AI-generated charts are only as good as the prompt and the data behind them. This post first covered Prompt To Data, PlotSet's in-app text-to-chart feature. That feature was retired in August 2026. The way to make charts from a prompt now is the PlotSet MCP server: you connect PlotSet to the AI assistant you already use, such as Claude, ChatGPT, Cursor, Claude Code or Codex CLI, and ask it for a chart. The chart lands in your PlotSet account, fully editable.
Below are ten prompts that work, what the agent does with each one, and the checks worth making before you publish.
How do AI-generated charts work in PlotSet?
You connect PlotSet once at mcp.plotset.com/mcp and sign in with OAuth; there is no API key. After that, your AI assistant can create charts from any of PlotSet's templates, upload data, fix column bindings, restyle, export PNG or SVG and return an embed link. The assistant supplies the data, so you check the numbers.
That last sentence matters. Model Context Protocol is an open standard that lets an assistant call tools; PlotSet's server gives it chart tools, not a database of facts. If you paste data, the chart shows your data. If you ask the assistant to find data, it uses its own search, and you should ask it to name the source.
The 10 prompts
1. Turn pasted data into a chart
Here is our monthly sign-ups CSV for 2025. Make a line chart in PlotSet titled "Monthly sign-ups, 2025", add the source "Product analytics" and give me the embed link.
The agent picks a line template, uploads the CSV, sets the title and source, and calls get_or_create_embed. This is the most reliable prompt of the ten because the data is yours. A line chart is the right default for one measure over time.
2. Build a chart from a named public source
Using ACEA's full-year 2025 figures, make a PlotSet pie chart of EU new car registrations by power source. Show percentages and put the source in the footer.
Naming the source does two jobs: it tells the agent where to look, and it gives you something to check. Here is a chart built that way through the MCP server, with the figures checked against ACEA's 2025 press release:
Hybrids took 34.5% of the market and battery-electric cars 17.4%, up from 13.6% in 2024. Six categories that add up to a whole is a fair job for a pie chart.
3. Ask for the right chart type, with a reason
Here are 14 countries and their average download speeds. Pick the best PlotSet template for this and tell me why before you build it.
The agent can list PlotSet's templates and their data slots. For fourteen named categories and one number it should land on a sorted horizontal bar chart, because long labels stay readable on horizontal bars. Asking for the reason first catches a bad choice before it is built.
4. Draw a decision chart with real thresholds
Make a PlotSet quadrant chart of EU countries: unemployment rate 2025 on x, real GDP growth 2025 on y, from Eurostat. Put the dividing lines at the EU averages and say so in the subtitle.
Quadrant lines only mean something if they sit on a stated value. In PlotSet's quadrant chart the lines cross at the centre of each axis range, so the agent sets custom axis limits centred on the averages, 6.0% and 1.4%:
5. Check the bindings
Open my chart "Q3 pipeline" and check that each column is bound to the right slot. Fix anything bound to the wrong field.
Bindings decide which column becomes labels and which becomes values, and auto-binding can get them wrong. The agent reads the chart, compares the bindings with the template's slots and rebinds. Run this prompt after any chart where the preview looks odd.
6. Restyle an existing chart
Restyle my "Q3 pipeline" chart: transparent background, our brand blue #3581F9 for the main series, grey for the rest, source line in the footer.
The agent reads the template's settings first, then changes only keys that exist. Give exact colour codes; "make it look nicer" gets you a guess.
7. Update the data and keep the design
Replace the data in my "Monthly sign-ups, 2025" chart with this new CSV that includes September. Keep the styling and the embed link.
Replacing data keeps the configuration and the embed URL, so any page that already embeds the chart updates on its own.
8. Export for a report or slide
Export that chart as an SVG and as a 1200 × 675 PNG.
PNG and SVG come straight back from the export tool. Use SVG for print and design tools, PNG for slides and email.
9. Turn a time series into a video
Turn this table of yearly values per country into a bar chart race in PlotSet. Show me a preview frame before you render anything.
The agent can pick a video template, build the layer and render a single preview frame for free. Ask for the preview first: rendering the final video uses your plan's video quota. See the bar chart race page for what the output looks like.
10. Make a set of matching charts
Make the same column chart for each of these five regions, same colours and axis range, and list the embed links in a table.
Consistent axes and colours are what make small multiples comparable. The agent repeats the build with one shared configuration and returns a table you can paste into a doc.
What makes a good chart prompt
| Include | Why it matters | Example |
|---|---|---|
| The data or its source | The server draws what it is given | "Using Eurostat's 2025 figures…" |
| The chart type, or a request to choose one | Avoids a default that does not fit | "a horizontal bar chart, sorted" |
| Title and source line | Readers need both, and agents skip them | "source: ACEA, 2025" |
| Exact colours | Brand colours are codes, not adjectives | "#3581F9 for the main series" |
| The output you want | Embed, PNG or SVG | "give me the embed link" |
| A check | Makes the agent verify its own work | "confirm the bindings before you finish" |
Always open the chart before you share it. AI assistants are quick at charts and poor at noticing when one misleads; we covered why in AI makes your charts now, and it can't tell when they lie.
FAQ
What happened to Prompt To Data?
PlotSet retired its in-app text-to-chart and image-to-chart features in August 2026. Prompt-based chart building now runs through the PlotSet MCP server, which works from the AI assistant you already use.
Which AI tools can make PlotSet charts?
Any client that supports remote MCP servers over HTTP. The MCP setup page has step-by-step instructions for Claude, ChatGPT, Cursor, Claude Code, Codex CLI and Hermes Agent.
Do I need an API key?
No. The first time your assistant calls a PlotSet tool, a browser window opens so you can sign in and approve access.
Can the AI find the data for me?
Only if your assistant has web search or file access of its own. PlotSet's server builds and styles charts; it does not look up facts. Ask the assistant to name its source and check the numbers.
Can I edit an AI-generated chart by hand?
Yes. Every chart is saved to your PlotSet account, so you can open it in the editor, change anything, then embed it or export it.
