Microsoft’s answer to the retirement of =COPILOT() is the Copilot side pane. The official line is that it “provides many of the same AI-powered capabilities, including summarizing text, classifying data, generating content, and retrieving information from the web.”
That is true, and it is also incomplete in a way that matters. This post is a direct comparison of what each one actually does, so you can work out which of your workflows survive the transition and which need a different answer entirely.
The One Difference That Explains Everything Else
=COPILOT() lived in cells. The side pane lives in a conversation.
Every other difference follows from that. A formula in a cell can be dragged down two hundred rows, producing two hundred independent results that sort, filter, and feed other formulas. A conversation produces an answer you read, and then copy somewhere if you want to keep it.
For some tasks that distinction is irrelevant. For others it is the entire job.
Copilot Side Pane vs =COPILOT(): Side by Side
| Capability | =COPILOT() function | Copilot side pane |
|---|---|---|
| Populates cells row by row | Yes — drag down like any formula | No — returns a conversational answer |
| Output is filterable and sortable | Yes, once in cells | Only after manual copying |
| Sees the whole workbook for context | No — only what you pass it | Yes |
| Summarises a dataset | Poorly — one cell at a time | Yes, this is its strength |
| Answers follow-up questions | No | Yes |
| Rate limited | Roughly 100 calculations per 10 minutes | Standard Copilot limits, far less restrictive per task |
| Works in Confidential labelled workbooks | No | Depends on tenant policy |
| Deterministic output | No — same input could vary on recalculation | No |
| Availability | Retired September 14, 2026 | Generally available |
Workflows That Transfer Cleanly
These were arguably always better suited to the side pane, and moving them is an improvement rather than a compromise.
Summarising a dataset
Asking “what are the main themes in these three hundred reviews” was awkward with the function — you would summarise each row individually and then have to summarise the summaries. The side pane reads the range and answers directly, and you can ask follow-ups.
Exploratory analysis
“Which product has the most complaints” or “what changed between Q1 and Q2” are questions, not columns. The side pane’s access to the whole workbook makes it materially better at these than the function ever was.
Drafting a single piece of text
Writing one customer response or one summary paragraph does not need a formula. The side pane handles it and lets you iterate on the wording conversationally.
Workflows That Do Not Transfer
This is where being honest matters more than being reassuring.
Per-row classification at scale
Tagging two hundred support tickets with a priority. Categorising eight hundred expense lines. Assigning a sentiment to every review in a column. The function did this in one drag. The side pane will discuss your data intelligently and will not write those two hundred values into cells.
You can ask it to analyse the column and it will give you a useful answer about the column. What you cannot get is a filterable value sitting next to each row.
Normalisation of messy values
Standardising vendor names where “MICROSOFT CORP”, “microsoft”, and “Microsoft Corporation Inc” all need to become “Microsoft”. This was the function’s strongest use case precisely because a lookup table cannot anticipate variations it has never seen. It is the loss that will be felt most by anyone doing data cleanup.
Extraction into structured columns
Pulling a dollar amount, a date, or an email address out of a free-text field and into its own column, across a whole sheet. Again: the side pane can tell you about the field. It cannot build the column.
What to Use Instead for the Ones That Do Not Transfer
Ranked by how well they actually match the pattern you are replacing:
- A third-party AI add-in that provides worksheet functions. The closest structural match — you get a formula that lives in a cell and drags down. The cost is a new vendor in your data path, which for anything touching client or financial data means a real review rather than a quick install.
- Office Scripts calling an external API. Full control, writes results back into cells, and schedulable through Power Automate. Requires someone comfortable in TypeScript, which for most small firms is the blocker.
- Power Query. Not an AI tool, and that is often the point. A surprising share of what people used
=COPILOT()for was actually rules-based work that a deterministic transformation handles better and more reliably. - Accept the manual step. For a monthly task on fifty rows, the side pane plus copy-paste may genuinely be cheaper than building anything. Not every workflow deserves automation.
An Honest Word About Determinism
One thing worth keeping in perspective as you rebuild: neither tool is deterministic, and neither should be trusted with anything that must be exactly right.
The function could return different output for identical input on recalculation. The side pane has the same property. Whatever you replace this with, if it involves a language model, the same rule applies — use it for a first pass on tasks where you can sample the output and catch errors, and never for arithmetic or for anything that goes into a financial record without review.
That was true of =COPILOT() when it existed, and we said so in our original review. It remains true of whatever comes next.
If You Are Still on the Function
You have until September 14. Convert anything you need to keep into static values first — our step-by-step conversion guide covers the process — and read the retirement guide for what Microsoft has and has not confirmed.