Verified against Claude for Excel · 2026-07-27
Get a spreadsheet analyzed with formulas explained, not just numbers
A spreadsheet-analysis prompt for Claude for Excel / Sheets that asks for findings, the exact formulas behind them, and flagged data-quality problems, so you get an auditable analysis instead of a black-box number.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Analyze the spreadsheet data below (or the sheet you have access to) to answer the question, and show your work in a form I can audit in the spreadsheet itself, not just a spoken summary. QUESTION Which three sales regions had the highest quarter-over-quarter growth in Q2, and by how much? DATA DESCRIPTION Sheet "Sales_2026" has columns: Date, Region, Rep, Amount, Product. Roughly 4,800 rows covering Jan-Jun 2026. WHAT TO PRODUCE 1. The direct answer to the question, as a number or short statement, stated first. 2. The exact formula or calculation method used to get there — written as an actual formula (e.g. a SUMIFS or a pivot logic description), not just "I calculated the average." 3. Any data-quality issue you noticed while working — blank cells treated as zero, duplicate rows, inconsistent date formats, mismatched units — that could change the answer if handled differently. Flag it even if you handled it a reasonable way; I need to know it was there. 4. One sentence on how sensitive the answer is to the data-quality issues in part 3, if any — would fixing them plausibly change the conclusion, or not? CONSTRAINTS - If a column or field the question depends on is ambiguous (e.g. two columns could both be "revenue"), stop and ask which one rather than guessing. - State any filtering you applied (date range, category, excluded rows) explicitly — do not silently narrow the dataset without saying so.
Customize the highlighted detailsoptional — the prompt above already works
Why this works
Claude for Excel and Claude for Sheets work directly against live cell data, which means an analysis has an actual formula it could show, not just a narrated conclusion — asking explicitly for the formula or calculation method used, rather than accepting a prose summary of the result, converts the output from a claim you have to trust into a method you can re-run or audit yourself in the sheet. Requiring data-quality issues to be flagged even when handled reasonably targets the single biggest source of silently wrong spreadsheet analysis: a blank cell treated as zero versus excluded entirely produces a different average, and a model that picks one convention without saying so hands you a number with an invisible assumption baked in. The instruction to stop and ask rather than guess when a column is ambiguous (two plausible 'revenue' columns, for instance) matters specifically in real business spreadsheets, which routinely carry legacy or duplicate-looking columns from past reporting changes — a model that picks one and answers confidently gives you no signal that a choice was even made. Requiring filters to be stated explicitly closes the same gap at a different point: a growth calculation that quietly excludes a partial month or a returns-adjustment row will look completely normal in the output while resting on a scope decision you never approved.
Verified against
Claude for Excel Beta (Sonnet 4.6) · 2026-07-27
Changelog
- 2026-07-27 — Initial publish, verified against Claude for Excel (beta) on Sonnet 4.6.
Building this for real?
This is a free starting point. If you'd rather have what Scult builds built and running for your business, that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
