ChatGPT Projects Not Reading Files
When a ChatGPT Project accepts your file but then cannot read it, or answers about the wrong content, the cause is usually the file or retrieval, not a broken account. ChatGPT fails to read an upload for five main reasons: it is a scanned, image-only PDF with no selectable text (ChatGPT does not run OCR by default), the format is unsupported, the document is too large or exceeds the token budget, the file is password-protected or corrupted, or the model simply did not invoke the retrieval tool. On top of that, some 2026 builds had a bug where files added on the web were not read reliably while the same files added via the desktop app worked. The fixes are to make the file machine-readable (OCR to text or DOCX), keep it a supported format and reasonable size, and explicitly tell the Project to consult the file. This guide covers each.
Affiliate link, we may earn a commission at no extra cost to you.
Symptoms You Might See
The Project accepts the upload but says it cannot access the file
ChatGPT answers from general knowledge instead of your document
It reads a DOCX or TXT fine but not your PDF
Only part of a long document seems to be used
Files added on the web are ignored, but the desktop app reads them
Common Causes
Scanned, image-only PDF
ChatGPT does not run OCR on uploaded PDFs by default. If the PDF is a scan or export with no selectable text, there is nothing for the model to read, so it appears to ignore the file.
Unsupported or awkward format
ChatGPT reliably reads PDF, DOCX, PPTX, TXT, CSV, JSON, XML, HTML, common code files, native spreadsheets (XLSX, XLS), and standard images (PNG, JPG). Obscure or non-standard formats extract poorly or not at all.
File too large or over the token budget
Very large files, or many files, can exceed what the Project can hold in context at once. Plus users get the most reliable results with files well under the theoretical limit (practically, tens of MB, not hundreds).
Retrieval did not fire
Like Custom GPTs, a Project does not always search its files on every turn. If you do not point it at the document, it may answer from general knowledge and skip retrieval entirely.
Web vs app upload bug
In some 2026 builds, files added through the web were not read reliably in Projects while the same files added via the desktop app worked, an upload-surface bug rather than a file problem.
Platform-wide file_search outage (May 2026)
Starting around May 7, 2026, OpenAI had a period where file_search and cross-chat reference inside Projects broke widely, reported on OpenAI's own developer forum. If your Project file access broke suddenly with no local change on your end, this kind of service-side incident, not a broken upload, is the more likely cause. Check status.openai.com for anything Projects or file-search related before re-uploading everything.
Desktop app redesign hid Projects (July 2026)
The July 9, 2026 rebuild of the ChatGPT Mac app, which merged in Codex, moved chat behind a toggle and hid Projects from the desktop sidebar entirely for some users (OpenAI partly reversed this on July 17). Files and Projects were not actually deleted, they were still visible on chatgpt.com in a browser and via desktop search, but the sidebar made them look gone. If Projects vanished specifically after a Mac app update, check the web version before assuming your files or Project were lost.
Step-by-Step Fixes
OCR scanned PDFs before uploading
When to try: First, if the file is a scanned or image PDF
If the file is a scan or image-only PDF, run OCR first (Adobe Acrobat, Google Docs 'Open with Google Docs', or an OCR app) and save the result as DOCX or TXT. Text-based files extract far more reliably than PDFs, and this fixes the most common 'ignores my file' case.
Explicitly tell it to use the file
When to try: When it answers without using your document
Do not assume the Project will search on its own. Prompt it: 'Using only the files in this project, answer... and quote the relevant section. If the answer is not in the files, say so rather than guessing.' This forces retrieval instead of a general-knowledge answer.
Use a supported, clean format and reasonable size
When to try: For large files or unusual formats
Convert to DOCX or TXT for text, keep spreadsheets as XLSX/CSV, and split very large documents into logical sections. Aim for files in the tens of MB rather than hundreds. Smaller, clean, supported files are read most reliably.
Re-upload, and try the desktop app
When to try: If files added on the web are not being read
Remove and re-add the file, then confirm it still appears. If web uploads are being ignored, add the file through the ChatGPT desktop app instead, which sidestepped the 2026 web-upload bug for many users.
Split long documents and synthesize
When to try: When only part of a long document is used
For a document that exceeds the context budget, split it into chapters or themes, ask ChatGPT to analyze each section, then ask for a combined summary. This often produces better results than forcing one huge file through at once.
Refresh the workspace and check basics
When to try: If nothing above resolves it
Refresh the Project, clear browser cache and cookies for chatgpt.com, disable conflicting extensions, and confirm the file is not password-protected or corrupted (open it yourself first). Check status.openai.com if uploads fail across files.
Check chatgpt.com in a browser before assuming files are lost
When to try: When Projects or files vanished specifically after a desktop/Mac app update
If Projects or its files disappeared right after updating the Mac desktop app, open chatgpt.com in a regular browser first. A July 2026 desktop redesign hid Projects from the Mac app sidebar for some users without actually deleting anything, the Project and its files were still reachable on the web and via desktop search. If the web version shows your files intact, the issue is the desktop app's display, not your data, update the app or use the web version until it is fixed.
Check status.openai.com for a file_search incident
When to try: When multiple files across multiple Projects all stopped being read at the same time
If file reading or cross-chat reference inside Projects broke suddenly across multiple files with no change on your end, check status.openai.com and OpenAI's developer community for a reported file_search or Projects incident before spending time re-uploading. These platform-side outages have happened before (notably starting around May 7, 2026) and resolve on OpenAI's end, not through anything you change locally.
How to Prevent This
Convert scanned PDFs to searchable text (OCR to DOCX/TXT) before uploading
Keep Project files in supported formats and reasonable sizes
Always instruct the Project to consult its files and quote from them
For big references, split into sections rather than one massive file
If web uploads act up, add files via the desktop app
When to Contact Support
Contact OpenAI via help.openai.com if a supported, text-based, reasonably sized file that opens fine on your computer is still not read after re-uploading on both web and desktop, and instructing retrieval explicitly. Include the file type and size. For scanned PDFs, format, or size issues, the fixes above resolve it faster than support can.
Frequently Asked Questions
Why won't ChatGPT read my uploaded PDF?▼
The most common reason is a scanned, image-only PDF with no selectable text, ChatGPT does not OCR PDFs by default. Run OCR (Acrobat, Google Docs, or an OCR app) and save as DOCX or TXT, then re-upload. Also check the file is not too large, password-protected, or corrupted.
The file uploaded but ChatGPT answers from general knowledge. Why?▼
A Project does not always search its files automatically. Tell it explicitly: 'using only the files in this project, answer X and quote the relevant part, and if it is not in the files, say so'. That forces it to actually read the document.
What file formats does ChatGPT read reliably?▼
PDF (with real text), DOCX, PPTX, TXT, CSV, JSON, XML, HTML, common code files, native spreadsheets (XLSX, XLS), and standard images (PNG, JPG). For text, DOCX or TXT extract most reliably. Convert anything obscure before uploading.
Files I add on the web are ignored but the desktop app works. Is that a bug?▼
Yes, some 2026 Project builds had exactly this web-upload issue while desktop-app uploads read fine. Add the file through the ChatGPT desktop app as a workaround, and re-upload on web after refreshing.
Only part of my long document seems to be used. What do I do?▼
The document likely exceeds the context budget. Split it into chapters or themes, have ChatGPT analyze each part, then ask for a combined summary. Chunked analysis beats forcing one oversized file through at once.
My ChatGPT Projects and their files disappeared from the Mac app. Are they gone?▼
Probably not. The July 9, 2026 rebuild of the ChatGPT Mac app (which merged in Codex) moved chat behind a toggle and hid Projects from the desktop sidebar for some users, OpenAI partly reversed this on July 17. Before assuming data loss, open chatgpt.com in a browser, Projects and their files were generally still there and reachable, the app's sidebar was just not showing them. Update the desktop app to the latest version and check the web version in the meantime.
File reading in Projects broke for everything at once, is this an outage?▼
It can be. OpenAI has had platform-side incidents where file_search and cross-chat reference inside Projects broke widely, one was reported starting around May 7, 2026 on OpenAI's developer forum. If every file across multiple Projects stopped being read at the same time with no change on your end, check status.openai.com and the OpenAI developer community before re-uploading everything, it may resolve on its own once OpenAI fixes the service.
Other Common AI Errors
ChatGPT Not Working
Service Issues
🚦Rate Limit Exceeded
Usage Limits
📏Context Length Exceeded
Input Limits
🔥Model Overloaded
Service Issues