How to Analyze a PDF with Claude: A Complete Walkthrough

Summarising a fifty-page report, pulling the numbers out of a study, checking a single clause buried in a contract - these are exactly the jobs Claude is good at, provided you ask in the right way. This guide walks through how to get useful answers out of a PDF, which prompts actually work, and what to watch for when a document is long, scanned, or stuffed with tables.

What Claude can actually do with a PDF

When you attach a PDF to a conversation, Claude does more than skim the raw text. It also reads the layout and the visual elements - charts, diagrams, tables, even scanned pages. In practice, that means you can ask it to:

• summarise the whole document or just one section you care about;
• answer specific questions whose answers live somewhere in the text;
• pull out data such as key figures, dates, names and amounts;
• turn a table inside the PDF into clean Markdown or a reusable CSV;
• compare several documents attached to the same chat;
• rephrase or translate a passage into a different tone or language.

The difference from a plain "Ctrl+F" search is real. Claude understands what the document means, so you can ask "what are the risks mentioned in this contract?" without knowing the exact wording the author chose.

Getting a PDF in: the three-step routine

On claude.ai, importing a file takes three small actions:

1. Open a fresh conversation (or an existing project).
2. Click the attachment icon, or simply drag and drop the file into the message box.
3. Send a clear instruction with the file in that very first message, rather than firing off the PDF on its own.

That last point matters more than it looks. A PDF dropped in with no instruction gives you a generic summary. The same PDF sent with "list the supplier's obligations mentioned in this contract, with the page number for each one" gives you something you can use straight away.

Size, pages and the limits worth knowing

The exact limits - file size, page count, how many attachments per conversation - depend on your plan and shift as the product is updated, so check Anthropic's official site if you are unsure. The principle to remember is simpler: an attached document eats into the context window, the model's working memory. The longer the PDF, the less room is left for the back-and-forth of the conversation.

For a genuinely large document, three approaches tend to work well:

• Split it up into thematic chunks and handle them one at a time.
• Point Claude at the right part by naming the sections you care about so it focuses its reading there.
• Summarise in layers - have each part summarised, then ask for a summary of the summaries.

The prompts that make the difference

The quality of the analysis depends directly on the instruction you give. A few patterns that earn their keep, ready to adapt to your own document:

• "Summarise this report in 10 points, and cite the page each point comes from." The page request lets you verify every claim.
• "Extract every financial figure mentioned, as a table with three columns: amount, purpose, page."
• "What questions should a sceptical reader ask after finishing this document?" - handy for prepping a meeting.
• "Compare the conclusions of these two PDFs and flag where they disagree."
• "Reproduce the table on page 12 in Markdown, without changing anything."

Four concrete cases

In all four, the winning reflex is the same: describe the output format you want - table, list, word count - on top of the content you are asking for.

Scanned PDFs and messy layouts

A scanned PDF is really just a stack of images. Claude can read those images if the scan is clean: sharp text, decent contrast, pages that sit straight. On a poor scan - a crooked photocopy, handwriting, overlapping stamps - reading errors climb quickly.

Two precautions help here. First, run the document through OCR (optical character recognition) before importing, so you have a PDF with a reliable text layer. Second, double-check the critical data - amounts, dates, references - page by page, because on a scan it is easy to mistake a 3 for an 8.

The same care applies to complicated layouts: multiple columns, footnotes, merged table cells. Claude usually copes, but flag the difficulty in your prompt ("this document is in two columns") to help it rebuild the reading order.

Working with the same documents over time

If you keep querying the same files - an internal knowledge base, a reference manual, product documentation - reattaching the PDFs every time gets tedious. Projects solve this: you drop your documents in once, and every conversation in the project can reach them. It is also the right place to set standing instructions, such as "always answer by citing the source document."

The limits to keep in mind

Three habits prevent most of the disappointments. The risk of hallucination never fully disappears - even with the document in front of it, Claude can misread an ambiguous passage, which is exactly why paginated citations are your safety net. Confidentiality comes first, so don't import sensitive documents without checking the privacy policy that applies to your plan, especially at work. And remember that Claude analyses while you decide: for a contract or legal document its work prepares the ground but replaces neither a human review nor professional advice.

In short

Analysing a PDF with Claude comes down to four reflexes: import the document with a precise instruction, insist on paginated citations, describe the output format you expect, and verify the critical data. Build those habits and Claude becomes a genuine research assistant - for monitoring, contract review, or meeting prep.