ChatGPT Can't Read Your PDF? Here's How to Fix It

You upload a PDF, ask a simple question, and get one of these back:

• "No text could be extracted from this file."
• An answer that only covers the first part of the document.
• A confident answer that mixes up two sections, or quotes a footnote as if it were the main text.

All three come from the same place. ChatGPT doesn't look at your PDF the way you do. It pulls the text out first, then reasons about whatever came out. When that extraction step goes wrong, the answer goes wrong with it.

Here are the four usual causes, how to tell which one you have, and what fixes each.

Cause 1: The PDF is a scan, so there's no text inside

How to tell: open the PDF and try to select a sentence with your cursor. If you can't highlight individual words, or the whole page selects as one block, the page is a picture of text, not text. Scanned contracts, photographed receipts and old printed reports are almost always like this.

Why it fails: a scan has no text layer to extract. This is exactly when ChatGPT shows "No text could be extracted from this file." OpenAI's forum has long-running threads about this error.

The fix: the text has to be recognised from the image first (that's called OCR). For a page or two, take a screenshot of each page and upload the images instead. ChatGPT reads images with vision rather than text extraction. MDSpin's image to Markdown converter does the same thing and gives you editable text. Our guide to extracting text from a screenshot walks through it.

For a long scanned document, run it through an OCR tool first. Adobe Acrobat and macOS Preview's "Live Text" both work. Then use the resulting PDF.

To be upfront: MDSpin's PDF converter can't help with scanned PDFs. It reads the text layer, and it tells you when a file has none rather than handing back an empty result.

Cause 2: The document is too long for one conversation

How to tell: answers about the beginning of the document are fine, but questions about later chapters get vague, generic, or "I couldn't find that in the document."

Why it fails: every AI model can hold only so much text in view at once (its context window). A long report, a thesis or a 300-page manual can exceed what fits in one go. OpenAI also caps how much text it takes from a single file. When a document is over the limit, the model works from part of it, and the cut isn't always obvious to you.

The fix: make the document smaller, then split it if you still need to.

1. Convert it to Markdown. Raw PDF text carries a lot of extraction junk: broken line endings, repeated headers and footers, page numbers mid-sentence. In our format benchmark, clean Markdown used about a third fewer tokens than raw PDF text for the same document, so more of it fits.
2. Split it at its headings. Because Markdown keeps the document's structure (# for chapters, ## for sections), you can paste one chapter at a time and ask about each in turn.
3. Ask about one part at a time. "In section 4, what does the author conclude about X?" works far better than "summarise everything" on a long file.

Cause 3: The layout scrambles the reading order

How to tell: answers splice together sentences that don't belong together, or attribute a sidebar's claim to the main text.

Why it fails: a PDF stores text as characters placed at positions on a page. It has no idea of "paragraph" or "column". Two-column layouts, sidebars, pull quotes and headers can all come out in the wrong order when the text is pulled out, and the AI then reasons about the jumbled version.

The fix: give the AI text where the order and structure are already settled. Convert the PDF to Markdown, then skim the output before pasting it. You'll see straight away whether sections come out in the right order. MDSpin detects multi-column layouts and keeps headings as headings, which is what the AI leans on to tell sections apart.

One honest caveat: tables are the weak spot for every PDF extractor, MDSpin included. If your question depends on numbers in a table, check those rows against the original, or paste the table in yourself.

Cause 4: The file itself is the problem

How to tell: the upload fails outright, spins forever, or ChatGPT says it can't open the file.

Why it fails: password-protected PDFs, files with copying disabled, and damaged files exported from design tools can all block text extraction entirely.

The fix: remove the password (you'll need it), or open the file and use Print → Save as PDF to write a clean copy. That simple re-save fixes most damaged exports.

Quick diagnosis

What you see	Likely cause	Fix
"No text could be extracted from this file"	Scanned PDF	OCR first, or upload page screenshots
Later sections ignored or vague	Too long	Convert to Markdown, split at headings
Sentences mixed up, sidebars quoted as main text	Complex layout	Convert to Markdown and skim the order
Upload fails or file won't open	Protected or damaged file	Remove the password, or Print → Save as PDF

The same problems happen in Claude and Gemini

None of this is unique to ChatGPT. Claude, for example, analyses both text and visuals for PDFs up to 100 pages but reads only the text for longer ones (checked November 2026). So a long scanned PDF hits the same wall there.

For a full walkthrough of converting a PDF and pasting it into any of the big assistants, see how to convert PDFs for ChatGPT, Claude & Gemini.

Try it on your PDF

If your PDF has selectable text but ChatGPT still misreads it, convert it below. You'll see the Markdown the AI would get, and how many tokens it saves, before you paste anything. The preview needs no account.