Convert a Spreadsheet to Text for an LLM

You have a workbook and a question about it. The chat box takes text, so the spreadsheet has to become text first, and that is where most of the answer quietly goes missing. A spreadsheet is not a document that happens to hold numbers. It is a grid, and the grid is half the meaning.

This guide covers converting a spreadsheet to text for an LLM prompt: what a flattened read destroys, what actually reaches the model, and the two things a workbook always loses on the way out. There is a spot to try it on your own files partway down. Nothing is uploaded to us; the extraction runs in your browser.

A row is only data while its cells stay apart

Take one row of a revenue sheet: the region, then two quarterly figures. Read it with the separators intact and it is three values. Read it without them and it is one long number that nobody, model or person, can take back apart.

RegionQ1Q2
EMEA12001350
APAC9801105

# Sheet: Revenue
Region,Q1,Q2
EMEA,1200,1350
APAC,980,1105

EMEA12001350 is not merely harder to read than EMEA,1200,1350. The two figures cannot be recovered from it at all, and neither can the boundary between them. A model given the left column will still answer your question; it will just answer it about numbers that were never in your sheet.

The same applies one level up. A workbook with a Revenue tab and a Headcount tab is two tables, and if the tab names go missing they arrive as one undivided block with nothing saying which figures belong together.

What reaches the model

Each sheet arrives under its own name, and each row keeps its cells separated in column order. That is the whole trick: the model can see which column a number sat in and which table it belonged to, which is what any question about a spreadsheet turns on.

1. 1

   Drop the workbook

   One file or a folder of them. Spreadsheets go in alongside PDFs, Word documents, slide decks, and any plain notes or markdown you drop with them.

2. 2

   Every sheet is read in your browser

   Each tab is read separately and written under its own name, so a multi-sheet workbook keeps its boundaries instead of merging into one block.

3. 3

   Copy one bundle

   You get a single document with each file labeled by name, plus a live token count so you know it fits before you paste it.

Try it on your own workbook

Drop a spreadsheet below, or a folder with a few of them mixed in with your other documents. It runs in your browser, there is no account, and nothing is uploaded to us.

What a workbook loses on the way out

Two things do not survive, and both are worth knowing before you trust an answer.

Formulas become their results. A cell holding =SUM(B2:B14) reaches the bundle as the number it computed, not as the formula. For almost every question this is what you want, since the model can read the figure and reason about it. If your question is about the calculation itself rather than its output, that is the one case where the text will not hold the answer.

Dates become numbers. Excel stores a date as a count of days and shows you a formatted date on top of it. Extraction reads what is stored, so a due-date column can arrive as 45000 rather than the date you see in the cell.

Charts, cell colours, and conditional formatting are not text either, so none of them come out. If the point of a sheet lives in its highlighting, say in the prompt what the highlighting meant.

Mind the token and cost budget

Spreadsheets are the format most likely to blow a context window without looking like it. A sheet that fits on one screen can carry ten thousand rows below the fold, and every one of them is tokens. FileConcat counts as you add and remove files and can show the cost for a given model, so you can cut to the sheets that matter before you paste rather than after the bill.