How do you analyse data in ChatGPT with interactive tables and charts?

ChatGPT has been able to work with data files for a while, but on May 16 OpenAI announced changes that make it far easier to see what is going on. You can now pull files straight from Google Drive or Microsoft OneDrive, explore your data in an interactive table, and shape a chart before you download it. The upgrades come with GPT-4o and are rolling out to paying subscribers on the Plus, Team and Enterprise plans during the next few weeks. Free users who are on GPT-4o can also analyze data and create charts, within their message limits.

Under the hood, nothing magic happens. When you ask a question about your data, ChatGPT writes a short program in Python, a common programming language for data work, runs it, and reports the result. That is useful to know, because it means every answer has a method behind it that you can ask to see.

What is new

Chart types that are not interactive yet still appear, as a regular image.

A worked example: a club's budget

Imagine you are treasurer of a community club and want to see how this year's spending compared with the budget. Your spreadsheet has three columns. The numbers are illustrative.

1. Add the file. Use the attachment button to pick the spreadsheet from Google Drive, OneDrive or your computer. ChatGPT opens it as an interactive table.
2. Ask a clear question:

This is our club's budget for the year, with planned (Budget) and real (Actual) spending in dollars.
1. Add a column for the difference (Actual minus Budget) and a column for the difference as a percentage of Budget.
2. Give the totals for Budget, Actual and Difference.
3. Make a bar chart comparing Budget and Actual for each category.
4. In three sentences, explain where we overspent and by how much.
Show the code you used.

1. Read the table. Expand it to full screen and check that all five categories are there. Click the Equipment row and ask, "Why is this the largest difference?" to see how it explains it.
2. Adjust the chart. Hover over the bars to see each value, change the colours if you like, and download it for the annual meeting.

For this data, the correct results are easy to work out by hand: the budget totals 3,250 dollars, actual spending 3,600, so the club spent 350 more than planned. Equipment was 250 over (about 31%), hall rental 150 over (12.5%) and printing 40 over (16%), while snacks came in 90 under and insurance was exactly on budget. Those figures are your answer key for the next section.

Keeping your data private

Spreadsheets often hold personal details: names, addresses, salaries. Before you upload, remove what the analysis does not need. OpenAI says data from its Team and Enterprise business plans is not used for training, and Plus subscribers can opt out of training in their Data Controls settings. If your file belongs to an employer or organization, check its rules first.

How do you check the numbers it reports?

Start with the cheapest check: recompute one total yourself. In the budget example, a calculator or a spreadsheet SUM tells you in seconds whether 3,600 is right. If one headline number matches, the method is probably sound; if it doesn't, stop and ask why before going further.

Next, check that nothing went missing. Ask ChatGPT, "How many rows did you read, and what are the column names?" A wrong count can mean a header was read as data, a blank row cut the file short, or a merged cell confused the import. The interactive table makes this easier, because you can scroll the full dataset instead of trusting a summary.

Then look at the method. Since ChatGPT works by running code, ask it to show the code and read the key line in plain words: which columns were added, which rows were filtered out. You do not need to know Python to spot "only rows where Category is not empty" and wonder whether that dropped something.

Finally, separate the numbers from the story. The calculated figures come from code, which is usually reliable once the input is right. The written explanation is the model's interpretation, and that is where confident-sounding guesses can creep in. If it claims equipment costs rose "because of new tables," that may be plausible, but it is not in your data. Keep the figures you have checked, and treat the reasons as questions for whoever approved the spending.