Skip to content

ChatGPT cannot read your scanned PDF

Error messages explainedLast checked

Short answer

Your PDF has no text in it. It is a photograph of pages. Every ChatGPT plan except Enterprise extracts digital text and discards images, so a scan leaves nothing behind. Run the file through OCR to create a real text layer and it behaves like any other document.

This is documented behaviour, not a bug

OpenAI's File Uploads FAQ addresses images inside documents directly. Visual retrieval for PDFs is a ChatGPT Enterprise feature. For everything else:

All other plans and document files only support text-based retrieval. This means that ChatGPT will extract digital text from the file and discard any images.

Apply that to a scan. A scanned page contains one thing: an image. Extract the digital text, of which there is none, and discard the images, which is all of it. Nothing remains.

The upload succeeds. That is what makes it feel broken rather than expected.

The two second test

  1. Open the PDF

    Any reader.

  2. Try to select a sentence

    Click and drag across some words.

  3. Read the result

    Words highlight, one at a time, following your cursor? Real text, and your problem is something else. A rectangle draws over the page instead? It is an image, and this page is for you.

Fixing it

OCR converts the picture of the words back into words. A few options depending on what you have.

Mac, no extra software. Open in Preview and export as PDF. Recent macOS versions add a text layer automatically. Test it afterwards with the selection check above.

Adobe Acrobat. Scan and OCR, then Recognise Text. The most reliable option and it handles poor scans better than most.

Microsoft OneNote or Word. Insert the image or PDF and use the text recognition, then copy the result out.

Free web tools. Plenty exist and most work fine.

Not for anything confidential

A free OCR site means uploading your document to a company you know nothing about. For contracts, medical records, anything under an NDA, or anything with personal data in it, use software on your own machine. The two minutes saved is not worth it.

Check the OCR worked before uploading

Run the selection test again on the new file. If words highlight, you have text. If you still get a rectangle, the OCR did not take and uploading will fail the same way.

Also skim the result. OCR on a poor scan produces errors, and a document full of mangled words gets you confidently wrong answers rather than obviously broken ones. Faded print, handwriting, and skewed pages are where it struggles most.

Once it has text

It becomes an ordinary document, with the ordinary limits: 512 MB of size and 2 million tokens of length. A long scanned book will be past the length cap after OCR, so you may go straight from one problem to the next.

If OCR is not an option

Two fallbacks.

Screenshot the pages that matter and upload them as images. Images are supported, capped at 20 MB each, with a ceiling of about 100 per conversation. Fine for a few pages, impractical for a few hundred.

Type out the part you need. Faster than it sounds when it is one clause of a contract, and it gives you a clean text you can be certain about.

Common questions

Why does ChatGPT say my PDF has no readable text?

Because it does not. Every plan except Enterprise extracts digital text from a document and discards the images. A scanned page is a photograph, so there is no digital text to extract and nothing comes back.

Will a paid plan fix it?

Only Enterprise, which supports visual retrieval for PDFs. Plus and Pro behave exactly like Free here, so upgrading for this specific reason will not help.

How do I know if my PDF is scanned?

Try to select a sentence with your cursor. If it highlights words, there is real text. If it draws a rectangle over the page, it is an image and needs OCR.

What is OCR and do I have to pay for it?

Optical character recognition converts a picture of words back into words. Preview on a Mac and Adobe Acrobat both do it, and free web tools exist. Do not use a web tool for anything confidential, since you are uploading the document to a stranger.

Keep reading