Last week, OpenAI made many previously paid features free for everyone.

One of those was data analysis (formerly “Code Interpreter”).

This means that—along with some fun use cases (see my deep dive)—ChatGPT can process data to produce all sorts of charts.

How many sorts?

At least 12, as per this article from OpenAI:

Line graph

Bar chart

Pie chart

Histogram

Scatter plot

Box plot (box-and-whisker plot)

Heatmap

Area chart

Radar chart

Treemap

Bubble charts

Waterfall charts

I’ll be honest: A bunch of those didn’t mean anything to me when I started this post. They were *uncharted territory* if you will.

They may as well have said something silly like “hat graph” or “sunburst chart.”1

But I figured it’d be a fun and useful exercise to have ChatGPT showcase them using fake data.

Now, armed with visuals, I’m here to walk you through the 12 primary chart types ChatGPT can make and what you might want to use them for.

Let’s get graphing!

## What are “interactive charts” in ChatGPT?

The help article from OpenAI points out that four chart types2 are “interactive”:

Bar

Line

Pie

Scatter

But what exactly is the difference between a static and an interactive chart?

As far as I can tell, the two main features of interactive charts are that they let you:

Hover your cursor over the chart to pinpoint specific values:

Change chart colors via a dropdown menu:

That’s pretty much it at this stage.

## The 12 primary types of ChatGPT charts

All charts below were made by ChatGPT using fake data it invented upon my request.

Also, because I’m a child, every chart is a meta nod to itself.

### 1. Line graph (interactive)

A line graph shows changes in something over time by connecting time snapshots with lines. You can plot more than one line, which is great for e.g. tracking how your mood changes as you consume more caffeine.3

### 2. Bar chart (interactive)

A bar chart compares different categories using rectangular bars. The higher the bar, the more of that thing there is. Perfect for figuring out which co-worker drinks the most coffee on any given day. (Spoiler alert: It’s Greg.)

### 3. Pie chart (interactive)

A pie chart divides your data into “slices” as proportions of the whole “pie.” It’s great for visualizing percentages and proving that it *was* Greg who drank most of the coffee from the pot, as you’ve been saying all along.

### 4. Histogram

A histogram shows a frequency distribution of your data. It can tell you how many people drink a specific number of coffee cups per day, but it won’t explain why Greg gets on your nerves so much.

### 5. Scatter plot (interactive)

A scatter plot shows clusters of values for two different variables. This lets you e.g. spot trends in coffee consumption vs. productivity and identify any outliers. (I won’t say who they are, but we all know.)

### 6. Box plot (box-and-whisker plot)

A box plot visually summarizes data using five data points: minimum, first quartile, median, third quartile, and maximum. This helps you figure out the length of the average coffee break and how it’s *of course* Greg who takes the longest breaks every single time! God, Greg, you’re the worst.

### 7. Heat Map

A heat map uses color gradients to represent the relative “strength” of data values. This can give you a handy way to see where most coffee spills occur in the office. Hot spots are where Greg’s been most active. Coincidence?

### 8. Area chart

An area chart happens when you take a line chart and color-fill the area below it. You can stack several line charts on top of each other to show combined changes over time. Useful for tracking seasonal changes in total coffee consumption by different people in the office.

### 9. Radar chart (interactive)

A radar chart tracks data along several axes, starting from a central point. This results in a sort of spiderweb (which is why it’s also called a “spider chart”). It’s used to compare one or more things across multiple dimensions, like different people’s coffee preferences in terms of strength, sweetness, or creaminess.

### 10. Treemap

A treemap groups hierarchical data into nested rectangles representing different categories and sub-categories. This lets you see how much is spent on coffee (lattes vs. cappuccinos) and pastry (scones vs. donuts). It explains exactly why Greg’s always broke.

### 11. Bubble chart (interactive)

A bubble chart is a fancier scatter plot, which can track a third dimension by turning data points into differently-sized bubbles. In our case, in addition to seeing everyone’s coffee consumption vs. productivity, we can use bubbles to represent how much every person hates Greg. Your bubble will be giant.

### 12. Waterfall chart

A waterfall chart breaks a category total into its components, resulting in cascading “waterfall” steps. There’s no better way to visualize Greg’s monthly budget and just how expensive his coffee addiction really is.

## Beyond the standard charts

The 12 charts above are what OpenAI lists as the default, but they’re far from an exhaustive list of what ChatGPT can work with.

It turns out, ChatGPT is perfectly capable of mimicking virtually any chart type you throw at it. Matplotlib is an excellent plotting library and is directly transferable to ChatGPT.

For instance, say I like this crazy-looking “Curve with error band”:

Matplotlib provides the code necessary to replicate it:

I can grab that code, paste it into ChatGPT, and ask it to reproduce it:

Or how about this “Stem Plot”?

Here’s the ChatGPT version:

What’s cool is that once you’ve given ChatGPT an example of what you’re after, you can ask it to manipulate the underlying data while reusing the chart type:

So if your analysis requires more than the basics, chances are ChatGPT can handle it, as long as you can feed it examples.

## Over to you…

Have you used ChatGPT’s chart-plotting ability for any real-world analysis? What charts do you find the most useful to work with?

Do you have any examples of charts or graphs that ChatGPT couldn’t reproduce?

Leave a comment or shoot me an email at whytryai@gmail.com.

Wait! Those are real? I’ll be damned!

In writing this article, I discovered that the radar and bubble charts are also interactive. (But the layout in the interactive version doesn’t always match the final product.)

Correlation doesn’t equal causation, but come on!

...uncharted territory 🤣

Never mind Greg. Let's talk about those cats in boxes. Are they all alive, and does the answer to this affect how long they spend in the boxes? Asking for a friend named Erwin.

Also: I'm getting closer to having the ability to sort my own data meaningfully for the first time, and charts like this could be a really good way to study it. I'm not in a huge hurry, but I definitely want to do this some time in the near-ish future.