Build a refreshable Power Query with Copilot Get Data, then prove next week's file still works
In Excel for the web, invoke Copilot's Get Data skill with explicit business rules, confirm it created a real Power Query (not pasted values), check rows and sums against the source, inspect the Advanced Editor for hardcoded columns, then refresh with next week's file before you trust it.

Tools used
- Microsoft Copilot in ExcelHosts the built-in Get Data skill (@get-data) that drafts the Power Query
- Power QueryStores the reviewed M steps you refresh next week
- Excel for the webWhere Microsoft listed the mid-October 2026 Get Data rollout
Step navigator
A reviewed Power Query in Queries & Connections that matches source row counts and totals, survives a refresh on next week's file, and has no silent hardcoded column list you did not accept.
Excel users on Excel for the web with Copilot who repeat the same import/cleanup every week or month and want a query they can refresh instead of a fresh chat.
Before you start
Microsoft rolled Copilot's built-in Get Data skill into Excel for the web in mid-October 2026: you describe the data and transformations in natural language, and Copilot builds a Power Query you can inspect and refresh. George Mount (Stringfest Analytics, 22 Oct 2026) argues the point of Get Data is not another chat you re-run every month — it is a reviewed query you keep. InflueXcel put @get-data through three real traps (US dates, folder+subfolder, blocky goals) and a "next week" schema change. This guide follows those two human write-ups plus Microsoft's announcement wording. Exact AppSource/license gates and every supported source stay gaps.
Step-by-step instructions
Confirm Get Data is available, then ask for a query — not a one-off paste
Microsoft's October 2026 Excel notes and Neowin's write-up of the AI-at-Work roadmap: Copilot in Excel for the web gets a built-in Get Data skill. You describe the data you need and how to shape it; Copilot connects, transforms, loads into the sheet, and leaves a Power Query script you can open in the Advanced Editor. Mount: treat the skill as the playbook and your prompt as this month's assignment — the durable artifact is the saved query, not the chat. Availability may still vary by tenant; if you do not see Get Data / @get-data, stop here (gap).
Import this month's sales export. Keep account codes as text, standardize the dates, retain all transactions, join the account reference table, and flag unmatched codes. Build a Power Query query I can review and refresh.

Why this works
Without a saved query you are back to hoping next month's chat remembers the rules.
Grant access, load only after Copilot asks, then prove it created a real query
InflueXcel's first live test (web orders CSV with US dates like 1/13/2027): invoke with @get-data, ask for a table on a new sheet with correct types. Copilot asked permission to read the file and asked again before loading; it dropped a nameless index column on its own and used en-US locale so 1/13/2027 became 13/01/2027. Their first of five checks: confirm Queries & Connections shows a query — not values pasted into cells. They verified 612 rows and sum 2.163.477,50 against the file.
Why this works
A pretty table of values has no refresh button next Monday.
Spell joins and exceptions the way Mount does for append / unpivot / merge
Mount's three refresh-worthy patterns when you brief Get Data: append recurring exports (combined rows should equal the sum of sources before intentional filters); unpivot wide category/month columns into fields you can Pivot; merge reference tables with an explicit left outer join when every transaction must survive, and flag unmatched codes instead of letting an inner join drop sales. He wants those decisions saved in the query, not re-decided by the model next month.
Why this works
Vague \"clean this\" leaves Copilot free to change which invoices survive between runs.
Open Advanced Editor and hunt for hardcoded columns, paths, and skip-N rows
InflueXcel's folder test (one CSV per month plus a trap antiguos subfolder): Copilot filtered the subfolder path and converted text amounts — row/sum matched — but some steps fixed the column list. Their goals/blocks case also over-used fixed values even when header detection by content was correct. Check four: look in Advanced Editor for hardcoded paths, years, column name lists, and "skip N rows". Those are the bugs that only appear when next week's file changes.
Why this works
Hardcoded column lists pass today and break the Monday refresh.
Refresh on next week's file; if it fails, expect a new query — delete the old one yourself
InflueXcel's decisive test: next week's orders CSV dropped coupon and added gift_wrap. Refresh All failed with The column 'coupon' of the table wasn't found because the type-change step had handwritten the column list. Asking Copilot to fix it: it diagnosed the step correctly but could not edit the existing query — it created a new corrected one and left the broken query in the workbook for the human to remove. Their fifth check is mandatory: does refresh survive next week's schema?
Why this works
Passing today's file is not the same as surviving next week's export.
End of guide · Reviewed Nov 1, 2026
Pro tips2
- Spell the business rules in the prompt: keep codes as text, which join, what to flag — Mount's example does this on purpose.
- Before you trust it, run InflueXcel's five checks: real query vs values, rows+sum vs source, types, hardcoded bits in Advanced Editor, refresh on next week's data.
Common mistakes2
- Re-asking Copilot every month instead of refreshing the reviewed query — Mount compares that to putting RANDBETWEEN in charge of which invoices survive.
- Assuming Copilot can edit the broken query in place; InflueXcel found it creates a new corrected query and leaves the failing one for you to remove.







