Prefer a purpose-built workbook
Start by defining what the AI needs to calculate, classify, or explain. Then create a new workbook with only the required columns, rows, and values. This reduces the number of formulas, hidden objects, historical notes, and internal references that must be inspected.
Copy values rather than formulas when the formulas themselves reveal internal file paths, server names, workbook links, pricing logic, or protected business rules—and when the AI task does not require those formulas. Keep the source workbook unchanged.
Give the new file and worksheets neutral names. Filenames and tab names can disclose customers, projects, incidents, or reporting periods even when the visible cells are transformed.
Transform identifiers as a set
Replace direct identifiers such as names, email addresses, employee IDs, account numbers, IP addresses, and customer domains. If relationships matter, map each original value to one consistent placeholder.
Then review combinations of role, office, date, amount, product, and unusual events. Generalise or bucket fields that are more precise than the analysis requires. For example, use month instead of day, region instead of branch, or a revenue band instead of an exact figure.
Recalculate uniqueness after the transformation. A row that is the only example in a group may remain identifiable even when obvious identifiers are gone.
Inspect more than visible cells
| Workbook area | What may remain | Review action |
|---|---|---|
| Hidden rows, columns, sheets | Raw data, lookup tables, excluded records | Unhide, inspect, and remove only after checking dependencies |
| Comments and notes | Names, explanations, review conversations | Delete unnecessary content and re-inspect |
| Names and links | Named ranges, external workbook paths, data connections | Inspect Name Manager, links, queries, and connections |
| Objects and caches | Charts, embedded files, PivotTable source data, images | Confirm sources and remove unsupported extras |
| Document properties | Author, company, last saved by, custom properties | Use Document Inspector and manual property review |
| Code | Macros, credentials, internal endpoints | Remove unless explicitly required and approved |
Microsoft states that Document Inspector can find several categories of hidden information, but it also documents limits. Some content cannot be removed automatically, and objects obscured by other objects or certain hidden locations may escape detection. Treat the tool as one inspection layer, not a guarantee.
Check formulas, values, and exported formats
Deleting a hidden sheet or external link can change calculations. Compare key totals and sample records against the source after each transformation. Open the clean copy on its own so that missing dependencies become visible.
CSV can reduce workbook complexity, but it is not automatically anonymous. The visible values, column names, free text, and filename can still expose sensitive information. CSV also discards formulas, formatting, and multiple sheets, which may or may not be acceptable for the task.
If the AI needs only a small table, a new CSV containing reviewed values can be a good minimisation step. If workbook structure matters, keep XLSX and inspect its additional structures.
Final Excel preflight
- Open the prepared file without the source workbook available.
- Search all worksheets for original identifiers and sensitive terms.
- Unhide every row, column, and sheet for the review.
- Check comments, notes, names, links, queries, connections, objects, and macros.
- Run Document Inspector and read each result instead of selecting Remove All blindly.
- Recheck totals, date ranges, formulas, and row counts needed by the task.
- Confirm the destination and have a person review the exact upload copy.
Official sources
This guide uses primary sources available on August 19, 2026. Product policies and software features can change, so confirm current terms before handling sensitive material.
- Remove hidden data and personal information with Document Inspector — Microsoft Support
- Privacy supplement for Microsoft Excel — Microsoft Support
- Data minimisation — UK Information Commissioner's Office
- NISTIR 8053: De-Identification of Personal Information — NIST