Many users limit Claude to conversation. Yet one of its most profitable uses is analysing your own documents. Here is what it can read — and how to make the most of it.
The formats Claude works with
- PDF — contracts, reports, invoices, presentations. Ideal for summarising, extracting key points or comparing versions.
- Spreadsheets (Excel / CSV) — to have data commented on, spot trends or write up the analysis of a report.
- Word and text documents — proofreading, rephrasing, getting the format right.
- Images — screenshots, diagrams, photos of documents: Claude can read and describe their content.
- Code and technical files — to explain, document or debug.
Format by format: strengths, limits and a sample prompt
| Format | What Claude does well | Limits to know | Sample prompt |
|---|---|---|---|
| Excel / CSV | Commenting on data, spotting trends and anomalies, writing up the analysis of a report | Very large or multi-sheet files: prefer a targeted export of the useful range | "Analyse this sales file: the 5 main trends, the anomalies, and a one-page commentary for management." |
| Summarising, extracting clauses, amounts and dates, comparing versions | A scanned (image) PDF reads less well than a native PDF with selectable text | "Summarise this contract in 10 points and list the clauses a lawyer should check." | |
| Word / text | Proofreading, rephrasing, harmonising the tone, getting the format right | Complex layouts (columns, text boxes) can be flattened on reading | "Review this document: correct it, harmonise the tone and flag ambiguous passages." |
| Images | Reading screenshots, diagrams and photos of documents; describing and extracting content | Small fonts, blurry or badly framed photos degrade reading | "Here is a screenshot of a dashboard: list what to remember and what to double-check." |
| PowerPoint | Summarising a deck, critiquing the structure, proposing a new outline | Purely visual elements (charts as images) are used less well than text | "Here is our sales deck: summarise the through-line and propose 3 structural improvements." |
The exact limits — maximum file size, page count, accepted formats — depend on your plan and change regularly. The up-to-date reference is Anthropic's official help centre: support.claude.com.
3 uses that save hours
1. Summarise a long document
Drop in a 40-page PDF and ask for the 10 key points + the items to check first. You go from an hour of reading to a few minutes.
2. Turn data into a story
From an export of figures, ask for a clear commentary for management: Claude writes the analysis, you keep control of the numbers.
3. Extract structured information
"Extract from this document every amount, date and name, as a table." Perfect for preparing data entry or follow-up. It is also the building block for more ambitious work, such as having the forecast of a business plan written with Claude commented on.
The rule: let Claude do the rough work, and keep the final validation of numerical and legal data for yourself.
Preparing your files for better results
The quality of the answer depends heavily on the quality of the input file. Three reflexes before uploading a document:
- Clean up the spreadsheet. One clear header row, one piece of data per column, no merged cells and no subtotals inserted in the middle of the data. Five minutes of cleaning prevents half of the approximate answers.
- Say what you're looking for. "Analyse this file" produces a general answer. "Compare margins by region and explain the Q2 gap" produces a useful analysis. Also state the audience and the expected format.
- Split large documents. For a 200-page report, work chapter by chapter, or send the table of contents first and ask Claude which sections to analyse as a priority.
When it goes wrong: the classic blockers
- File too big. Export only the useful range or columns, or split the file into several parts. A targeted extract almost always gives a better analysis than a complete raw file.
- Scanned PDF. If the text isn't selectable, reading is less reliable. Run the document through an OCR tool, or get the native version from the sender when possible.
- Merged cells and nested tables. They break the row-and-column structure Claude interprets. Unmerge and flatten the table before sending.
- Multi-sheet workbook. State explicitly which sheet you care about, or export it alone as CSV to remove any ambiguity.
- Approximate answers on numbers. Ask Claude to cite the cell, row or page each key figure comes from: errors become immediately visible.
Two important precautions
- Confidentiality. Don't upload sensitive data without knowing the processing framework — we cover the topic in detail in our article on Claude, GDPR and company data. In the training, we show you how to work cleanly and GDPR-compliant.
- Verification. On figures or clauses, always cross-check. AI speeds things up; it doesn't excuse you from reviewing.
Finally, if you're still hesitating between tools for this kind of document work, our Claude vs ChatGPT comparison explains why document analysis is precisely the ground where Claude stands out.
Capabilities and file-size limits evolve. Check the up-to-date terms of your plan at support.claude.com for technical details.
Use your files like a pro
In the training, we build your analysis assistants on your real documents.
Take my free assessment