Check Your Copilot Report Before Publishing the Spreadsheet

Excel Canvas is Microsoft’s announced Copilot-generated view of workbook visualizations, metrics and insights—not a substitute for checking the data or preparing a public release. Before sharing a report, verify its calculations against the source table, then publish a separate file containing only approved public data. As of October 3, 2026, its availability remains a rollout plan rather than a guarantee for every account.

Enter the source-table totals to check the completion rate; the defaults use the fictional example below.

The calculator checks arithmetic, not whether Copilot selected the correct rows. Enter totals you have independently checked in the source table, rather than copying both numbers from the generated report. Otherwise, a plausible percentage can conceal an incorrect denominator.

Excel Canvas Is an Announced Workbook Report View

Microsoft’s roadmap entry 569424 describes Excel Canvas as an interactive view of visualizations, metrics and insights generated from workbook data. It is designed to update automatically when that data changes, and users can modify it by chatting with Copilot.

Those capabilities describe a report connected to a workbook. They do not establish that exporting the workbook will export the Canvas presentation, or that publishing its data elsewhere will preserve the connection. Treat the generated report, the underlying table and the public dataset as separate objects that need separate checks.

The roadmap still lists September 2026 and “In development.” A public archive of Microsoft’s Message Center notice MC1485411 gives a later schedule: worldwide rollout beginning in mid-October, with completion expected by the end of October 2026. That notice names Excel for Windows and Excel Online, for users with access to Copilot in Excel.

The September date in XDA’s earlier report therefore is not confirmation that everyone already has the feature. Microsoft also states that roadmap dates are estimates and can change. The cited notice does not establish availability for other Excel platforms or users without access to Copilot in Excel.

You can prepare and verify the source table before Canvas appears in your account. What you cannot yet establish from an announcement alone is whether a particular generated report refreshes correctly in your workbook.

Define the Metric Before Asking Copilot to Report It

Suppose you are preparing a public report on service-request completion. This fictional sample has one row per district per month:

month	district	requests	completed
2026-01	North	120	100
2026-01	Central	80	60
2026-01	South	150	120

The independently checked results are 350 requests, 280 completed and an overall completion rate of 80%. Calculate the overall completion rate as total completed divided by total requests, multiplied by 100.

For this example, completed counts completed requests from the same group represented by requests. That definition makes the numerator a subset of the denominator. Without that relationship, the percentage could describe a different measure even when the arithmetic is correct.

For example, “requests received during the month” and “requests completed during the month” need not describe the same group. Some completions could belong to requests received earlier. If that is how your real table works, do not present the division as the share of that month’s requests completed. Define the measure to match the data, or prepare a table that tracks the intended group.

Use one header row with explicit field names, consistent month values and district labels, and numeric counts. Keep subtotal rows out of the observations used for calculation. A total row included alongside the district rows could cause the report to count the same requests twice.

Write down which requests qualify as completed and the cutoff date used to determine their status. The sample’s month label identifies the reporting period; it does not, by itself, explain when completion status was assessed. A public reader needs that distinction if requests can be completed after the reporting month ends.

Also distinguish a missing count from a confirmed zero. A blank completed cell does not establish that no requests were completed. Resolve its meaning before allowing it to affect the denominator or the report narrative.

Specify the Calculation, Not Just the Chart

When Canvas becomes available in your account, an illustrative request to Copilot could be:

Create a report comparing requests and completed requests by district. Show total requests, total completed, and overall completion rate calculated as total completed divided by total requests. Label the reporting month and explain the rate definition.

This is a suggested analytical specification, not a required command or a promise about the resulting layout. It identifies the comparison, the headline metrics and the calculation you intend to verify.

The phrase “overall completion rate” needs the explicit formula. An unweighted average of district percentages gives every district equal influence, regardless of its request count. The combined completion rate instead gives each request equal influence. These are different calculations, even if both could be described casually as an average.

For this sample, use the combined totals of 280 and 350 to check the reported 80%. Do not accept a district-rate average merely because it appears under an overall-rate heading. If you actually want an equal-weight district comparison, label it separately rather than substituting it for the share of all requests completed.

Ask for the reporting month and rate definition because those labels constrain the interpretation. A correctly calculated percentage with an ambiguous label can still mislead a reader about what was completed and when.

Verify Totals, Rows and Claims Separately

Compare the generated report with the source table before sharing it. Start with the totals, then inspect the chart and narrative. A correct headline number does not establish that every district is represented correctly.

Report Element	Expected Result	Check Against
Total requests	350	All three request counts
Total completed	280	All three completed counts
Overall rate	80%	280 divided by 350
District coverage	North, Central, South	All three source rows

For each plotted district, compare both series with its row. North should use 120 requests and 100 completed, Central 80 and 60, and South 150 and 120. Check the series labels as well as the values: swapping requests and completions can leave a chart looking orderly while reversing its meaning.

Check whether the report applies filters or silently excludes blanks. A filtered result can be internally correct but wrong for a report intended to cover the whole table. The report’s scope must match the rows you used for the independent calculation.

Inspect any written insight for claims the source supports. This sample supports a comparison of recorded requests and completions for the listed districts and month. It does not establish why districts differ, how long individual requests took, or whether performance improved over time. A convincing sentence is not a substitute for a calculation or evidence of a cause.

Keep a short verification record tied to the version you intend to publish: the approved totals, the rate definition and any exclusions. Its purpose is to make discrepancies traceable. If a later report shows a different percentage, you should be able to identify whether the source changed, the selected rows changed, or the calculation changed.

If Copilot also edits the underlying workbook, review its change history before publishing, then validate the final table independently. Checking a report against an earlier table is not enough when the data itself has been modified.

Test Automatic Updates in a Disposable Copy

To evaluate the announced automatic updating, use a disposable workbook copy. Change North’s completed count from 100 to 110. The new expected results are 290 completed and approximately 82.86% overall completion; total requests should remain 350.

This controlled change separates several checks. The North chart value should reflect the new count. The completed total should reflect the increase. The overall percentage should be recalculated using the unchanged denominator. Any written insight that depends on completion counts should remain consistent with the revised data.

Do not count the test as successful just because one visible number changes. A refreshed headline metric beside a stale chart or stale narrative is still an inconsistent report. Conversely, do not infer a product-wide limitation from one workbook without first establishing what changed and what the report includes.

Restore the value or discard the test copy afterward. Keep the test out of the release file: the change exists to exercise the report, not to revise the fictional source’s approved result.

Repeat the relevant checks if you change the table’s structure or the report’s scope. A successful value-update test does not establish that adding a district, changing a field name or filtering the report will produce the intended result. Verify the actual version you plan to share, rather than relying on an earlier successful test.

Publish a Release File, Not the Working Workbook

Create a release file containing only the intended public fields and rows. Do not upload sensitive information or rely on hidden worksheets and columns as a privacy boundary. The file being uploaded should itself be suitable for public release, without depending on which parts a viewer happens to display.

For the sample, the intended public table contains month, district, requests and completed. In a real workbook, inspect whether other sheets, columns or notes contain information outside that release scope. Remove unapproved content from the release copy rather than merely hiding it in the working workbook.

TablePage accepts CSV, TSV, XLSX and XLS uploads and generates a public dataset page with a shareable link, charts and a filterable table that readers can explore without signing up. This is a separate publishing workflow. Do not assume an uploaded workbook will reproduce the Excel Canvas layout or inherit its live-update behavior.

For a single flat table, consider CSV; for format trade-offs, see CSV vs XLSX. Choose the release format for the approved data, not as a way to preserve a report layout that the publishing workflow has not promised to support.

After export, compare row counts, totals, labels and dates with the approved source. For this sample, the release should retain the three district rows and the verified totals of 350 requests and 280 completed. Inspect the exported file itself, not only the workbook from which it was created.

After upload, check the public page against that same release file. Confirm that the displayed fields and district labels match, and that the table and charts represent the intended values. If readers can filter the page, make clear whether the stated 80% describes the full released dataset or a particular selection.

Give Readers Enough Context to Reproduce the Rate

Alongside the shared link, provide the source, reporting period, field definitions, exclusions and release date. These details explain the numbers without requiring access to your working workbook or Copilot conversation.

For this fictional example, identify the data as fictional rather than implying that it is an operational service report. State that the completion rate uses completed requests from the same group as the request count, and that the overall rate is calculated from combined totals rather than an average of district percentages.

For a real release, supply the actual completion cutoff and any excluded rows. The draft provides no real service-data source or cutoff date; those details must come from the dataset owner, not from Copilot-generated prose.

If you later revise the approved data, validate the replacement release and its public presentation again. Automatic updating inside Excel Canvas does not establish that a separately uploaded dataset has changed. The public link should point readers to data whose version and definitions you can identify, with headline metrics they can reproduce from the released table.