How to Use ChatGPT and AI With Excel: Choose the Right Connection

You can use AI to summarize customer comments, draft descriptions or explain an Excel formula. First choose the connection you actually need: ChatGPT working with a spreadsheet, Microsoft’s Copilot inside Excel, or a custom OpenAI API workflow. These have different accounts, permissions and bills. Power Query’s From Web command does not connect a ChatGPT subscription to your cells by itself.

Start safely: work on a copy, use a small non-sensitive sample, preserve row IDs, and put generated text in new columns. Do not overwrite the only copy of your source data.

Choose Between ChatGPT, Microsoft Copilot and the OpenAI API

• ChatGPT: suitable for an interactive task with an uploaded workbook or an available supported spreadsheet integration. OpenAI also offers ChatGPT for Excel; check the current integration and your account’s availability rather than installing an add-in merely because its name contains “GPT.”
• Copilot in Excel: Microsoft’s built-in assistant. Its entitlement comes from your Microsoft account or organization, not a ChatGPT subscription.
• OpenAI API: a developer integration for repeatable processing. Requests authenticated with your standard OpenAI API key use API billing. Paying for ChatGPT does not fund that API project.

OpenAI distinguishes subscription usage from API pricing. Review the specific integration’s authentication and billing before authorizing it; an API-key connection and an eligible ChatGPT-plan connection are not interchangeable.

Use ChatGPT for an Excel Text Task Without Writing an API Query

1. Prepare a copy containing only the rows and columns needed. Include a stable ID for every row and remove information you are not authorized to send.
2. Attach the spreadsheet in ChatGPT. Identify the worksheet, source column and desired output column.
3. Start with a small request, for example: “For rows 2–6 on Feedback, summarize each comment in column B in one sentence. Return the original ID and a new Summary column. Do not change source text or invent missing details.”
4. Review the results against the original comments. Check that every returned summary still belongs to the correct ID.
5. Ask for the remaining approved rows and an Excel or CSV result. Inspect the exported file before merging it into the working workbook.

This is a file-based workflow, not a live connection: later edits to your original workbook are not automatically included. OpenAI’s usage guide explains attachments and reviewing generated files.

Use Copilot Directly in Excel

Microsoft lists eligible Personal or Family subscriptions with an AI credits plan, Premium, commercial Copilot subscriptions and eligible business/enterprise Copilot Chat subscriptions. Exact features vary by license, account and tenant. For shared consumer subscriptions, do not assume every family member inherits the owner’s AI benefits. Check Microsoft’s current Excel requirements and your license.

1. Open a copy of the workbook in a supported, updated Excel application and sign in with the eligible account.
2. Open Copilot and name the relevant sheet and range. Ask it to summarize a few rows into a new column while leaving the originals intact.
3. Choose a chat or planning experience, when available, if you want a read-only proposal first. Editing experiences may change the workbook directly; do not assume a universal approval step or an Insert button.
4. Inspect the actual workbook edits, including formulas, totals and row alignment. If the changes are wrong, use Undo or an available version-history restore, and keep the unchanged copy until the result is verified.

If Copilot is unavailable, check the account and license, app version, organization policy and connected-experience requirements. A visible ribbon icon alone does not prove full access.

When a Custom API and Power Query Workflow Makes Sense

For a recurring batch job, ask a developer to build an authenticated service that reads approved rows, calls the OpenAI API, and returns validated results keyed by row ID. Store the secret in a server-side environment variable or secret manager, not in a shared workbook, cell, query parameter or pasted screenshot.

Power Query can import the service’s finished results into Excel. Calling a paid model from a refreshable row function requires additional engineering: request deduplication, explicit run controls, error handling, result storage and usage monitoring. A refresh or retry must not silently regenerate and rebill every row.

Web.Contents supports HTTP requests, but its POST authentication constraints and your service’s credential scheme must be designed together. Power Query’s “Anonymous” credential mode does not mean the OpenAI API accepts unauthenticated requests. Response parsing must also match the chosen endpoint; a generic instruction to expand a field named choices is not a complete implementation.

Test costs and quality with a small batch, check current API pricing, and retain successful outputs so a failed job can resume without repeating completed rows. For a one-off writing task, the interactive methods above usually need less custom infrastructure.

Official documentation checked: October 14, 2026. Product access and labels can differ by account and release.