ChatGPT prompts can speed up most of the work around a chart: picking the form, cleaning a messy table, choosing colours and writing the headline. They work best when you give the model your real column names, your audience and the point you want to make. This guide lists ten data visualization prompts we would use today, and what to check in each answer.
How do you use ChatGPT for data visualization?
Use ChatGPT for data visualization by describing your data, audience and message, then asking for one specific decision: a chart type, a cleaning step, a palette or a caption. Paste column names and a few sample rows. Then check every number and suggestion against your source before you publish.
A vague prompt gets a generic answer. Compare these two:
| Weak prompt | Better prompt |
|---|---|
| What chart should I use? | I have 12 months of revenue for 3 products. I want to show that product B overtook product A in June. The audience is our board. Which chart fits, and why? |
| Make my chart better. | This column chart has 14 categories in alphabetical order and 5 colours. The message is that two regions lead. What should I change? |
The OpenAI prompting guide gives the same advice in general terms: be specific about context, format and the result you want.
Ten ChatGPT prompts for better charts
Replace the text in square brackets with your own details.
1. Choose the chart type
I have [describe columns and row count]. The one thing I want readers to take away is [message]. Suggest the best chart type and one alternative, and explain what each would hide.
Asking what each option hides gets you past the default answer. Our guide to choosing the right chart for your data is a useful second opinion. For market share among a few companies, for example, the usual answer is a pie or donut chart:
2. Clean a messy table
Here are the first 20 rows of my CSV: [paste]. List the problems that would stop this from charting cleanly, such as mixed date formats, numbers stored as text, merged headers and duplicates. Then give me the fixed table.
Check row counts after cleaning. Models sometimes drop or merge rows without saying so.
3. Reshape data for a template
Convert this table from wide format to long format, with columns [A, B, C]. Keep every value unchanged and don't add rows.
Most chart templates are specific about layout, so this saves a lot of manual work. Compare totals before and after.
4. Normalise values for a fair comparison
I'm comparing [metric] across [countries or groups] of very different sizes. What should I divide by so the comparison is fair, and what would each choice change?
Raw totals often just show which places are largest. The map below uses prisoners per 100,000 people rather than total prisoners, so a small country can rank above a large one.
5. Build a palette with a purpose
Suggest a palette for a [sequential, diverging or categorical] chart showing [topic], as hex codes. It must work for colour-blind readers and have enough contrast on a [white or dark] background.
Don't accept the result on trust, because generated hex codes don't always match their names. Test them with ColorBrewer or a colour-blindness simulator, and read our guide to designing colour-blind-friendly charts.
6. Write a headline that states the finding
This chart shows [data]. The main finding is [finding]. Write five headlines of under 10 words that state the finding, not the topic.
"Rents rose faster than wages" beats "Wages and rents over time". Check that any number in the headline is in your data.
7. Write alt text
Write alt text of under 40 words for this chart. Include the chart type, what it shows and the main trend: [describe the chart or paste the data].
Good alt text states the finding instead of listing every value. The W3C accessibility guidelines cover text alternatives and contrast in detail.
8. Find what stands out
Here's my data: [paste]. List the three largest outliers and the biggest change between periods. Show how you worked out each one.
Asking for the working makes errors easier to spot. Once you know the outlier, highlight it. Here, the map's colour scale does that job, and readers go straight to the darkest countries.
9. Critique a finished chart
Here's a description of my chart: [type, axes, colours, labels, order]. Critique it as a data editor. List what could mislead a reader, then suggest three fixes in order of impact.
If you can upload an image, include it. Then compare the answers with our list of data visualization dos and don'ts.
10. Adapt a chart for mobile
This chart has [number] categories and [number] series and will be read on a phone. What should I remove, merge or rotate so it stays readable at 360 pixels wide?
The usual answer is fewer labels and horizontal bars, which is also a good default for long category names. A horizontal bar chart handles this well.
What to check in every AI answer
ChatGPT writes plausible text, and plausible isn't the same as correct. Before you publish, check:
- Numbers. Recalculate totals, percentages and rankings. Never use a statistic from the model without finding its source.
- Rows. Compare row counts and totals before and after any data change.
- Sources. Cite the original dataset, not the chat.
- Chart rules. Bars should start at zero, and a pie's slices must add up to a whole.
- Private data. Check your company's policy before you paste customer or financial data into any AI tool.
From prompt to published chart
Once the data and plan are ready, upload a CSV to PlotSet, paste the table, or link a Google Sheet, then pick a template. Our bar chart guide shows the steps from start to finish.
You can also let your assistant do the building. The PlotSet MCP server connects AI clients that support remote MCP, including ChatGPT and Claude, to your PlotSet account, so the assistant can create a chart, set its data and export it for you.
Frequently asked questions
Can ChatGPT make charts?
Yes. With data analysis tools turned on, ChatGPT can generate static charts from a file you upload. For interactive charts you can embed, connect it to a chart tool. PlotSet does this through its MCP server.
What is the best prompt for data visualization?
The best prompt names the data, the audience and the message, then asks for one decision. "Which chart shows that B overtook A in June, for a board audience?" works far better than "visualise this".
Can ChatGPT analyse a CSV file?
Yes, when file upload is available on your plan. Ask it to show its working, and verify a few results yourself in a spreadsheet, because it can misread formats and make calculation errors.
Is it safe to paste data into ChatGPT?
Treat it like sending the data to a third party. Remove personal or confidential details, and check your organisation's AI policy and your account's data settings first.
Should I use the charts ChatGPT generates?
They're fine for exploring data. For publishing, a dedicated tool gives you more control over labels, branding, interactivity and embedding, and you can export SVG, PNG or JPG.
