also tried the same CSV, the same screenshots and the same instructions in Gemini. In my test, it handled the task in under a minute. That does not establish that Gemini is always better or never fails. It does explain why I am so frustrated when this turns into a prolonged discussion about my inputs instead of resolving the failure. A later attempt through another access route could read the CSV. That still does not tell me why the original conversation failed, or establish that the workflow is now reliable. 4. A generated PNG would not download either Support kept discussing CSVs, but there was also a separate image failure. I tried to download a generated PNG on my MacBook Pro. The message was: “Failed to download file. Please try again later.” I took a screenshot when it happened. That screenshot was supplied to support. It showed the error I could see. There was no useful explanation of the cause and no detailed file path for me to investigate. The latest support response contained a particularly concerning finding: “OpenAI's logs show that the PNG's file record could not be found when it was requested, about 30 seconds after its link was successfully saved. This explains how the chat could contain a generated link that failed to serve the file.” That is what support told me its logs showed. I do not have independent access to those logs. But if a generated file link can be saved and the file record cannot be found about 30 seconds later, this clearly needs more than another suggestion to try downloading again. What failed? Why did it fail? Can the file be recovered? What stops the next file from doing the same thing? Those are the questions I need answered. 5. This is happening across conversations I have tried new conversations. I have uploaded things again. The problems have continued. That is why I keep asking whether there is something account-specific going on, a file-storage problem, a problem with how files are passed to the tools, or something else. I am not claiming to know which explanation is correct. I am saying that treating each incident as an isolated user mistake is not addressing the pattern I am reporting. It is also not enough to say an earlier incident was resolved if I am still experiencing the same kind of failure afterward. I understand that similar symptoms can have different causes. If that is the explanation, investigate the current failures and tell me that. 6. There is substantial work at stake One of the documents involved is approximately 195 pages. There are also months of work, planning, project decisions, notes and follow-up steps tied to these conversations. The concern is bigger than downloading one attachment. When I tell ChatGPT to keep track of a decision, preserve a requirement or return to something at a later stage, I build subsequent work around that continuity. When I can no longer reliably access the work or determine which version includes which changes, I do not know what I can trust. I have said in frustration that everything is gone. To be precise, I cannot prove that everything has been permanently deleted. Some original files and saved planning documents have subsequently been located and opened through other access methods. That is important, and I am not going to hide it. But locating some files does not establish that every approved change, every note and every commitment has survived. It also does not mean I can download the files myself. “Your file exists somewhere” and “you can reliably access and use your completed work” are not the same outcome. I need to know what is actually recoverable, what version I have, and how I can get it out. 7. The support exchange has become another loop I have supplied information over multiple rounds, including affected conversations, file details, timestamps, device information, screenshots and the exact error text. The exchanges have included requests for more information, escalation notices, replies under individual names, discussion of workarounds, account-status checks and, eventually, the log findings described above. I have repeatedly asked for a person to take ownership of the technical problem. At different points, the replies have not been able to confirm a dedicated file-storage engineering assignment or a verified engineering timeline. Being referred to specialist support is not the same as being told who is investigating, what they have found and what happens next. The latest response did provide more concrete information than earlier ones. I want to acknowledge that fairly. But it also told me that my subscription was active and there was no recorded account deactivation. I was not asking whether my account had been cancelled. I was asking why files were not working. The response ended with: “If this doesn't help with your issue, please respond to this message.” Of course it does not resolve the issue. Explaining that a file record was missing does not make the download work. I need the investigation to continue from that finding, rather than another round in which I have to explain that the problem is still a problem. 8. I do not trust the claims about human support Some replies have been presented under a person's name, and I have been told I am dealing with a person. The wording feels extremely familiar to me as someone who uses AI extensively: the same polished phrases, acknowledgments and apparent understanding, followed by answers that miss the actual issue. I strongly suspect AI is being used to produce those replies. I cannot prove who wrote an individual message simply by its wording. A human could be using templates or AI assistance. A name in a signature does not settle the question either. What I can say is that a later reply was explicitly signed “OpenAI AI Support Agent.” So there has definitely been an AI-labelled response in this exchange. I cannot establish the authorship of every other message. The reason this matters to me is that I feel trapped in a loop: ChatGPT tells me to contact support; ChatGPT helps explain the failure; support sends back a response that feels like more ChatGPT; and I am still left with a broken workflow. I need someone accountable for investigating what happened. A sentence saying someone understands my frustration does not demonstrate that they have addressed the evidence. 9. I cannot screen-record something after it happened I have been asked for a screen recording showing the download attempt and failure. I understand that recordings can help diagnose a reproducible problem. If I deliberately test it again, a recording might capture a new occurrence. But I cannot record a past event. I took a screenshot when the failure happened, and supplied it. I also cannot reasonably keep a screen recording running every time I use ChatGPT just in case the next file fails. The interface gives me a generic error. It does not give me the server logs or the underlying cause. Support apparently has access to information I cannot see, as the PNG finding demonstrates. Use the evidence already supplied, tell me exactly what is still missing, and explain why it is needed. Do not make the absence of a retrospective recording the end of the investigation. 10. The trust problem extends beyond downloads The file failures are the clearest technical examples, but they sit alongside another problem: the assistant's statements do not reliably match its actions. I have repeatedly had experiences where it says it will do something, says it will not do something, or implies that a step has been completed, and later the outcome does not match. Then I get an explanation along the lines of what it should have done. That does not undo the mistake. It also leaves me checking whether the next assurance means anything. I have been told to put instructions into settings, and I have done that, yet I still see instructions disregarded. I understand that saved instructions are not a guarantee of perfect behaviour. But being told to use them as the solution, then having the same problem continue, is infuriating. There has even been a case where the assistant handling correspondence said I would be able to review a message first, then sent it without that review. When I have to verify the file, the contents, the version, the instructions and whether an action actually happened, the supposed time savings disappear. 11. Compensation is not my main goal I am not principally looking for a credit. I want a service I pay for and rely on to work. A temporary workaround may help me retrieve an urgent file. It does not explain or resolve repeated failures across different files and conversations. I want: • A technical investigation that addresses the actual upload and download failures. • A clear explanation of what support has verified and what remains unknown. • Confirmation of whether the relevant technical team is investigating. • A practical recovery plan for affected files and project work. • A way to verify that the problem is fixed, using the workflows that were failing. • A meaningful next update, rather than another invitation to repeat that it is still broken. What I am asking Reddit Has anyone experienced this specific combination of problems and actually got it resolved? In particular: 1. Have you had files appear uploaded, but remain unreadable to ChatGPT in that conversation? 2. Have freshly generated download links failed almost immediately? 3. Has support told you a file record was missing shortly after the link was created? 4. Have these failures continued across new conversations? 5. Did anyone get a confirmed technical investigation and a lasting fix? What was the route? 6. Is there a reliable way to recover original files and verify the latest version without rebuilding everything? 7. What specific evidence was genuinely useful to support, beyond screenshots, affected conversations, timestamps and file details? Please distinguish a workaround that got one file out from something that actually stopped the failures. Both could help, but I need to know which one it is. I am not asking anyone to speculate about my account or promise they can recover everything. I need practical help from people who have seen this and know what worked. I am angry because I have invested real time and work in this, and I still do not have a reliable answer. I do not need another polished paragraph telling me my frustration is understandable. I need my work accessible, my files downloadable, and the underlying problem investigated.

retained text (truncated)

I need help from someone who has dealt with this and found an actual resolution.

I pay for ChatGPT Pro 100 and rely on it heavily. Over the last several days, basic file workflows have repeatedly failed across different conversations, including new ones. Uploaded files appear to be there, but ChatGPT cannot read them. Generated files appear to be ready, but I cannot download them.

This involves documents, spreadsheets and images. It is not just one CSV, one oversized document or one conversation.

Meanwhile, I have spent days going back and forth with support, supplying information, explaining that the problem is continuing, asking for an investigation and receiving replies that often do not answer what I am asking.

I am exhausted. Something that was supposed to save time has become a job in itself.

I am leaving personal information, account details, private project names and support identifiers out of this post. I want help fixing the problem, not people contacting individual support staff.

1. The basic workflow I need is failing

I am not asking ChatGPT to do something wildly unusual.

I need to upload documents, have it work with them, and download the result. Sometimes that means comparing two documents. Sometimes it means reconciling spreadsheet information against screenshots. Sometimes it means getting an image it generated.

If I cannot reliably get the finished work out, the fact that ChatGPT can describe what it supposedly did is not much use.

There are at least two different failures here:

• Files that appear to have uploaded successfully, but cannot actually be read by the assistant in that conversation.  
• Files that ChatGPT generates and presents for download, but that I cannot download.

Those may have different technical causes. I cannot diagnose that from the interface. But both break the work I am trying to do.

2. Seeing a filename does not mean ChatGPT can use the file

One of the most frustrating parts is that a file can appear to exist. ChatGPT can refer to its name. That makes it look as though we are ready to work.

Then it turns out it cannot access the contents.

In one document workflow, the explanation was that the file record existed, but the actual file contents had not reached the environment where the document needed to be opened and edited. The failure happened before it could actually open or parse the Word document.

That is different from a document being corrupt. It is also different from the file simply not being there.

As a user, though, I am left trying to figure out why the interface says one thing and the assistant can do another.

I need a clear indication of whether a file is actually available for use. I should not have to spend a conversation discovering that a filename is visible but the document behind it cannot be opened.

3. A small CSV also failed

This was not confined to a large document.

I uploaded a CSV of roughly 6 KB along with five screenshots for a reconciliation task. The file could be identified, but the assistant could not read it in the conversation where I needed the work done.

Support later said it found an upload record matching the conversation, size and timestamp I supplied. It also said the analysis tool logged file-read errors in that conversation, although those errors did not identify a filename.

That distinction matters: support found a matching upload and file-read errors. I am not claiming those logs prove every detail of the underlying cause.

The practical result was that I could not get the task done there.

I also tried the same CSV, the same screenshots and the same instructions in Gemini. In my test, it handled the task in under a minute. That does not establish that Gemini is always better or never fails. It does explain why I am so frustrated when this turns into a prolonged discussion about my inputs instead of resolving the failure.

A later attempt through another access route could read the CSV. That still does not tell me why the original conversation failed, or establish that the workflow is now reliable.

4. A generated PNG would not download either

Support kept discussing CSVs, but there was also a separate image failure.

I tried to download a generated PNG on my MacBook Pro. The message was:

“Failed to download file. Please try again later.”

I took a screenshot when it happened. That screenshot was supplied to support. It showed the error I could see. There was no useful explanation of the cause and no detailed file path for me to investigate.

The latest support response contained a particularly concerning finding:

“OpenAI's logs show that the PNG's file record could not be found when it was requested, about 30 seconds after its link was successfully saved. This explains how the chat could contain a generated link that failed to serve the file.”

That is what support told me its logs showed. I do not have independent access to those logs.

But if a generated file link can be saved and the file record cannot be found about 30 seconds later, this clearly needs more than another suggestion to try downloading again.

What failed? Why did it fail? Can the file be recovered? What stops the next file from doing the same thing?

Those are the questions I need answered.

5. This is happening across conversations

I have tried new conversations. I have uploaded things again. The problems have continued.

That is why I keep asking whether there is something account-specific going on, a file-storage problem, a problem with how files are passed to the tools, or something else.

I am not claiming to know which explanation is correct.

I am saying that treating each incident as an isolated user mistake is not addressing the pattern I am reporting.

It is also not enough to say an earlier incident was resolved if I am still experiencing the same kind of failure afterward. I understand that similar symptoms can have different causes. If that is the explanation, investigate the current failures and tell me that.

6. There is substantial work at stake

One of the documents involved is approximately 195 pages. There are also months of work, planning, project decisions, notes and follow-up steps tied to these conversations.

The concern is bigger than downloading one attachment.

When I tell ChatGPT to keep track of a decision, preserve a requirement or return to something at a later stage, I build subsequent work around that continuity. When I can no longer reliably access the work or determine which version includes which changes, I do not know what I can trust.

I have said in frustration that everything is gone. To be precise, I cannot prove that everything has been permanently deleted.

Some original files and saved planning documents have subsequently been located and opened through other access methods. That is important, and I am not going to hide it.

But locating some files does not establish that every approved change, every note and every commitment has survived. It also does not mean I can download the files myself.

“Your file exists somewhere” and “you can reliably access and use your completed work” are not the same outcome.

I need to know what is actually recoverable, what version I have, and how I can get it out.

7. The support exchange has become another loop

I have supplied information over multiple rounds, including affected conversations, file details, timestamps, device information, screenshots and the exact error text.

The exchanges have included requests for more information, escalation notices, replies under individual names, discussion of workarounds, account-status checks and, eventually, the log findings described above.

I have repeatedly asked for a person to take ownership of the technical problem.

At different points, the replies have not been able to confirm a dedicated file-storage engineering assignment or a verified engineering timeline. Being referred to specialist support is not the same as being told who is investigating, what they have found and what happens next.

The latest response did provide more concrete information than earlier ones. I want to acknowledge that fairly.

But it also told me that my subscription was active and there was no recorded account deactivation.

I was not asking whether my account had been cancelled. I was asking why files were not working.

The response ended with:

“If this doesn't help with your issue, please respond to this message.”

Of course it does not resolve the issue. Explaining that a file record was missing does not make the download work.

I need the investigation to continue from that finding, rather than another round in which I have to explain that the problem is still a problem.

8. I do not trust the claims about human support

Some replies have been presented under a person's name, and I have been told I am dealing with a person. The wording feels extremely familiar to me as someone who uses AI extensively: the same polished phrases, acknowledgments and apparent understanding, followed by answers that miss the actual issue.

I strongly suspect AI is being used to produce those replies.

I cannot prove who wrote an individual message simply by its wording. A human could be using templates or AI assistance. A name in a signature does not settle the question either.

What I can say is that a later reply was explicitly signed “OpenAI AI Support Agent.”

So there has definitely been an AI-labelled response in this exchange. I cannot establish the authorship of every other message.

The reason this matters to me is that I feel trapped in a loop: ChatGPT tells me to contact support; ChatGPT helps explain the failure; support sends back a response that feels like more ChatGPT; and I am still left with a broken workflow.

I need someone accountable for investigating what happened. A sentence saying someone understands my frustration does not demonstrate that they have addressed the evidence.

9. I cannot screen-record something after it happened

I have been asked for a screen recording showing the download attempt and failure.

I understand that recordings can help diagnose a reproducible problem. If I deliberately test it again, a recording might capture a new occurrence.

But I cannot record a past event. I took a screenshot when the failure happened, and supplied it.

I also cannot reasonably keep a screen recording running every time I use ChatGPT just in case the next file fails.

The interface gives me a generic error. It does not give me the server logs or the underlying cause. Support apparently has access to information I cannot see, as the PNG finding demonstrates.

Use the evidence already supplied, tell me exactly what is still missing, and explain why it is needed. Do not make the absence of a retrospective recording the end of the investigation.

10. The trust problem extends beyond downloads

The file failures are the clearest technical examples, but they sit alongside another problem: the assistant's statements do not reliably match its actions.

I have repeatedly had experiences where it says it will do something, says it will not do something, or implies that a step has been completed, and later the outcome does not match.

Then I get an explanation along the lines of what it should have done.

That does not undo the mistake. It also leaves me checking whether the next assurance means anything.

I have been told to put instructions into settings, and I have done that, yet I still see instructions disregarded. I understand that saved instructions are not a guarantee of perfect behaviour. But being told to use them as the solution, then having the same problem continue, is infuriating.

There has even been a case where the assistant handling correspondence said I would be able to review a message first, then sent it without that review.

When I have to verify the file, the contents, the version, the instructions and whether an action actually happened, 

document fetched 2026-11-12 00:20 · canonical https://reddit.com/r/ChatGPT/comments/1xtxg3n/paid_chatgpt_pro_100_user_dealing_with_repeated