Skip to content
FlowArray help

Documents as inputs: PDF, Word, Excel

Give each item a document — a PDF spec sheet, a Word brief, an Excel price list — and a Claude or Gemini step reads it.

In this article (4)

A lot of real work starts from a document. Add a file column to a batch and each item can carry a PDF, a Word document, an Excel workbook or a text file; a Claude or Gemini step then reads it directly — "write three listings from this spec sheet", "pull every SKU and price from this catalogue page as JSON", "turn this brief into a shot list" — across the whole batch.

Setting it up

  1. In Map view, add a column of type file to the batch card (the Spec sheet to listings template already has one).
  2. In the items grid, drop a PDF, Word (.docx), Excel (.xlsx), .txt, .md or .csv into each row's file cell. PDFs show their page count as soon as they're uploaded.
  3. Use a Claude or Gemini text step and refer to the document in the instruction — "Read the attached spec sheet and…".

Both read every PDF page as text and as an image, so tables, charts, diagrams and product photos count, not just the words. Text files are read as text. Gemini image models can take text and Word/Excel files but not PDFs — read the PDF in a text step first and pass its text on.

Word and Excel

Word and Excel files are turned into text the moment you upload them, and that text is what the step reads:

  • Word keeps headings, lists, bold, links and tables.
  • Excel keeps every sheet, each one labelled with its name, as rows of values — the results of formulas, not the formulas, and dates written out as dates. Hidden sheets are included and marked hidden.
  • Pictures, charts, colours and layout are left out. If how the document looks matters — a designed brochure, a chart someone needs to read — save it as a PDF instead; Claude then sees every page.
  • Save old .doc / .xls files and files with macros as .docx / .xlsx first. Password-protected files are refused — remove the password.

Limits

  • PDFs: up to 25 MB and 100 pages each. Password-protected PDFs are refused at upload — remove the password first. Split anything bigger.
  • Text files: up to 500 KB.
  • Word and Excel: up to 10 MB each, and at most 500 KB of text once converted (about 170,000 tokens). Workbooks: up to 20 sheets and 50,000 rows. Split anything bigger, or delete the parts the step doesn't need.
  • Claude and Gemini steps read files. A step on another model (or a video step) fails straight away with a clear message, before anything is spent — it never answers a question about a document it didn't see.
  • CSV import: a file column becomes a <name>_file_url column; put a public link to the file in it.

What it costs

PDFs are charged by the page: roughly 1,500–3,000 tokens of text per page, plus the page image. As a guide on Claude Sonnet 5.5, a 10-page brochure is about $0.10 and a 100-page catalogue about $1; Haiku 5.5 is around a twentieth of that. Gemini reads a PDF page for about 258 tokens, so the same 100-page catalogue costs cents on Gemini Flash. Word and Excel files are charged as text only, which is usually far cheaper than the same document as a PDF. Set the step's Speed to Economy to halve it for a big stack, and your spending limits apply as always.