Use Claude to Explain a Spreadsheet Before You Change It

A spreadsheet can look simple and still hide a formula copied one row too far, a blank that means “not applicable,” or a column whose label no longer matches its contents. Before asking AI to clean it up, ask for an explanation.

This documentation-based tutorial uses a fictional spreadsheet and an illustrative response. The spreadsheet was not uploaded to Claude in a recorded test here, so the example is not a reproduced demo.

What you need

• Claude with file upload available for your account.
• A safe copy of a spreadsheet or CSV. Do not start with a live finance, payroll, customer, or production file.
• A spreadsheet surface that lets you inspect the original values and formulas.

Claude’s help documentation lists CSV and XLSX among supported document types. It notes that XLSX uploads require code execution and file creation to be enabled on the account. Account, plan, and workspace settings can vary.

The fictional sheet

Create a small practice file called q3-renewals-sample.csv with this exact content:

account,region,renewal_month,contract_value,renewal_owner,status
Atlas Labs,West,October,12000,Mina,Review
Beacon Works,North,October,8500,Jon,Confirmed
Cedar Health,West,November,14500,,Review
Delta Retail,South,November,-3000,Arun,Review
Eon Systems,East,December,22000,Lea,Confirmed

The negative contract value is deliberate. The blank owner is deliberate. Neither one should be silently repaired.

1. Upload a safe copy

In Claude, use the attachment control to add q3-renewals-sample.csv. If you use an XLSX file instead, confirm that the required file-creation and code-execution settings are available before uploading it.

For a real workbook, remove unrelated sheets and sensitive columns first. Keep the original file unchanged so you can compare any proposed edit later.

2. Ask for an explanation, not a cleanup

Paste this prompt:

Review only the attached q3-renewals-sample.csv.

Do not edit the file, create a replacement, calculate a new business total,
or assume that a blank or negative value is an error. This is an explanation
pass only.

Return:
1. A plain-English description of each column and its apparent data type.
2. Every blank, negative, duplicate, or unusual value you can identify.
3. The exact row and column for each issue.
4. Any assumptions you would need before changing the file.
5. Three questions a human owner should answer.
6. A short list of safe next steps, with no edits performed.

For each observation, quote the original value. If the file does not provide
enough information, say "Not determined from this file."

The boundary matters. “Clean this spreadsheet” asks for an action. “Explain this spreadsheet” asks for a review that a person can check first.

3. Review the illustrative response

A suitable illustrative response might identify:

- contract_value is numeric in the sample, but the file does not define its currency.
- Cedar Health has a blank renewal_owner in row 4. The cause is not determined from this file.
- Delta Retail has contract_value = -3000 in row 5. The file does not say whether this is a credit, correction, or error.
- No exact duplicate rows were observed in the five-row sample.

Questions for the owner:
1. What currency does contract_value use?
2. Is a negative value allowed for credits or adjustments?
3. Who owns Cedar Health's November renewal?

This is an example of the requested format, not a recorded Claude response. Check every observation against the original file yourself.

4. Apply the acceptance check

The explanation passes only when:

1. Every cited value exists in the original file.
2. Row numbers are counted consistently, with the header treated as row 1 or explicitly excluded.
3. Blank, negative, and unusual values are reported separately.
4. The response does not turn an unknown business rule into a correction.
5. No output file was created or overwritten during the explanation pass.
6. A named human answers the open questions before any edit is approved.

If you later ask for a proposed correction, save it as a new file and request a change table containing original value, proposed value, reason, and confidence. Keep the original beside it.

Troubleshooting

• Claude misses a column: Use a plain CSV or a smaller workbook, then state the exact sheet and range to inspect.
• The spreadsheet contains charts or embedded images: Treat this as a data extraction problem. Export the relevant table as CSV and compare it with the visual source.
• An XLSX upload is unavailable: Use CSV for the explanation pass, or check whether your account allows the required code execution and file creation settings.
• The response sounds certain: Add “Separate observed values from assumptions” and require an exact quote for every issue.

Limits and privacy

File support and limits depend on the current Claude account and workspace. Use a redacted practice file when possible. Do not upload credentials, personal data, confidential customer records, or regulated information without your organization’s approval. An AI explanation is a review aid, not an accounting or compliance determination.

Verdict

TRY this before changing a messy spreadsheet with unclear business rules. SKIP it for a file you are not authorized to upload. USE the explanation only after checking its quoted values against the original.

See Claude’s file upload documentation for supported file types and Claude’s Google Workspace guide for the separate Sheets sidebar workflow.

For another practical AI-for-work lesson, join the Weekly Verdict.