ClaudeSummarizationLevel 2: Context

How Do You Summarize a Very Long PDF With Claude?

Attach the PDF directly to a chat and ask for the central argument first, then a section-by-section summary. Anthropic's limits decide the rest. Claude reads text and visuals for up to 100 pages. From 101 to 1,000 pages it reads text only, and above 1,000 the upload fails. Split long PDFs if the charts matter.

Prompt to copy
<document>
[attach the PDF]
</document>

Before summarizing, tell me how many pages you can see, and whether you can read the charts and tables in this file, or text only.

Then:
1. In one sentence, state the document's central argument or purpose.
2. For each major section, give its page range, its point in 2 or 3 sentences, and the key figures, with page numbers.
3. Write a brief of up to [length] for [reader] that follows the argument from step 1, keeping only what supports or challenges it.
4. List any places where sections contradict each other, and anything you couldn't read.

Quote the source when you give a figure. Don't estimate figures you can't see.
  • Chat, text and visuals100 pages
  • API, per request600 pages
    100 if the window is under 1M
  • Chat, text only1,000 pages
Maximum PDF pages. Sources: Anthropic's upload help page and PDF support docs, checked 6 November 2026. In chat, charts are read only up to 100 pages.
  1. 1Split at chapters
  2. 2Summarize each
  3. 3Merge and reconcile
  4. 4One brief
A rule of thumb, not an Anthropic chart. Use it for documents past the 100-page mark, where charts stop being read in chat.

Try it now

  1. 1Check your PDF's page count. If it's over 100 and the charts matter, split it at chapter breaks into files of 100 pages or fewer, using any PDF tool.
  2. 2In a new chat, click +, choose Add files or photos, and attach up to 20 parts.
  3. 3Paste the prompt and read Claude's page count first, to confirm it saw the whole document.
  4. 4You'll know it worked when the summary cites figures from charts, and you can find two or three of them on the pages it names.

Use it when

Annual reports, academic books, long contracts, government reports, and transcripts of a whole conference. Use the split approach when charts and tables carry the argument, as in a financial report or scientific paper.

Skip it when

When you only need one fact from a large document. Ask for it directly, with a quote. And a section-by-section summary is the wrong choice for a document you must fully understand, such as something you're signing. Use it as a reading map rather than a replacement.

Why it works

Anthropic's upload page sets the limits: 500MB per file, 20 files per chat, and up to 1,000 PDF pages. The trap is easy to miss. A 300-page report is accepted, but Claude processes only its text, so a summary can miss numbers that appear only in charts. Other document types such as DOCX give text only, at any length.

Where you upload matters too. A chat file goes straight into the conversation. Project files are limited to 30MB each, and on paid plans a large project switches to retrieval, where Claude searches for relevant pieces rather than loading everything. That's fine for questions. Rule of thumb, not a vendor figure: for a whole-document summary, use a chat. Anthropic's long-document advice applies in both: place the document first, and for over 20,000 tokens, ask for quotes.

Example

Say you have a 320-page annual report, and the segment results are mainly in charts. Uploaded whole, it would be read as text only. You split it into four files of 80 pages each at its chapter breaks, and attach them all to one chat. Now Claude can read the visuals in each part. You ask for the central argument, then a section summary of each file, then a merged 300-word brief. Contradictions between the CEO letter's claims and the segment tables are worth your attention, so step 4 asks for them.

Going further

On the API, the limits are different. Anthropic's PDF docs give 128MB per request and 600 pages, or 100 when the request's context window is under 1M tokens. It adds that dense PDFs can fill the context window before reaching the page limit, and suggests splitting them. Place PDFs before text in your request, and refer to pages by their PDF viewer numbers.

In chat, a PDF over 30MB can be processed in Claude's computing environment without being loaded into context, Anthropic's file creation page says. That's handy for extraction jobs, but ask Claude which method it used, since it affects what it read.

Common mistakes

  • Uploading a 300-page PDF and asking about its charts. Beyond 100 pages Claude reads text only, so ask what it can see before you rely on the answer.
  • Using a Word file with key figures in images. Anthropic says non-PDF documents give text only, so export to PDF first.
  • Adding the book to project knowledge for a whole-book summary. With retrieval on, Claude searches for relevant pieces rather than reading all of it.
  • Merging summaries without asking for contradictions. A smooth final brief can hide that chapter 2 and chapter 9 disagree.

Sources

Checked 6 November 2026. Products change often, so confirm details with the vendor.

Related

Beyond the course