Guide

How to Feed Multiple PDFs to Claude or ChatGPT (Without Hitting the Limit)

A 5-minute read

You've got five papers and two book chapters open, you want the AI to work across all of them at once — and the app tells you that's too many files, or the upload is too big, or it only seems to "read" the first couple. Working from multiple PDFs is where these tools get genuinely useful, and it's also where they push back hardest.

The short version: don't hand over the raw files. Pull the pages that matter from each one, combine them into a single small packet, and keep the filenames so you can still tell the sources apart. Here's why, and how.

Why you hit the wall

Two limits stack up when you add files:

• A cap on how many files you can attach. The chat apps limit attachments per message, so a stack of separate PDFs gets rejected before the model even looks at them.
• A cap on how much it can actually read. Even if the files upload, everything has to fit in the model's context window. Several full documents rarely do — so it silently reads part and ignores the rest.

Why "just zip them" doesn't work

The obvious workaround is to zip everything into one archive and upload that. It usually backfires: the model can't reliably open a zip and pull the documents out of it, so you either get an error or a vague answer based on whatever it managed to extract. One archive isn't the same as one readable document.

The approach that works: trim, then combine

Instead of shrinking the number of files, shrink what's inside them:

• Take only the relevant pages from each source. You almost never need all 40 pages of a paper — you need three. Pulling just those is what makes several sources fit at once. (Same idea as when a single PDF is too long.)
• Combine them into one small packet — or a handful of trimmed PDFs, well under the file cap.
• Keep the original filenames. This is the part people skip, and it matters: the filename is how you and the model tell the sources apart. Merge five papers into one nameless blob and every citation collapses into "the document" — useless.

Keep citations straight across sources

With several sources in play, be explicit about attribution in your prompt:

These pages come from several sources; the filename marks each one. When you cite, name the source it came from and the page number printed on the page. If two sources disagree, point out the disagreement rather than blending them. If a claim isn't in the sources, say so.

That keeps the model from quietly merging five papers into one averaged-out answer — and it means every claim traces back to a specific source you can check. (More on stopping invented page numbers here.)

In one line

Don't fight the file limit with a zip. Trim each source to the pages that matter, combine them into one small packet with the filenames kept, and tell the model to cite each source by name. Several books' worth of research suddenly fits in a single conversation.

Related: When ChatGPT says your PDF is too long · Why AI invents page numbers.