The COPILOT function was one of the clearest signals that Excel’s AI direction was moving deeper into the grid. Instead of staying only in a side panel, AI could sit inside worksheet logic itself and produce values directly in cells — until Microsoft retired the function on 14 September 2026 without ever taking it out of preview.
That is powerful, but it also changes the review burden. A side-panel answer is easy to question. A cell result can disappear into a model very quickly if you do not build the right checking habits.
Note: As of 25 September 2026, the COPILOT worksheet function is no longer available in Excel. It only ever shipped as a preview to the Frontier and Microsoft 365 Insider programmes, and since 14 September 2026 any cell that still calls it returns #NAME? when it recalculates; results that were already calculated stay in the workbook as cached values. Microsoft’s replacement is the Copilot pane in Excel, where you describe the same summarising, classifying, or web lookup task in your own words, so the review habits below now apply to whatever you bring back from there.
Quick answer
The COPILOT function was useful for bounded AI assistance inside worksheet logic, especially when you wanted a result in a cell rather than a side conversation. It was never a substitute for deterministic formulas in any workflow that must be fully auditable or reproducible, and since its retirement the same test applies to AI output you bring in from the Copilot pane.
- You need AI assistance directly in worksheet output.
- The task is advisory, classificatory, or interpretive rather than purely deterministic.
- You can review and document where the AI-derived values are being used.
Where it can genuinely help
The strongest use cases are the ones where traditional formulas are possible but awkward, or where the task involves interpretation: classifying free-text comments, drafting a short summary, or converting an informal request into a structured answer.
Why review matters more than syntax
If an AI-produced value feeds a metric, a filter, or a business rule, someone needs to know where that value came from and how trustworthy it is.
Label the column clearly, sample the output, and use a manual review pass before the results affect dashboards or decisions.
- Label AI-derived columns clearly.
- Review a sample before filling the function down a whole table.
- Keep the prompt objective and specific enough to reduce drift.
The real risks
The bigger risk is false confidence. A neat result in a cell feels more official than a paragraph in chat, so teams can accidentally treat it like a normal formula even when it behaves differently.
Worked example: classifying support comments
A support team has 1,200 comments from a customer survey. They want a quick first-pass label for each row: billing, delivery, product quality, or account issue.
The COPILOT function could create that first label column; with the function retired, the team now asks the Copilot pane to do it, but should still sample the outputs and avoid treating the result as final without review.
Common mistakes
- Using AI-derived cell values in a model without labelling them clearly.
- Replacing deterministic logic with the COPILOT function simply because it feels faster.
- Ignoring availability and rollout caveats in team documentation.
When to use something else
If you need a formula written for you, single-cell formula generation is the better topic. If you need a broader comparison of AI tools for spreadsheet work, go to the 2026 comparison guide.
Frequently asked questions
What is the COPILOT function for?
It was for bounded AI assistance inside worksheet logic when you wanted a result in a cell rather than a side chat. Microsoft retired it on 14 September 2026, so do those tasks in the Copilot pane in Excel now; neither is a replacement for deterministic formulas in anything that must be fully auditable or reproducible.
Where does it genuinely help?
Tasks that are awkward for formulas or need interpretation: classifying free-text comments, drafting a short summary, or turning an informal request into a structured answer.
Why does review matter more than syntax?
If an AI-produced value feeds a metric, a filter, or a business rule, someone must know where it came from and how trustworthy it is. The syntax is the easy part.
What is the real risk?
False confidence. A neat value in a cell feels more official than a chat paragraph, so teams treat it like a normal formula even though it can behave non-deterministically.
Is the COPILOT function deterministic?
Do not assume so. AI outputs can vary and depend on context, unlike SUM or XLOOKUP, so avoid it in cells that must recompute identically every time for audit or reconciliation.
When should I avoid it?
In auditable or reproducible workflows such as financial reconciliation or compliance figures. Use deterministic formulas there, and reserve COPILOT for interpretive, lower-stakes helper cells.
Related guides on this site
If you want to keep going without opening dead ends, these are the most useful next reads from this site.
- Generate Single-Cell Formulas With Copilot in Excel: Fast Wins and Failure Modes
- How to Review AI-Generated Excel Formulas Before You Trust Them
- ChatGPT vs Claude vs Copilot vs Gemini for Excel in 2026
- How to Use Microsoft Copilot for Data Analysis in Excel
Official references
These official references are useful if you need the product or framework documentation alongside this guide.