Context over vibes
Describe Excel version, whether you use Microsoft 365 dynamic arrays, table names, and sample headers. Vague prompts produce brittle formulas.
How to use language models as a spreadsheet co-pilot: faster drafts, stricter reviews, and habits that keep production workbooks safe.
AI is excellent at first drafts and explanations. You remain accountable for correctness, security, and fit for your data.
Describe Excel version, whether you use Microsoft 365 dynamic arrays, table names, and sample headers. Vague prompts produce brittle formulas.
Test on a copy. Check blanks, duplicates, and edge dates. Compare totals against a known-good baseline before overwriting the live file.
Share structure and fake rows, not payroll, health, or customer PII. Strip real identifiers before pasting sheets into any chat tool.
Ready-to-adapt prompts. Replace bracketed parts with your sheet reality.
“I have an Excel Table named Sales with columns Date, Region, Amount. Write a dynamic-array formula that returns Region totals for the current month, sorted high to low. Prefer FILTER / SUMIFs as appropriate. Explain edge cases for empty results.”
“Here are sample column names and two messy date formats in a CSV. Write M steps to promote headers, set types, parse dates with locale en-GB, and remove null Amount rows. Comment each step briefly.”
“Generate a VBA module with Option Explicit that exports the active sheet as CSV to a path in cell Config!B2. Include error handling, turn off screen updating, and restore settings in a Finally-style exit. No clipboard tricks.”
“Review this formula / M / VBA for performance and silent failures. List assumptions, suggest tests, and flag any use of volatile functions or full-column references.”
Explaining error messages, drafting LET blocks, converting nested logic into readable steps, sketching Power Query transforms, writing unit-style test cases for edge data, and producing handover notes for complex workbooks.
Anything that touches credentials, external APIs with secrets, bulk email from Excel, or “delete rows if…” logic without a backup. Prefer human sign-off before macros that mutate many sheets or files on disk.
One sentence outcome, inputs, outputs, and constraints (Excel 365 vs older, desktop only, no add-ins).
Ask for the simplest correct approach first. Request a second pass that uses Tables or Power Query if the first is range-heavy.
Run tests, lock assumptions on a Notes sheet, and keep a short changelog when the model helped design the solution.
AI accelerates learning; Power Query, structure, and clear VBA keep systems maintainable after the chat window closes.