Claude's document superpower is comprehension, not search. Upload a document and Claude holds the entire thing in context at once. Not excerpts, the full document. You can ask questions that require connecting page 3 and page 41 in the same answer.
The mistake most people make: asking for generic summaries. "Summarize this contract" gets you a watered-down version of the document you already have.
The better approach: targeted questions that extract specific intelligence. Not "summarize this contract" but "find every clause that could cost me money if I miss a deadline." Claude doesn't get tired reading a 200-page PDF. It reads every word with the same attention. Your job is to ask the right questions, not to read the document yourself.
Targeted Extraction Prompts
Copy-paste patterns for the most common document tasks:
Contract review:
"Review this contract. Find: 1) auto-renewal clauses and cancellation notice periods, 2) liability caps that favor the vendor, 3) who owns data and IP, 4) non-standard clauses. Quote the exact clause language for each issue."
Proposal comparison:
"Build a comparison table of these three proposals: total cost, timeline, SLA guarantees, support hours, exit terms, red flags. Use the exact numbers from each."
Meeting notes:
"Extract all action items: owner, deliverable, deadline. Format as a table. Flag items with no owner or deadline."
The pattern: specify what to find, what format to return, and how to handle edge cases. Structure in gets structure out.
Cross-Document Analysis: The Power Move
Single-document analysis is useful. Cross-document analysis is where Claude becomes irreplaceable. Upload three quarterly reports and ask what trends emerge that no single report shows. Or a vendor contract plus your standard terms: where does it deviate, and in whose favor?
"I've uploaded [number] documents. Identify: 1) information that appears in all of them, 2) information that contradicts between them, 3) significant information that appears in only one."
For trends:
"Compare these 3 quarterly reports. Which KPIs improved, declined, or show volatility? For any metric that moved more than 15% in either direction, quote the exact numbers from each report."
Tell Claude what to look for across documents. It executes the comparison framework you specify.
Workflows by Document Type
PDFs and research papers: ask for the core argument in three sentences and the three most important tables or figures, then drill into the sections that matter: "What is the methodology? What limitations do the authors acknowledge? Does the conclusion match the data?"
Contracts: ask Claude to flag non-standard clauses first. Standard language is background noise; the deviations are where the risk lives. Then drill into each flag.
Financial documents: extract the numbers into tables before asking for analysis. "Build a table of revenue figures, time periods, and page references. Then tell me what story the trend tells."
Transcripts and meeting notes: "Separate this into: decisions made, commitments given, open questions, next steps." Those categories get conflated constantly.
Upload documents and ask targeted extraction questions, not generic summaries. Use the prompt patterns for contracts, proposals, reports, and meeting notes. Cross-document analysis, uploading multiple files and asking comparative questions, is where Claude's document analysis becomes genuinely irreplaceable.