Use Synthetic Examples Before Giving an AI Tool a Real Spreadsheet
When an assistant needs to understand a spreadsheet problem, it often needs the structure rather than the real people or transactions. A synthetic sample lets you explain that structure using invented records. Treat it as a working example, not as a guarantee that every privacy concern has been removed.
Identify what the task actually needs
If you need a formula for counting incomplete tasks, the useful fields may be Task ID, Status and Due date. Names, contact details, free-text notes and financial amounts may be irrelevant. Describe the intended calculation before choosing any rows to share.
Create a new sample with fictional identifiers such as TASK-001 and TASK-002. Include the same column names and data types as the original. Replacing only a person's name while retaining distinctive notes or exact dates can leave unnecessary identifying detail in the example.
Reproduce the difficult cases
Include a blank status, an unexpected status, a missing date and a duplicated identifier if those occur in your real workflow. Write the expected result next to each case. For example, decide whether a blank status means incomplete or unknown; the assistant should not invent that business rule.
Ask for an explanation and a formula against the sample. Check that any referenced ranges correspond to your actual spreadsheet application and layout before using them on a copy of the real file.
Transfer cautiously
Keep the original workbook unchanged while testing. Compare known examples manually and inspect error cells. A formula that works on five invented rows may fail when dates are stored as text or when the real sheet contains merged cells.
Share the smallest useful sample, inspect it before submission, and follow the rules that apply to your actual data. If the task requires protected information, a synthetic demonstration may be the appropriate stopping point.
A worked example you can check before uploading
The following records are invented for this article. They are not disguised customer records. The question is deliberately narrow: how many rows have Status exactly equal to Open? Due dates are present to reproduce the column structure, but are not used in this count.
| Task ID | Status | Due date |
|---|---|---|
| TASK-001 | Open | 2026-10-20 |
| TASK-002 | Done | 2026-10-21 |
| TASK-003 | (blank cell) | (blank cell) |
| TASK-001 | Open | 2026-10-22 |
Leave the marked blank cells empty when recreating this table. Before asking an assistant, write down the expected result: two Open rows, one Done row and one blank Status. There are four rows but only three distinct task identifiers. Counting rows and counting unique tasks are different requirements.
A prompt with explicit acceptance checks
These four records are entirely fictional. Status is in B2:B5. Suggest a Google Sheets formula that counts rows whose Status equals Open. Explain the result for this sample. Do not infer a status for blank cells, deduplicate task identifiers, or use Due date. If the requirement is ambiguous, ask before changing it.
A response passes the sample check only if it returns 2 for the requested count and explains that both TASK-001 rows count. If it returns 1 by deduplicating identifiers, it has changed the task. If it treats the blank Status as Open, it has invented a rule. Test another sample with no Open rows before using the answer in your real workbook.
Keep the transfer boundary visible
Copy the approved calculation into the original workbook yourself, then test it locally against cases you understand. The assistant does not need the original file merely to demonstrate this counting rule. A synthetic example reduces unnecessary sharing; it does not grant permission to upload confidential data or prove that a third-party service meets your organisation's requirements.
Comments