Claude Can't Read Your PDF? Causes and Fixes (2026)

Claude says it can't read your PDF, or its answers skip half the file? The page and size budgets, dense scans, encryption and links behind it, and the fixes.

Lewis Hadden5 min read

Claude reads a PDF by turning every page into an image and extracting its text, then working from both. So when Claude can't read your PDF, the cause is almost never the file being a scan - it's usually the page or size budget, a dense document filling the context early, a password, or a link instead of a file. Each has a quick fix, and there's one point that surprises people: on scanned pages, Claude is one of the better tools.

Cause 1 - The PDF is over the size or page budget

Anthropic publishes hard limits for PDF processing: 32 MB per request and 100 pages per request (600 pages when the request has a 1 million token context window). The limit covers the whole request, so a 90-page PDF plus a long prompt can still tip over.

The fix: check the page count before uploading. Split a long PDF into sections, or extract only the chapters you need, and ask about each part in turn.

Cause 2 - The pages are dense, so the context fills up early

Because each page is processed as an image and as text, every page costs context, and dense pages - small fonts, big tables, image-heavy scans - cost more. Anthropic's own documentation notes that dense PDFs can fill the context window before the page limit is reached. The symptom is an answer that only reflects the first part of the file.

The fix: split dense documents into smaller chunks, or drop pages that are pure images. If the document is a scan, running it through OCR first and uploading the searchable copy gives Claude clean text to work from - see how to make a PDF searchable.

Cause 3 - The PDF is password-protected

Anthropic's PDF support requires a standard PDF with no password or encryption. A locked file fails outright.

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

A link to a PDF in Google Drive, SharePoint, a wiki or any page behind a login isn't something Claude can open. Even public links are unreliable compared with uploading the file itself.

The fix: download the PDF and upload it directly, or paste the relevant text into the chat. Better still, use a reader that works on the page you already have open, so there's nothing to download and re-upload.

Cause 5 - The file isn't a standard PDF

Renamed images, PDF portfolios, forms built on XFA data and files exported by unusual tools can trip the parser even when they open fine in a viewer.

The fix: print the document to a fresh PDF from your viewer (Print, then Save as PDF), which flattens it into a standard file, and upload that copy.

The deeper problem: reading all of it, and proving it

Even when the upload works, two things remain. A long document has to fit in one conversation, so you're always managing budget. And while Claude can quote passages, checking that a specific claim really matches the source still means opening the PDF and finding the page yourself.

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

Sidenote is 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 for the cases that trip Claude up:

  • Long documents are indexed in full, and each question retrieves the relevant passages, so there's no page budget to manage.
  • Scanned PDFs are run through OCR automatically, 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 session, read-only.
  • Every answer carries a citation. Click it and the document scrolls to the exact sentence and highlights it; claims that can't be matched to a real passage are dropped server-side before you see them.

Claude vs an in-browser PDF reader

ClaudeIn-browser reader (Sidenote)
Page and size budget100 pages and 32 MB per requestIndexes the whole document
Scanned PDFsReads page images, within the budgetOCRs automatically before answering
Files behind a loginDownload and uploadReads in place, read-only
Verify an answerQuotes, checked by handClick a citation to the exact passage
Best forReasoning and drafting around a documentReading and questioning long, real documents

Claude is an excellent thinking partner once the document is in front of it. The friction is getting a long, scanned or private PDF in front of it in the first place, and that's the part an in-browser reader removes. The same pattern holds for Gemini and Copilot.

Frequently asked questions
Usually, yes - and better than most chat tools. Claude converts each page to an image and analyses it alongside any extracted text, so a scan with no text layer can still be read. The limits are the page and size budgets: a long or dense scan fills the context window quickly, and pages past the budget are never seen.
Anthropic's published API limit is 100 pages per request, or 600 when the request has a 1 million token context window, with a 32 MB cap on the whole request. The Claude app applies its own per-file and per-chat caps, listed in Anthropic's help centre. Dense pages can exhaust the context before the page limit is reached.
The document filled the context window before Claude reached the end, so later pages were never read. Split the PDF into sections, remove pages you don't need, or use a reader that indexes the whole document and retrieves the relevant passages for each question.
No. Anthropic's PDF support requires a standard PDF with no password or encryption. Remove the password in your PDF viewer (if you're allowed to), save an unprotected copy, and upload that.
All guides
Ready when you are

Stop digging. Start asking.

Add Sidenote to your browser, open any page in your wiki, and ask it the question you’ve been Slacking the team about.

7-day Pro trial · No card required · Free plan forever