Asking questions of a spreadsheet instead of building formulas

What it is good at

• Handles pivot-table-style questions without you needing to know pivot tables.
• Can spot outliers and data quality issues you'd likely miss scanning manually.
• Explains its calculation method, so you can sanity-check the logic, not just trust the number.
• Produces quick charts from the same data without switching tools.

Where it falls short

• Large or highly complex spreadsheets can exceed what it can process reliably in one go.
• It can misread ambiguous column headers, check its interpretation before trusting the output.
• Sensitive or client data needs the same handling care as any other tool you upload it to.

How to use it for creative work

1. 01

   Upload and ask it to describe the data first

   Before asking any real question, ask it to describe what each column contains and flag anything odd, blank cells, mixed formats, duplicate rows. This catches misread headers early.

2. 02

   Ask business questions, not formula questions

   Phrase requests the way you'd ask a colleague: 'which region underperformed this quarter and by how much', not 'write me a SUMIFS formula'. Let the model choose the method.

3. 03

   Ask it to show its working

   Request the calculation steps or the formula it used alongside the answer, so you can verify the logic rather than take the number on faith.

4. 04

   Cross-check one number by hand

   Pick one output and verify it manually or in the spreadsheet directly. This single check catches most misinterpretations before they reach a report.

5. 05

   Turn the finding into a chart and a sentence

   Ask for a simple chart of the key finding plus a one-sentence summary you could put in front of a client or manager without further editing.

Ready-to-use prompts

I've uploaded a spreadsheet. Before answering anything, describe each column, flag any blank cells, inconsistent formats or likely duplicate rows, and tell me what you think this dataset is tracking.

Using this data, which three [products/campaigns/clients] contributed most to the change in [metric] between [period A] and [period B]? Show the numbers, not just the ranking.

For the answer above, show me the exact calculation or formula-equivalent logic you used, step by step, so I can verify it against the raw data.

Scan this dataset for any row or value that looks like an outlier or a likely data entry error. List each one with the row reference and why it looks unusual.

Create a simple bar chart showing [metric] by [category] from this data, and write a one-sentence summary of the key takeaway suitable for a client update.

Questions people ask

Is this safe for client financial data?

Check your data handling policy and the tool's enterprise settings first; for sensitive client data, use a business-tier account with data retention controls turned off, not a free consumer account.

How big a spreadsheet can it handle?

It varies by tool and file size, but tens of thousands of rows is usually fine; very large multi-sheet workbooks are better split before uploading.

Does Gemini or ChatGPT do this better?

Both handle it well; Gemini has an edge if your data already lives in Google Sheets, since it can work inside the sheet directly.