How to use Claude to analyze documents: from useless summaries to hours-saving extractions

Roberto runs a labor agency. Every month you receive dozens of PDF payrolls and invoices from your clients. Some are born in modern programs. Others come scanned, crooked, with stamps above the text and pen notes in the margins. For years his routine was always the same: print, underline, write down by hand and then upload what was important to Excel. Six hours a month were spent just transferring data from one site to another.

When he heard that Claude “read PDFs,” he thought he could finally get that job off his back. He uploaded an invoice and asked, “Summarize this invoice for me.” The result was correct. And completely useless for what I needed. It told him what the invoice was for, who issued it and the total amount. Everything that Roberto already saw with the naked eye. He tried again with several. More of the same.

The problem wasn’t Claude. The problem was the question. Roberto didn’t need anyonewould understandthe invoice. I needed someoneextract specific datawithout inventing anything. When that changed, the six hours started to disappear.

Reading is not the same as analyzing

When you upload a document to Claude, several things happen at the same time. Claudereadthe text it can detect. Thentry to understandthat text according to what you ask. And, if you’re not careful,inferswhat’s missing to give you a “nice” answer.

Here is the fine cut. For serious work, you want the first and the second. Never the third. This is why asking for “summaries” is usually a bad idea in business documents. A summary invites you to fill in the gaps. A well-defined extraction does not.

The difference between using Claude toanalyze documentsas a summarizer or as an analyst is not in technology. It’s in how you ask for what you need.

What formats can Claude read (and with what limits)

Claude can work with several common file types: PDF with real text, scanned PDF, images (JPG, PNG), Word documents and spreadsheets. That doesn’t mean they all read equally well.

A PDF generated by an accounting program usually reads almost perfectly. A scanned PDF depends on the quality: skew, resolution, smudges, stamps. Images with small or poorly contrasted text generate more errors. Claude does what he can, but he doesn’t see like you.

That is why it is key to ask him to tell youwhat have you not been able to read. That phrase alone transforms an extraction into a reliable tool: if something is missing, it appears as a lack, not a guess.

The analyst rule: don’t ask for summaries, ask for extractions

The change that saved Roberto time was this: he stopped asking “summarize the bill” and started asking:

“Extract these fields and if any are not readable, say so explicitly.”

That transforms Claude from editor to analyst. A good extraction has three characteristics:

• Specific fields: not “the important data”, but “invoice number, tax base, VAT, total, date of issue, NIF of the issuer.”
• Fixed format: table, structured list, JSON. Whatever allows you to copy and paste without touching.
• Explicit prohibition on inventing: If something is not clear, let them know. Not that I guess.

So, if something is missing, it appears as missing. Not as an assumption. The difference between “not legible” and a made-up number can be the difference between a well-recorded invoice and an accounting error.

Verifying is also part of the job

When you work with documents, it’s not enough to get a pretty table. You need to know how reliable it is. A good habit is to always ask for a final line:

“List the fields or pages that you could not read clearly.”

If Claude tells you that a tax base is not readable, you know you have to review that file. If he doesn’t say it, you can move forward with peace of mind. Verifying is not distrusting AI. It is to use it professionally. A reliable employee doesn’t just deliver results: he tells you where he isn’t safe.

The extraction prompt that works

Un effective promptTo analyze documents with Claude, follow the structure ROLE · CONTEXT · TASK · FORMAT:

• Role: accounting analyst specialized in billing.
• Context: small customer invoices, some scanned with low quality.
• Task: extract specific fields (number, date, base, VAT, total, NIF, supplier).
• Format: table with one row per invoice. If a field is not readable, write “not readable” and do not invent.

This prompt converts one hour of manual work into three minutes of supervised extraction. And most importantly, it produces results you can audit.

When you should NOT upload a document

There are documents that, although technically they can be uploaded, you should not do so without thinking about it. Confidential contracts. Health data. Sensitive tax information of third parties. In those cases, the answer is not “don’t use AI.” It’s “choose where.” Not all readings need to be done in the cloud. Later you will see alternatives with local models for the sensitive. For now, stick with one idea: every document you upload is a decision. Make her conscious.

---