Analysing data and spreadsheets
Upload or paste data, ask questions in plain English, make charts and, most importantly, check the numbers independently.
In this lesson you will
- Prepare data for analysis
- Ask clear questions and request the working
- Verify numbers independently
- Read charts critically
About 18 minutes
Ask your spreadsheet a question in plain English
Nadia has a sheet of four months of sales and a shaky memory of how pivot tables work. She wants to know which product earns more and which month was best. A year ago that meant an hour on the internet looking up formulas. Now she uploads the file and types the question. Many versions of ChatGPT can read spreadsheets and often run small programs behind the scenes to calculate answers and draw charts. This lesson shows how to use that power well, and, crucially, how to check it, because numbers are the one place where a confident wrong answer does real damage.
Data analysis features, file limits and chart tools depend on your plan and version. If you cannot upload a file, you can paste a small table of data directly into the chat.
Step 1: prepare your data
The cleaner the input, the better the output. Before you upload:
- Put the data in one table with a single header row and no blank rows or merged cells.
- Give columns clear names: units, price, not Column 3.
- Keep one type per column (all numbers, all dates, all text).
- Remove private information: customer names, phone numbers, emails. Replace with codes if you need to tell rows apart.
- Keep your original safe and upload a copy.
Here is the small sample we will use through this lesson. You can paste it into a chat or save it as a CSV file.
month,product,units,price
Jan,Candle,40,900
Jan,Soap,25,350
Feb,Candle,55,900
Feb,Soap,30,350
Mar,Candle,35,900
Mar,Soap,60,350
Apr,Candle,70,900
Apr,Soap,45,350
Step 2: ask for a description first
I have uploaded sales.csv. First, describe the data: columns, number of rows, any missing or odd values. Do not analyse yet.
This catches problems early: text where numbers should be, blank cells, duplicates. A short first look prevents confident analysis of broken data.
Step 3: ask clear questions
Add a column called revenue = units x price. Then show: total revenue per product, total revenue per month, and the best month. Show the table of results and the calculation you used.
Notice the pattern: define any new column, name what you want, and ask it to show the working. For a data task, the working matters as much as the answer.
Step 4: verify with an independent check
This is the step beginners skip and professionals never do. Check the result another way. For this tiny sample, here is the correct answer computed independently with a short Python program, so you can see what a right answer looks like:
Revenue Candle: 180000
Revenue Soap: 56000
Total revenue: 236000
Jan 44750
Feb 60000
Mar 52500
Apr 78750
Best month: Apr
If ChatGPT’s answer matches these figures, you can trust it for this question. If not, ask it to show its steps and find the difference. You can do the same quick check with a calculator or a spreadsheet formula: =SUMPRODUCT over the units and price columns gives the total revenue.
Even if you never check everything, recompute one or two figures yourself: a single row, one subtotal. If those match, the method is probably right. If they do not, stop and investigate.
Step 5: charts
Make a bar chart of revenue by month with clear labels, a title and the numbers on top of each bar. Use two colours, one per product, in a stacked chart. Then explain in one sentence what the chart shows.
- Ask for a title, axis labels and units.
- Choose the chart type: bars to compare, lines for change over time, pie only for a few parts of a whole.
- Look closely: do the bar heights match the numbers? Is the vertical axis starting at zero?
- Ask for the chart as an image or a file you can download, if your version allows it.
Questions that uncover insights
Looking at the revenue data, what are three interesting patterns or surprises? For each one, say how confident you are and what extra data would test it.
A pattern in eight rows may be pure chance. Good analysts always ask how much data supports a conclusion, and good prompts invite that scepticism.
Cleaning messy data
The column 'date' has dates in different formats (3/4/2026, 2026-04-03, 'April 3'). Convert them all to YYYY-MM-DD. List any rows you were not sure about instead of guessing.
- Ask it to list the changes it made so you can audit them.
- Be careful with dates: 3/4/2026 means March 4 in some countries and 3 April in others. Tell it which format you use.
- Ask for the cleaned data as a new file; never overwrite your original.
Formulas and spreadsheet help
You do not have to upload anything to get help with formulas. Describe your sheet and the result you want.
In Google Sheets, column B has units and column C has price. In column D I want revenue, and in cell F1 the total revenue. Give me the exact formulas and explain each in a sentence.
D2: =B2*C2 (then drag down)
F1: =SUM(D2:D9)
Explanation: D2 multiplies units by price for one row; SUM adds the whole revenue column. Alternatively, =SUMPRODUCT(B2:B9,C2:C9) gives the total in a single cell.
What can go wrong
Trusting a number you did not check. Wrong column picked, a text value treated as zero, a filter applied silently: the answer looks professional and is simply wrong.
Uploading sensitive data, such as customers’ personal details or confidential finances, without permission. Remove or anonymise first and follow your organisation’s rules.
Over-reading small samples. Four months of sales cannot show a seasonal pattern.
A data checklist
- Clean copy of the data, private details removed.
- Ask for a description of the data first.
- Define new columns and show the working.
- Verify at least one number independently.
- Look at charts critically, including the axis.
- Say how confident you are, and what you cannot conclude.
- Keep the original file untouched.
Check what you learned
Five quick questions. Pick an answer and you will see straight away whether it is right and why.
What should you do before uploading a spreadsheet?
- Merge all the cells
- Add colours
- Clean it and remove private details
- Delete the headers
Clean data gives better results and protects privacy.
Why ask for a description of the data first?
- To catch odd or missing values early
- To get a chart
- To make it longer
- To change the file
Problems are easier to fix before analysis.
What does independent verification mean?
- Asking the same question twice
- Checking the result another way
- Trusting the first answer
- Using a bigger font
Recompute with a calculator, a formula or a script.
What should you look at in a chart?
- Only the colours
- The title font
- Nothing
- Whether bars match the numbers and the axis starts at zero
Charts can mislead if the scale is off.
Dates like 3/4/2026 are risky because…
- They are too long
- They mean different dates in different countries
- They are numbers
- They have slashes
Tell it which date format you use.
Summary
- Clean and anonymise data first.
- Describe before analysing.
- Always verify at least one result another way.
- Charts and small samples can mislead.
Questions
Can ChatGPT replace Excel?
No. It is a helper for questions, formulas and quick charts. Spreadsheets remain best for ongoing, shared, auditable work.
How big can my file be?
Limits depend on your plan and change over time. For very large data, summarise or sample it first, or use a spreadsheet or database tool.
Why did it give different totals on a second try?
If the method is not stated, it can choose different interpretations. Ask it to show the calculation, and ask again with exact definitions.
