Why ChatGPT Can’t Create a File: Causes and Fixes for Generation and Download Errors

You ask ChatGPT to create a PDF or Excel workbook, but it returns only text. On another attempt, the task stays on “Generating” or “Working.” Sometimes a file name appears, yet selecting it leads to “Download failed” or “File Not Found.” These symptoms look similar, but they do not have the same cause.

The practical answer is to identify whether the task failed before creation, during generation, or at download time. The problem may involve the workflow you chose, a Custom GPT setting, file size, an expired link, the browser or network, or a temporary service incident—not just the wording of your prompt.

Based on OpenAI’s official guidance available on July 22, 2026, this article organizes the likely causes by symptom and provides a recovery sequence designed to preserve your work. It covers standard ChatGPT and Custom GPTs, not file-generation problems in an API application.

First, Identify Where the File Process Stopped

“ChatGPT cannot create my file” can describe several different failures:

• The response contains text, but there is no file name or download control.
• The task remains on “Generating” or “Working.”
• A file appears, but selecting it shows “Download failed” or “File Not Found.”
• File creation works in standard ChatGPT but fails in one Custom GPT.
• The source material fails to upload because of a format or limit error.

If the draft text or table is still visible, copy it somewhere safe first. Then start with the section matching the point of failure. This is usually more efficient than immediately clearing every setting in your browser.

If You Only Get Text, Check the File-Creation Workflow

A draft displayed in the chat is not necessarily a finished downloadable file. For writing-focused documents, one supported route is to open the content in Canvas, review it, and export it from the Download control in the upper-right corner.

OpenAI’s Canvas documentation says general documents can be exported as PDF, Markdown, or Word (.docx). A code Canvas can use an extension appropriate to its language, such as .py, .js, or .sql.

If ChatGPT keeps returning text, specify both the deliverable and the workflow. For example: “Open this content in a general-document Canvas and prepare it so I can download it as a Word file.” For a workbook with multiple sheets or structured calculations, use a feature that supports data analysis and file creation, then begin with a small sample that is easy to verify.

Canvas has the advantage of letting you inspect and revise the document before downloading it. The tradeoff is that availability can depend on the model and Custom GPT configuration, so a missing Canvas control may require the checks below.

If Generation Never Finishes, Try a New Chat and a Clean Connection

For a task stuck on “Thinking,” “Generating,” or “Working,” OpenAI’s troubleshooting guide recommends waiting 30–60 seconds first. If it still does not finish, stop the response and regenerate it. In a long conversation, start a new chat and bring over only the source material and requirements needed for the file.

Next, reload the page or app and try an incognito window or another browser. Content blockers, privacy or security extensions, VPNs, proxies, and secure DNS tools can interfere with the connection. Where policy allows, temporarily disable them to test whether they are involved. Trying another network or device can also help isolate the cause.

Do not move confidential material to an unauthorized personal device or network. In a managed company or school environment, firewall rules or workspace controls may be involved, so contacting the administrator is safer than attempting to bypass the restriction.

If the same failure appears across devices and networks, check OpenAI Status for an active incident. During a service problem, preserve the draft and retry after recovery instead of repeatedly changing local settings.

For “Download Failed” or “File Not Found,” Check Link Age and Size

OpenAI’s troubleshooting guidance warns that links to ChatGPT-generated files can expire quickly. When a file appears, review it and download it promptly.

If the link may have expired, do not keep selecting the old link. Ask ChatGPT to regenerate the file in the same conversation. A request such as “Keep the content unchanged and generate a new file in the same format” makes the goal clear. Download and open the replacement immediately.

The same official guide lists keeping the downloadable file below 512 MB as another check. If the output contains many images, sheets, or pages, reduce image sizes, split the date range or sheets, or test with only the required pages. Splitting creates more files to manage, but it also makes it easier to see whether the failure occurs during generation or download.

If a new, smaller file still fails, disable a VPN or proxy and try another browser or connection. A displayed file name does not always prove that the underlying generation step completed, so ask ChatGPT to confirm completion and regenerate the file if necessary.

If Only a Custom GPT Fails, Review Its Capabilities

When one Custom GPT cannot return a file, open GPT Builder and confirm that Code Interpreter & Data Analysis is enabled. OpenAI’s GPT troubleshooting guide also says to update the GPT after changing the setting. Then use Preview to create a small file and test generation and download separately.

Canvas in a Custom GPT has its own capability setting. As of July 22, 2026, OpenAI’s Canvas documentation says the capability is off by default for existing Custom GPTs and on by default for newly created ones. It also says Custom GPT Canvas is not supported by GPT-5.5 or later models; when Canvas is enabled, incompatible models cannot be selected.

If the required capability is missing entirely, check the account plan, workspace policy, and your role permissions. In a managed workspace, the user may not be allowed to change the relevant setting. Ask the administrator rather than attempting to work around the restriction.

If the Source File Will Not Upload, Treat It as a Pre-Generation Problem

When a task depends on uploaded source material, a failed upload prevents file generation from starting. This is separate from a failure to download the finished output.

OpenAI’s File Uploads FAQ lists a hard limit of 512 MB per file, a limit of 2 million tokens for a text or document file, approximately 50 MB for a CSV or spreadsheet depending on row size, and 20 MB for an image. Common document formats are supported, but the Google Docs .gdoc format is not; export it to PDF or DOCX before attaching it.

The FAQ also describes rolling upload-rate and shared-storage limits, and notes that failed attempts can sometimes count toward the rate limit. If an upload-limit message appears, confirm that you are signed into the intended account and plan, then remove unneeded files from recent chats or Custom GPTs. Settings shows Library storage use, but it does not show the remaining rolling upload allowance.

Breaking a large source into the necessary pages, columns, or date ranges can make a retry easier. However, splitting may remove context or change an aggregation range. Label each part and tell ChatGPT how the parts should be combined in the final output.

A Recovery Sequence to Follow When the Cause Is Unclear

If you cannot identify one clear cause, use this order to preserve your work and limit unnecessary changes:

1. Copy and save the visible draft, table, and requirements.
2. Remove large images or extra sheets and retry with a smaller file.
3. Move only the required context into a new chat.
4. For a document, open Canvas and use one of its supported export formats.
5. For a Custom GPT, verify Code Interpreter & Data Analysis, Canvas, model compatibility, and saved settings.
6. Download the newly generated file immediately.
7. Reload, then try an incognito window or another browser.
8. Test without interfering extensions, VPNs, proxies, or secure DNS tools.
9. Use another approved network or device.
10. Check OpenAI Status for an incident.

If the issue still reproduces across environments, contact OpenAI Support with the timestamp and time zone, model, conversation URL or ID, device, app or browser, error screenshot, and request ID when available. OpenAI also recommends browser console logs and a HAR file for persistent browser problems. HAR files can contain sensitive data, so handle them only according to the official support instructions.

Summary

When ChatGPT cannot create a file, do not begin by repeatedly rewriting the prompt. First identify whether the process stopped before creation, during generation, or at download time.

• If you only receive text, request a supported downloadable workflow such as Canvas.
• If generation stalls, test a new chat, browser, connection, and OpenAI Status.
• If download fails, check whether the link expired, the file is below 512 MB, or a VPN or proxy is interfering.
• For a Custom GPT, verify Code Interpreter & Data Analysis, Canvas, model compatibility, and permissions.
• If a source file cannot be attached, investigate its format, size, upload rate, and storage allowance separately.

The safest first step is to save the visible content and reproduce the task with a small file in a new chat.

File creation can fail because of a small workflow or timing mismatch—not only because of the AI itself—so locating the exact stopping point is often the fastest route back to a usable download.