Why Can't ChatGPT Read My PDF? 5 Causes and Fixes (2026)

ChatGPT reads the text layer of a PDF and discards everything else. So if ChatGPT can't read your PDF, the file is one of five things: scanned with no text layer, over the size or token limit, blocked by an upload cap, password-protected, or reachable only through a link it can't open. Each has a quick fix, and each is below, along with the official limits and what the error messages actually mean.

Cause 1 - Your PDF is scanned (no text layer)

This is the single most common reason. A PDF can look like text on screen but actually be a picture of text - a scan, a photo, an export from a fax or a signing tool. There's no selectable text underneath, so there's nothing for ChatGPT to read. OpenAI is explicit about this in its File Uploads FAQ: on every plan except Enterprise, document files support text-based retrieval only - ChatGPT extracts the digital text and discards any images.

How to check: open the PDF and try to select a sentence with your cursor. If nothing highlights, the file is image-only. (Why can't I copy text from a PDF? covers the other reasons selection fails.)

The fix: the document needs OCR - optical character recognition - to turn the picture of text back into real, machine-readable text. Run it through OCR and upload the searchable copy, following how to make a PDF searchable, or use a reader that does OCR automatically before answering. Our full walkthrough on extracting text from a scanned PDF covers the tools in depth.

Cause 2 - The file is over the size or token limit

OpenAI's hard limits at the time of writing: 512 MB per file, and 2 million tokens per text or document file (roughly 1.5 million words). Long or image-heavy PDFs can also exceed what fits in a single conversation. When that happens, the document may be truncated - ChatGPT reads the first part and answers as if that's the whole thing, which is worse than a clean failure because the answer looks complete.

The fix: split the document into sections and work through them one at a time, or strip out large embedded images so the text fits. If you keep hitting this on the same big files, that's a sign you want a tool built to read whole documents rather than paste them into a chat box.

Cause 3 - You've hit an upload cap

This one surprises people because the file itself is fine. OpenAI applies a rolling rate limit of 80 file uploads every 3 hours, and Free users get 3 uploads per day. Separately, storage is capped at 25 GB per user (100 GB per organisation) across chats, Projects and custom GPTs, and failed upload attempts can count toward the rate limit. Both show up as "upload limit reached".

The fix: check Settings, then Storage, and delete files from old chats or GPTs you no longer need. If it's the rate cap, wait it out or, on Free, plan the day's three uploads. OpenAI also notes it may lower these limits at peak times.

Cause 4 - The PDF is password-protected or encrypted

Locked PDFs can't be parsed until they're unlocked. ChatGPT will usually tell you it can't open the file rather than guess.

The fix: remove the password in your PDF viewer (if you have the right to), save an unprotected copy, and upload that.

Cause 5 - You gave it a link, not a file

If your PDF lives in Google Drive, SharePoint, a company wiki, or behind any login, pasting the URL rarely works - ChatGPT can't open links behind authentication, and even public links are hit-or-miss.

The fix: download the file and upload it directly, or paste the relevant text. Better still, use a reader that works on the page you're already looking at, so there's no download-and-re-upload dance.

ChatGPT's file upload limits (the official numbers)

All from OpenAI's File Uploads FAQ at the time of writing; OpenAI adjusts them, so treat the table as a guide and check the FAQ before planning around a specific figure.

ChatGPT's error messages, decoded

The exact wording changes between versions, but the messages map onto the causes above:

• "Upload limit reached" - the rate cap or the storage cap (Cause 3). Check storage first; it's the one that persists.
• "This file is too large" or an upload that never completes - over 512 MB, or over the token limit for a text file (Cause 2). Split it.
• "Unable to extract text", an answer about an empty document, or a summary of a file that has nothing to do with yours - a scanned PDF (Cause 1). OCR it.
• "I can't access external links" or a summary of the login page instead of the document - a link, not a file (Cause 5). Upload it.
• "There was an error processing your file" - usually a locked, corrupted or unusually built PDF (Cause 4). Print it to a fresh PDF from your viewer and try again.

The deeper problem: no way to verify the answer

Even when ChatGPT does read your PDF, there's a second issue. It gives you an answer, but not a reliable way to check it against the document. For a casual question that's fine. For a contract clause, a policy, or a number you'll act on, "it sounds right" isn't good enough - and ChatGPT summaries can hallucinate details that were never in the source.

A different approach: read it in the browser, with citations

Instead of uploading a file into a chat, you can read and question the PDF where it already is. That's the idea behind Sidenote - an AI reading assistant that lives in your browser (Chrome, Edge and Firefox) and answers questions about whatever PDF you have open, without an upload step.

It's built to handle exactly the cases that trip ChatGPT up:

• Scanned and photographed PDFs are run through OCR first, so an image-only file becomes readable and citable.
• Files behind a login - Drive, SharePoint, a wiki - are read in place from your signed-in browser session, read-only, so there's nothing to download and re-upload.
• Long documents are indexed in full and each question retrieves the relevant passages, so there's no upload cap or token budget to manage.
• Every answer carries a citation. Click it and the document scrolls to the exact sentence and highlights it, so you verify the answer against the source instead of trusting it. Claims that can't be matched to a real passage are dropped server-side before you see them.

ChatGPT vs an in-browser PDF reader