Click the + on the left of the ChatGPT message box, choose Add photos & files, pick your Excel or CSV file, and before you ask anything else, ask ChatGPT to describe the columns, count the rows and list any duplicates or blank cells. Fix what it finds, then ask your real question, as you do in the free lesson Analyze a Spreadsheet.
That middle step is the one most people skip. ChatGPT will happily add up a column that holds the same order twice. Asking it to check the file first takes one message and saves you from a confident wrong number.
What are the steps?
- Export the data. An .xlsx or .csv file both work. If the sheet has several tabs, say which tab you mean. Delete columns the question does not need, especially names, emails and account numbers.
- Attach it. Click +, then Add photos & files, pick the file and click Open.
- Ask for a check, not insights. For example: "Tell me what each column means, the number of rows, the date range, and any duplicates or blank cells. No analysis yet."
- Decide how to clean it. You choose: remove duplicates, label blanks as Unknown, or move doubtful rows to a "needs review" group. Tell ChatGPT, and ask it to use the cleaned data from now on and give you the cleaned file.
- Ask your question as a table. "Total revenue by month and which category grew most, as tables" is easier to check than a paragraph.
- Ask for a chart if it helps. ChatGPT can draw one and you can download it.
- Finish with takeaways and one thing to double-check. Ask for three takeaways, each with its number.
Before you upload company data at all, read is it safe to upload a company spreadsheet to ChatGPT.
What does this look like in a real lesson?
In Analyze a Spreadsheet, a made-up garden center has six months of sales in a CSV, about 1,200 orders. You attach it and paste a request that asks only for the columns, the row count, the date range and problems. ChatGPT finds that three orders appear twice and 12 rows have no sales channel.
You tell it to remove the duplicates and label the blank channels "Unknown", and it makes a new cleaned file you can open in a panel like a spreadsheet. Then you ask for revenue by month and the fastest-growing category as tables: May was the peak and Plants grew the most. A line chart shows the same thing, you download it, and you end with three takeaways plus one thing to double-check.
The same pattern works for other files. In Analyze a Client's P&L Export the check finds four duplicate bank-feed entries and nine rows with no account, and you park those nine in "Needs review" instead of letting them into the cost lines. In the donor gift lesson, the check flags gifts that synced twice and pledges that were never received, and you keep the pledges out of every total before the numbers go to the board.
What kinds of questions work best?
- Totals and groupings: sales by month, cost by category, gifts by source.
- Changes over time: which line grew, which shrank, and by how much.
- Shares: each cost as a share of sales in two months you name.
- Problems: duplicates, blanks, outliers, rows that do not fit.
- Plain-English summaries: three points for a meeting, each with its number.
Questions that need knowledge outside the file, such as "why did sales drop?", get guesses. Ask what the data shows, then add the reasons yourself. For more on what AI can and cannot answer from your data, see can AI answer questions about your data.
What mistakes should I avoid?
- Asking for insights first. Check the file, then ask. Duplicates and blanks change totals.
- Uploading more than the question needs. Remove personal details and columns you do not use.
- Taking a total on trust. Pick one number and check it in Excel with a filter or a SUM.
- Letting ChatGPT decide what to do with bad rows. You know whether a blank means zero, unknown or a mistake. Tell it.
- Forgetting which tab it read. On multi-tab workbooks, name the tab in your question.
Which other tools can do this?
- Claude also takes spreadsheet uploads and can analyze them in a chat.
- Microsoft Excel with Copilot works inside Excel itself, so you can ask for formulas, charts and PivotTables without leaving the file.
The guide to the best AI tools for spreadsheets compares these and add-ons, and what AI tool can analyze my Excel spreadsheet helps you pick one.
Key takeaways
- Attach the file with + and Add photos & files.
- Ask ChatGPT to check columns, rows, dates, duplicates and blanks before anything else.
- You decide how bad rows are handled, then ask for the cleaned file.
- Ask for tables and charts, and check one number yourself.
- Remove personal details the question does not need.
Frequently asked questions
- Can ChatGPT read Excel files with several tabs?
Yes, but say which tab to use, or ask it to list the tabs first. Otherwise it may answer from the wrong one.
- Can ChatGPT edit my spreadsheet?
It can make a new, changed file for you to download, such as a cleaned copy. It does not change the original file on your computer.
- Does file upload work on the free plan?
Yes, with limits on how many files and how much analysis you can do. Paid plans raise those limits.