How to Set Up ChatGPT for Excel

ChatGPT for Excel is a native Office Add-in that connects your OpenAI account to Microsoft Excel, letting you build spreadsheet models, generate charts, and explain formulas through plain-language prompts. It launched in March 2026, requires ChatGPT Plus ($20/month) or higher, and is available in the US, Canada, and Australia. Setup takes under five minutes.

What the Add-in Does

The add-in runs as a sidebar inside Excel. You type a request in plain English and it reads your active worksheet, proposes changes, explains what it plans to do, and asks permission before editing any cells. Core capabilities include:

• Building financial models from a plain-language description
• Running scenario analysis on existing data (for example, testing three different growth rate assumptions)
• Explaining or rewriting formulas in plain language
• Generating charts from selected data
• Pulling live financial data through integrations with FactSet, Dow Jones Factiva, LSEG, and S&P Global (enterprise tier)

The financial data integrations are aimed at professional finance teams. For most analysts and ops managers, the practical value is in formula explanation, model scaffolding, and chart generation on clean datasets.

Requirements Before Installing

ChatGPT plan. The add-in requires ChatGPT Plus ($20/month), Pro, Business, Enterprise, Edu, or Teachers. The free tier does not support it.

Microsoft 365. You need an active M365 subscription. The add-in installs through the Office Add-ins store, which is part of the M365 environment.

Region. ChatGPT for Excel is available in the United States, Canada, and Australia at launch. Users in the EU cannot access the add-in as of April 2026. No rollout timeline for other regions has been published.

Operating system. The add-in works on Excel for Windows and Mac. Support for browser-based Excel has not been confirmed.

Enterprise and Edu workspaces. If your organization uses a Microsoft Business, Enterprise, or Edu account, the add-in is disabled by default. An IT administrator must enable it through role-based permissions in the Microsoft 365 admin center before individual users can install it.

Two Ways to Install

Via the Home Ribbon

1. Open Excel on your computer.
2. Click the Home tab in the ribbon.
3. Select Add-ins from the ribbon options.
4. In the search box, type ChatGPT.
5. Select ChatGPT for Excel and click Add.
6. Sign in with your OpenAI account credentials.

The sidebar appears on the right side of Excel and persists across sessions.

Via the Insert Menu

1. Open Excel.
2. Click Insert in the ribbon.
3. Select Get Add-ins.
4. In the Office Add-ins dialog, click the Store tab.
5. Search for ChatGPT for Excel.
6. Click Add, then click Continue to confirm.
7. Sign in with your OpenAI account.

Both methods install the same sidebar. The Home ribbon path is faster on Windows. The Insert menu path is more consistent across platforms.

Using the Add-in

Once the sidebar is open, type a request in the text field. Practical starting points that work reliably:

• "Build a simple revenue projection model using the data in columns A through C."
• "Explain what this formula does: =SUMIFS(D2:D100,A2:A100,G2)"
• "Create a bar chart comparing Q1 and Q2 totals by region."
• "Run three scenarios on the growth rate in cell B5: 5%, 10%, 15%."

ChatGPT reads the active sheet, lists the cells it plans to change, and waits for your confirmation before applying anything. This confirmation step is not optional and cannot be turned off. That design choice matters given the accuracy issues described in the limitations section.

Thinking Modes

The add-in offers three response modes, introduced after users reported slow response times at launch:

• Fast: prioritizes speed with lighter reasoning
• Standard: balanced default for most tasks
• Heavy: deeper reasoning, slower output

For formula questions and single-chart requests, Fast mode returns results in seconds. For multi-step financial models or complex scenario analysis, Standard or Heavy mode produces more accurate outputs. Begin with Standard and switch to Fast when the task is straightforward.

Limitations Confirmed at Launch

The launch discussion on Hacker News documented several issues not covered in OpenAI's official documentation. These are worth knowing before using the add-in in any high-stakes context.

Latency on complex models. Users building multi-tab financial projections reported response times of 15 to 20 minutes. OpenAI acknowledged the issue and noted that Fast mode was introduced to let users trade reasoning depth for speed. For simple tasks, the add-in is fast. For models with many interdependencies, expect meaningful wait times.

Cell reference errors. One user building a financial model found approximately 30 incorrect cell references in a single output. The add-in pointed formulas at the wrong input rows, likely because blank spacer rows in the workbook disrupted its structural reading of the sheet. The errors were not obvious visually. When confronted directly, ChatGPT acknowledged the mistakes. This means every formula the add-in generates needs manual verification before acceptance, particularly in revenue models or cost trackers that feed downstream calculations.

Hardcoded values. Separate user reports documented cases where the model inserted hardcoded numeric values into formulas instead of referencing the correct input cells. A hardcoded 1.12 instead of a reference to =$B$5 looks correct on screen but breaks the model the moment the assumption changes. Trace inputs manually on any accepted formula.

Single-sheet context. The add-in reads only the active worksheet. It has no awareness of other sheets in the same workbook, other open workbooks, or linked external data. For multi-tab models, you need to switch sheets manually and re-prompt for each section.

Row scale. Performance degrades above 50,000 rows. For large datasets, expect slower and less reliable results.

OpenAI's own documentation states that "certain responses may take longer than expected while the company improves performance," which covers all of the above.

ChatGPT for Excel vs. Microsoft Copilot

Many Excel users already have access to Microsoft Copilot through their M365 subscription and may not realize the two are different products.

Microsoft Copilot is embedded in M365 at $30 per user per month. It requires no separate subscription and covers the full Office suite including Word, PowerPoint, Outlook, and Teams. For organizations that want AI across all Office apps from a single budget line, Copilot has less friction.

ChatGPT for Excel runs on GPT-5.4 and includes live financial data integrations with FactSet, Dow Jones Factiva, LSEG, and S&P Global. Those integrations are the reason finance teams in particular are evaluating the add-in over Copilot. For users who spend most of their time in Excel with financial or market data, ChatGPT for Excel offers tools that Copilot does not.

The tradeoff: ChatGPT for Excel requires a separate OpenAI subscription on top of M365, whereas Copilot is a single line-item. If you primarily need Excel-specific AI help and work with financial data, ChatGPT for Excel is the stronger tool. If cross-app coverage matters, Copilot is simpler to manage.

A Note on Alternatives

If your work involves querying structured data across multiple sources without staying in a spreadsheet interface, dedicated tools built around plain-English data querying handle this without requiring formula knowledge. VSLZ AI, for example, takes uploaded data and returns insights, statistical analysis, and charts from a single prompt, which removes the formula-verification step entirely. Whether that approach suits you depends on whether Excel is your fixed working environment or just one option among several.

Practical Starting Point

Install the add-in on a copy of a non-critical workbook. Test with a formula explanation request first. It is the most reliable use case and gives you a clear sense of how the sidebar confirms changes before you attempt anything involving model-building. Once you are comfortable with the confirmation flow, move to scenario analysis on lower-stakes data.