Guide

How to summarize a long PDF with AI

A 200-page report is too long to read today, but a vague AI summary is worse than none. Here are the upload limits that matter in ChatGPT, Claude and Gemini, a prompt that produces a summary you can use, and a ten-minute check against the original.

Updated 30 October 2026

Quick answer: attach the PDF directly to a new chat in ChatGPT, Claude or Gemini, say who the summary is for and what you need to decide, and ask for key points with page numbers. Then spot-check three or four of those points against the pages cited before you rely on it. If the file is past a tool's limit, split it by chapter.

Limits and menu names below were checked on OpenAI's, Anthropic's and Google's help pages as of October 2026.

Step 1: Check the file fits the tool

Each assistant has different limits for documents attached to a chat:

Tool	Per-file limit	What else to know
ChatGPT	512 MB, and up to 2 million tokens of text per document	Free accounts can upload 3 files a day; others up to 80 files every 3 hours.
Claude	500 MB and up to 1,000 PDF pages, 20 files per chat	Charts and images are read only in PDFs of 100 pages or fewer.
Gemini	100 MB, up to 10 files per prompt	Google AI Pro and Ultra get a 1 million token context window, which Google puts at about 1,500 pages of text.

File size is rarely the problem with a PDF. Length is. Gemini's help page warns that very large uploads can lead to answers that miss details or connections, especially when your question depends on facts scattered across the document. If you need precision, smaller files work better in any tool.

Step 2: Attach the PDF

ChatGPT: select + in the prompt area, then Add photos & files, and pick the PDF. Attach it to the chat itself rather than relying on a copy in project files or GPT knowledge, because OpenAI notes that embedded images may not be included there.
Claude: attach the file in the chat box. A PDF over 1,000 pages is rejected with "Uploaded file is too large".
Gemini: at gemini.google.com, click Add files, then Upload Files (or Add from Drive, which needs Keep Activity on and Google Workspace connected), and click Submit.

Start a fresh chat for each document so earlier conversation doesn't color the summary.

Step 3: Ask for a summary you can act on

"Summarize this" gets you a generic paragraph. Tell the AI who will read the result and why:

"This PDF is [what it is, e.g. a 180-page vendor contract]. I'm [your role] and I need to [decision or task]. Summarize it in this order: 1) a five-sentence overview; 2) the 10 most important points for my purpose, each with the page number it comes from; 3) any numbers, dates, deadlines or obligations, as a table with page numbers; 4) anything unclear, contradictory or missing. Use only what is in the document. If something I'd expect isn't there, say so instead of filling it in."

Why this works:

The purpose tells the AI what to keep and what to drop. A lawyer and a buyer need different summaries of the same contract.
Page numbers make every point checkable.
A separate table for numbers and dates pulls out the details that are most costly to get wrong.
"Only what is in the document" discourages the AI from adding general knowledge that sounds right but isn't in your file.

Then drill down with follow-ups: "Explain section 4 in plain language", or "Quote the exact sentences about termination."

Step 4: Check the summary against the source

Spot-check three or four points. Open the PDF at the cited pages and confirm each point says what the page says. Start with the ones you'll act on.
Check every number and date in the table. These are the easiest errors to miss and the most expensive.
Ask for what's missing: "Which sections did you not cover in the summary?" Compare the answer with the PDF's table of contents.
Challenge one point: "Quote the sentence that supports point 3." If the AI can't find it, treat that point as unconfirmed.

AI summaries can leave things out or get details wrong, and Google's help page notes this gets more likely as uploads get larger. A summary is a map of the document, not a replacement for the pages you'll sign, quote or act on.

Good to know

Very long PDFs: if a file is past a limit, or answers start missing things, split it by chapter, summarize each part with the same prompt (say which pages it covers), then paste the part summaries into one chat and ask for a combined summary.
Visuals: Claude reads charts and images only in PDFs of up to 100 pages; past that it uses the text alone. ChatGPT reads both in PDFs attached directly to a chat. If a chart matters, you can upload it separately as an image.
Privacy: on personal ChatGPT accounts, whether your files help improve models depends on your data controls; OpenAI doesn't use content from business offerings such as ChatGPT Enterprise for training. Check your company's rules before uploading confidential documents to a personal account.
Saved copies: in ChatGPT, deleting a chat doesn't delete a file saved separately in your Library. Delete it there too if you want it gone.

If it doesn't work

"Upload limit reached" in ChatGPT. Check you're signed in to the right plan and wait for the retry time shown. Failed attempts can count toward the limit.
Claude says the file is too large. The PDF is over 1,000 pages. Split it into parts.
Gemini analyzed the file wrongly. Google suggests uploading it again; a smaller file with less content also helps.
The summary is vague. Add your purpose and the output order from the prompt above, and ask for page numbers on every point.

For tool-by-tool upload steps, see uploading files to ChatGPT and uploading files to Claude.

Keep the summary where you'll use it

A good summary usually turns into follow-up work. In Workbranch, answers come from Google Gemini and each point becomes a checklist item with progress, so you can tick off the points you've verified. You can highlight key lines in five colors and open a side thread on one point without losing the rest. There's a 3-day free trial.

Sources

Uploading files and audio to ChatGPT (OpenAI Help Center, checked 30 October 2026)
Upload files to Claude (Claude Help Center, checked 30 October 2026)
Upload & analyze files in Gemini Apps (Gemini Apps Help, checked 30 October 2026)

Frequently asked questions

Should I split a long PDF into parts?

Only if it is past a tool's limit or the answers start missing details. Splitting by chapter and summarizing each part, then asking for a summary of the summaries, keeps each pass small. Tell the AI which part it is reading so page references stay correct.

Does the AI read charts and tables in the PDF?

ChatGPT says it can analyze text and images in PDFs attached directly to a chat. Claude reads visuals only in PDFs up to 100 pages; for longer ones it reads the text alone. If a chart matters, ask about it directly or upload it as an image.

Can it summarize a scanned PDF?

A scan is a set of page images rather than text. Results vary, so check the summary especially carefully, or run the file through a text-recognition (OCR) tool first.