Prompting AI for Data Visualisation (Matplotlib, Seaborn, Plotly)

A weak chart hides the story; a great one makes it impossible to miss. AI can produce excellent visualisation code in any of the major Python libraries — if you tell it what the data is, who the audience is, and what the chart needs to say. This topic gives you the prompt patterns to consistently get presentation-ready charts on the first try.

1. Introduction

Most data scientists know enough Matplotlib to produce a chart, but not enough to produce a good one. The difference shows up in the small things: clear axis labels, sensible tick intervals, colour-blind safe palettes, an annotated headline value, and a subtitle that summarises the takeaway. AI can handle all of that — usually faster than you can — but only if the prompt names the headline insight the chart should communicate. This tutorial shows you how to brief AI like a design editor briefs a junior designer.

2. The Concept Explained

A good visualisation prompt answers five questions: what data, what library, what chart type, what story, and what audience. If any of those are missing, the output is generic. Beginners often skip "what story" — and so produce technically correct charts that fail to land with stakeholders. Always state the headline takeaway you want the chart to support: "show that churn rises sharply for plan_type=basic after month 3" is far more useful than "plot churn over time".

It helps to think of a chart as a sentence. Data is the vocabulary, the chart type is the grammar, and the story is the meaning. AI can construct fluent sentences quickly, but it needs you to choose the meaning.

Matplotlib, Seaborn, or Plotly?

The library choice changes the prompt. Matplotlib is the workhorse — verbose but flexible, ideal for paper figures and notebook artefacts. Seaborn wraps Matplotlib with statistical defaults — perfect for EDA. Plotly produces interactive HTML — ideal for dashboards and stakeholder-facing reports. Always name the library at the top of the prompt so the AI doesn't waste tokens explaining choices.

3. The Problem Without This Technique

Plot my sales data.

Act as a senior data analyst preparing a board-ready chart
in Seaborn (Matplotlib backend).

DataFrame: orders_df with columns
  order_date (datetime64), region (str: NA|EU|APAC),
  net_revenue_gbp (float64).

Chart goal: show that EU revenue overtook NA revenue
in Q3 of the current year and is now widening the gap.

Requirements:
- Line chart, x = month, y = net_revenue_gbp, hue = region.
- Aggregate by month (pd.Grouper, freq='MS') then sum.
- Title: "EU overtakes NA in Q3 — gap widening into Q4".
- Annotate the crossover point with an arrow + month label.
- Use a colour-blind safe palette (Okabe-Ito).
- Despine top and right.
- Output figsize=(10, 5). Save as 'eu_overtake.png' at 200 dpi.
- Return only the code.

4. The Solution

The pattern is: DataFrame brief → library → chart type → story → styling rules → output target. The story line is the one most people forget; without it, AI produces a chart, not an argument. Once you make the chart's narrative explicit, the AI starts choosing annotations, titles, and emphasis colours that support that narrative.

For Plotly, add one more constraint: state whether you want a plotly.express one-liner (fast EDA) or a plotly.graph_objects figure (more control, dashboard-ready). For Matplotlib, mention whether axes should follow the journal/paper style (grid off, serif font) or the slide-deck style (bold title, larger fonts).

5. Step-by-Step Breakdown

1. State the headline insight in one sentence. "Show that churn rises after month 3 on the basic plan." This single sentence guides every other decision.
2. Brief the DataFrame. Name, columns, dtypes — exactly as for Pandas prompts. The AI needs to know what's available before it can plot.
3. Choose the library on purpose. Seaborn for stats, Matplotlib for full control, Plotly for interactive. Naming it stops the AI hedging across all three.
4. Specify the chart geometry. Type, x, y, hue/colour, facet — using the same vocabulary the library uses keeps the code idiomatic.
5. List styling requirements. Title, axis labels, palette, font size, despine, figsize. Each item closes a degree of freedom and improves the output.
6. Demand the output target. Inline display, PNG at a specific DPI, or HTML for a dashboard — the AI will tailor savefig or fig.write_html accordingly.

Tip: Keep a "house style" prompt fragment that lists your preferred palette, font, despine settings, and figsize. Paste it at the top of every visualisation prompt and your charts will look consistent across an entire report deck.

6. Practice Exercises

Critique this chart for clarity and accessibility, then rewrite it. Improvements should include: clearer title summarising the insight, colour-blind safe palette, larger axis labels, despined frame, and an annotation highlighting the key data point.

7. Key Takeaways