Analyzing data with ChatGPT – OpenAI Academy

TL;DR

OpenAI Academy published a guide on April 10, 2026 that explains how to use ChatGPT for data analysis both directly in the chat and inside Excel or Google Sheets, providing a workflow for uploading data, asking focused questions, requesting visualizations, and verifying results.

Using ChatGPT for direct data analysis

You can move from raw data to insights in ChatGPT with minimal setup by uploading a CSV or Excel file, pasting a table, or connecting a supported data source, then asking questions in plain language. The process works best when you start with the decision you are trying to support, provide the data plus any critical context such as definitions, timeframe, and column meanings, ask for an approach rather than just an answer, request visuals explicitly if they would help, and ask for reusable outputs like a clean table or an executive summary. The guide includes example prompts for three common tasks: summarizing key insights from a Shopify store dataset, analyzing sales funnel data from a connected analytics app, and identifying process inefficiencies using ticket data and a process document.

Using ChatGPT in Excel or Google Sheets

If most of your work occurs inside a spreadsheet, the ChatGPT for Excel and Google Sheets add‑in provides a sidebar where you can ask for changes in plain language and see the results directly in the sheet. Effective prompts specify which tabs to use, which tabs to avoid changing, the desired output, and whether you want ChatGPT to explain its plan before editing. Example prompts cover cleaning up a workbook for executive review, explaining how revenue flows through a model into margin and EBITDA, updating assumptions on the Inputs tab only, building a monthly KPI pack from Revenue, Opex, and Headcount tabs, and preparing for review by outlining the plan before any edits.

Tips for successful analysis

Help ChatGPT help you by stating what “good” looks like up front, including the success metric you care about, the timeframe, and the groups or segments you want to compare. If the numbers matter, ask it to show its work: the assumptions it made, any formulas it used, and quick checks for missing data or unusual spikes. Set simple ground rules to keep the analysis trustworthy, such as telling it not to treat correlations as causes, to point out data limitations, and to flag anything that looks off. Before sharing results or making a decision, do a quick reality check by spot‑verifying a couple of key numbers to ensure everything adds up.