Microsoft shipped a June 2026 update for Excel that transforms Copilot from a generic assistant into a personalized, rule-bound analyst that works across more file formats than ever. The release also introduces PivotTable diagnostics and tighter admin controls for AI workflows in commercial tenants.

For the first time, Excel’s Copilot can remember your preferences across sessions—everything from how you format PivotTables to the chart types you favor—and apply them automatically when you ask the natural-language pane to build a report. This personalization engine is unique to Excel Copilot among the Office apps, signaling Microsoft’s intent to make the spreadsheet the flagship for its adaptive AI tools.

Workbook-level rules, the second pillar of the update, allow users to save a set of instructions that Copilot must obey inside a given file. Think of it as a governance layer that catches errors before they happen: a rule can mandate that all currency formulas round to two decimal places, that every chart on a sheet uses the company color palette, or that any VLOOKUP cross-references only certified named ranges. IT admins can push these rules via group policy, enforcing consistency across a fleet of analysts without locking down individual cell edits.

Copilot Personalization: AI That Adapts to You

Until now, asking Copilot to “analyze sales by region” produced a one-size-fits-all response. With the June update, the assistant builds a user profile that tracks repeated actions: the slicers you drag onto a PivotTable first, the conditional formatting styles you apply, even the specific number formatting you use for financial models. After a few sessions, Copilot begins to generate outputs that match your historical work, dramatically reducing the need to tweak its suggestions.

Microsoft has embedded the profile at the account level, so the preferences roam with your Microsoft 365 identity across devices. In a briefing for enterprise customers, product managers demonstrated how the system learned a user’s preference for stacked bar charts over clustered bars after just three consecutive requests. By the fifth query, every chart Copilot produced was stacked, with the legend placed at the bottom—without the user ever explicitly setting a default.

Privacy-conscious users can review and delete their Copilot profile data through a new “AI preferences” pane in the Excel Options dialog. The data is stored in the same encrypted vault that holds Office theme and ribbon customizations; it never leaves the Microsoft 365 tenant boundary for commercial customers. Microsoft says the profiling does not analyze the contents of cells—only the patterns of tool usage and formatting commands—so sensitive numbers remain opaque to the personalization engine.

Workbook Rules: A Belt-and-Suspenders Approach to AI Output

Workbook rules represent a fundamental shift in how Copilot interacts with mission‑critical spreadsheets. Instead of hoping the AI guesses correctly, authors define a JSON-based rule set that Copilot consults before finalizing any output. The rules are scoped to the workbook and can be authored by typing plain‑language statements into the new Rules Manager pane—no JSON knowledge required.

For example, a rule that reads “Never generate formulas that reference cells outside columns A through T” will prevent Copilot from accidentally pulling in a hidden helper column. Another common pattern: “If a PivotTable contains a date field, always add a timeline slicer.” The Rules Manager validates each statement for logic conflicts and offers a natural-language summary so non‑technical stakeholders can review the guardrails before approving them.

Commercial customers gain an extra layer: administrators can lock rules to specific compliance labels. A workbook tagged “QBR – Confidential” might inherit a rule that forces Copilot to redact customer names in any generated text summary. These label‑driven rules are managed through the Microsoft Purview compliance portal, unifying data protection across the entire Microsoft 365 suite.

Broader File Grounding for Commercial Users

Copilot’s ability to ground its answers in organizational data—the so‑called “Microsoft Graph grounding”—now extends to more than 30 additional analysis file types. Previously limited to Office documents and a handful of CSV and text formats, the grounding engine can now ingest data from .parquet, .avro, .orc, .json, .xml, and a variety of statistical package files including SAS, Stata, and SPSS formats.

This matters because a financial analyst asking Copilot “forecast Q4 revenue using the latest pipeline data” no longer needs to manually convert an exported .parquet file to .xlsx first. Copilot can read the raw file directly from SharePoint or OneDrive, extract the relevant schema, and even merge it with data already in the open workbook. Microsoft demoed a walk‑through in which the user dragged a 2 GB Parquet file from a Teams channel into the Copilot chat pane; within 45 seconds the AI surfaced summary statistics and, after a single follow‑up, plotted a 12‑month rolling forecast.

Administrators control which file types are available for grounding through the Microsoft 365 admin center. The default set includes all 30+ new formats, but IT can disable specific types if, for instance, the organization has no licensed tooling to audit SAS data sets. Grounding always respects existing SharePoint access controls—Copilot cannot read a file the user doesn’t have permission to open.

PivotTable Diagnostics Debut

Tucked into the update is a new PivotTable Diagnostics tool, accessible from the “PivotTable Analyze” tab. Aimed at power users who inherit sprawling workbooks from predecessors, the tool scans any PivotTable and produces a report identifying broken data connections, slicers pointing to deleted fields, formulas that reference stale cached values, and even redundant calculated items that bloat the workbook size.

The diagnostics output appears as an interactive task pane with clickable items that jump directly to the problematic cell or field list entry. A “Fix It” button appears next to many issues, and Copilot steps in to suggest repairs for the rest. In internal Microsoft testing, the tool reduced the time spent debugging a complex model by an average of 62 percent, according to data shared during the June technical community call.

Safer AI Workflows Through Admin Controls and Auditing

With AI generating more formula logic and data connections, the risk of inadvertent data leakage or compliance violations grows. Microsoft addresses this with a trio of security‑focused features debuting in the June release.

First, all Copilot‑generated formulas are now tagged with a unique metadata marker visible in the formula bar. When a user hovers over the marker, a tooltip shows the exact prompt that produced the formula and a “human‑verification” checkbox that, once ticked, registers in the compliance log as reviewed. This creates an auditable chain of custody for every AI‑authored calculation.

Second, the update introduces a “Sensitive Investigation Mode” that, when activated by an admin, prevents Copilot from generating any output that includes strings matching predefined patterns—social security numbers, credit card numbers, or custom regex patterns defined by the tenant. If a user asks Copilot to analyze a column that contains such data, the AI blocks the request and records the event in the unified audit log.

Third, Excel now integrates with Microsoft Purview Data Loss Prevention policies during Copilot sessions. If a DLP policy tags a cell as “Highly Confidential,” Copilot cannot reference that cell in any generated report or recommendation, even if the user has read access. The guardrail is enforced server‑side, so no amount of creative prompting can circumvent it.

Support for More Than 30 Analysis File Types Natively

Beyond the grounding engine, Excel’s Power Query backend has been updated to directly import over 30 new file formats without requiring external connectors. The list includes Parquet, Avro, ORC, Apache Arrow IPC, Feather, and modern JSON lines formats—all common in data engineering pipelines—plus the statistical formats mentioned earlier. Power Query’s M language gets new functions to handle nested structures in JSON and XML, and the Get & Transform experience now surfaces schema detection for semi‑structured files, letting users click to expand arrays and objects.

This is not merely a “Save As” convenience. By bringing these formats into the native import pathway, Excel unlocks them for Data Types, Power Pivot, and the new Python in Excel integration. A data scientist can pull a Parquet file straight into a Python cell, manipulate it with pandas, and return the result as an Excel table—all while Copilot watches and learns.

What the Update Means for Everyday Users

For the millions of information workers who live inside spreadsheets, the June 2026 update removes friction at three critical pain points: repetitive reformatting, unintentional AI mistakes, and the grind of hunting down PivotTable breakage. Copilot personalization should, over a week or two of use, start anticipating formatting needs so precisely that the AI suggestions become “It’s just the way I would have done it,” as one early tester at a Fortune 500 bank put it during a private preview call.

Workbook rules, while perhaps the least flashy feature, may prove the most transformative in regulated industries. Compliance officers at pharmaceutical and financial services firms have already begun crafting rule libraries that enforce GAAP formatting, rounding conventions, and disclosure language—rules that live with the file, not with the user, and survive email attachments and SharePoint migrations.

The broader file grounding likewise removes a major hurdle for organizations that operate data lakes. Dropping the need to pre‑process analytics files into Excel format cuts a step that, in large‑scale reporting, can consume hours of an analyst’s week. With DLP and Purview integration baked in, the security story is stronger than it was when Copilot first launched, which should encourage more cautious CISOs to green‑light the tool.

Looking Ahead

The June release is part of a cadence Microsoft has signaled for the second half of 2026, with Copilot becoming more context‑aware and more tightly woven into the compliance fabric of Microsoft 365. Upcoming builds will likely extend personalization to Word and PowerPoint, though Excel remains the proving ground because of the structured nature of spreadsheet data. Workbook rules are also expected to gain sharing capabilities—imagine a “Rules Gallery” where teams can publish and subscribe to each other’s rule sets, much like template sharing.

For now, the message to Excel power users is clear: Copilot is no longer a curious add‑on; it is becoming the primary interface through which analysis happens, and Microsoft is investing heavily in making that interface both smarter and safer.