Microsoft brings Copilot LLM directly into Excel cells with new function

Microsoft is set to revolutionize data interaction within its ubiquitous Excel spreadsheets by integrating advanced large language model (LLM) capabilities directly into individual cells. This new Copilot function promises to streamline a host of tasks, from sophisticated data analysis and text classification to dynamic content generation, all without the need for manual editing or toggling between separate applications.

The core of this innovation lies in a straightforward formula: =COPILOT(prompt_part1, [context1], [prompt_part2], [context2], ...). Users can simply type their requests in natural language, optionally referencing specific cell ranges to provide context for the AI. For instance, classifying customer feedback on a product becomes as simple as entering =COPILOT("Classify this feedback", D4:D18). Demonstrations have also shown Copilot adeptly looking up information, such as airport codes across the United States, based on a user’s prompt.

A significant leap forward, this Copilot function is seamlessly built into Excel’s underlying calculation engine. This means that as soon as the source data changes, the AI-generated results update automatically, eliminating the need to manually re-run scripts or refresh add-ins. Furthermore, the function integrates smoothly with existing Excel formulas like IF, SWITCH, WRAPROWS, or LAMBDA, allowing the output of other calculations to feed directly into Copilot prompts, enabling more complex and interconnected analytical workflows.

Microsoft highlights several key applications for this powerful new capability. Users will be able to generate ideas, transforming product descriptions into SEO-optimized keywords or rewriting marketing copy with ease. The function can also summarize large datasets or distill complex calculations into plain language for reports, making data more accessible. Text classification becomes effortless, allowing for the rapid sorting of customer feedback, support tickets, or survey responses directly within the spreadsheet. Additionally, Copilot can assist in creating structured lists or tables that align with specific data models, generating test data, industry benchmarks, or even project plans.

Addressing crucial concerns regarding data privacy and security, Microsoft assures users that any data processed through Copilot is not utilized to train its AI models. The information remains private, used solely for generating the requested results. It’s also important to note that, for now, Copilot does not access live web data or internal company files; any external information must first be imported into Excel. Microsoft has indicated plans to remove this limitation in future updates, significantly expanding the tool’s reach.

There are current usage limits, with users restricted to 100 calls every ten minutes, totaling up to 300 calls per hour. To optimize usage, Microsoft advises employing arrays rather than single cells, as a single call covering a larger data range still counts as only one use.

Currently, the new Copilot function is available exclusively to Beta Channel users holding a Microsoft 365 Copilot license, requiring specific Windows (version 2509, build 19212.20000 or newer) or Mac (version 16.101, build 25081334 or newer) versions. The web version is expected to follow suit through the Frontier program. While the exact OpenAI model powering Copilot for Excel is not explicitly stated, it is likely GPT-4o, consistent with other Office AI features, with plans for an upgrade to the GPT-5 family in the near future. This on-cell integration marks a significant evolution from earlier Copilot iterations, which were limited to a chat sidebar, often disrupting the user’s workflow by forcing a constant switch between chat interface and spreadsheet data.