Can ChatGPT read EPUB files? Yes, with 2 catches

What an EPUB actually is

Rename an EPUB to .zip and open it. Inside you will find a folder of XHTML files, one per chapter or thereabouts, plus a stylesheet, a manifest and an images folder. That is the whole format. An ebook is a small website in a box.

This matters because it tells you exactly what ChatGPT will do with it. It extracts digital text and discards presentation, and an EPUB hands over its text without much of a fight. OpenAI publishes no list of extensions, only the statement that it accepts "All common file extensions for text files, spreadsheets, presentations, and documents." An ebook is a document in common use, so it sits inside that, and uploads generally work.

Generally, not always. The container holds more than your chapters: a table of contents file, cover metadata, sometimes a per-chapter split that produces odd ordering once flattened. So an EPUB upload can come back with boilerplate mixed into the text, or with chapters in a sequence that does not match the book. Converting first removes that whole class of problem.

DRM is where it stops

This is the real reason most people's EPUB will not work, and no amount of fiddling changes it.

A book bought from Amazon is not an EPUB. It is AZW3 or KFX, Amazon's own formats, readable only by Kindle software. Books from Apple Books, Google Play Books and Kobo are usually EPUBs wrapped in DRM, which means the text inside is encrypted. A library loan through Libby or OverDrive is the same thing with an expiry date attached.

In all of those cases the file on your disk contains no readable text. Not for ChatGPT, not for a converter, not for anything except the app that was issued the key. There is no setting to change and no upload trick that helps.

Converting an EPUB properly

If you have a DRM-free file, converting takes a minute and makes everything downstream more predictable.

1. Open it in Calibre

   Free, runs everywhere, and it is the standard tool for this. Add the book, then choose Convert.

2. Convert to TXT

   Plain text is what ChatGPT works from anyway. You lose fonts and layout, which were being discarded regardless.

3. Split it by chapter

   For anything book length, save each chapter as its own file rather than one long dump. The reason is in the next section.

If you would rather not install anything, rename the file to .zip, unzip it, and open the XHTML files in the text folder. It is uglier and it works.

Converting an ebook to PDF is the common instinct and the weakest option. You are adding page breaks, headers and pagination to something that had none, and getting a bigger file that reads no better.

The length problem nobody mentions

Here is the part that surprises people. The size limits are not what stops you.

A typical novel runs 80,000 to 120,000 words. As plain text that is a file of a few hundred kilobytes, nowhere near 512 MB, and its token count is a small fraction of 2 million tokens. The upload will be accepted without complaint.

Then you ask a question about chapter fourteen and get an answer that is clearly built out of the first fifty pages. Nothing failed, nothing warned you. Quality falls off with length long before any published cap is reached, and the drop is silent. Textbooks, reference works and omnibus editions can be long enough to hit the token cap as well, at which point the tail of the book is simply not there.

The fix is unglamorous. Feed it a chapter at a time, keep a running set of notes you re-paste as you move forward, and check what it actually read by asking it to name the last section it saw.

Fixed-layout ebooks read as blank

Comics, children's books, cookbooks and some illustrated textbooks ship as fixed-layout EPUB, where each page is an image with the words baked into it. The container looks identical from outside.

ChatGPT handles document files as text only, except Enterprise, extracting digital text and discarding images. A fixed-layout ebook has no digital text to extract, so you get an empty or nearly empty read, with no error explaining why.

Test it in two seconds: open the book in any reader and try to select a word. If your cursor will not grab text, there is no text, and you are looking at an OCR job rather than an upload problem.

What to do

If the book is DRM-free, convert it to plain text, split it by chapter, and work through it in order. If it came from Kindle or a library app, the file is closed to everything but its own reader and no conversion route exists. And if the pages are pictures, no upload of any kind will help until the words have been recovered from the images.