Which Chart Should You Use? A Field Guide for AI Data Viz

TLDR: To decide which chart to use, start with the question your data answers, not the shape you like. Comparing categories is a bar chart. Change over time is a line chart. A relationship between two numbers is a scatter plot. Parts of a whole with a few big slices can be a pie, but a bar is usually clearer. A single headline number is a stat tile, not a chart at all. Pick the job first and the chart type names itself. This guide walks through each job with an example, the AI prompt to get it right, and the traps that make good data unreadable.

The question “which chart should you use” almost never has a hard answer, and yet most charts still get it wrong. Not because the design is ugly, but because the shape was chosen before anyone decided what the chart should say. Someone reaches for a pie because it looks approachable, or a dual-axis line because it looks smart, and then bends the data to fit. The result is a chart that is accurate and unreadable at the same time.

That mistake used to cost you an hour in a charting tool. Now AI draws the chart in seconds, which sounds like it should help, and mostly it does. But the model will render exactly what you ask for, including the wrong chart, beautifully. So the one skill that still matters is knowing which chart to ask for. This is a field guide for exactly that.

Start with the question, not the shape

Every chart answers one question. Figure out the question and the chart type stops being a matter of taste. Here is the whole decision in one move: name what you want the viewer to see, then read across.

Notice that two of the seven rows are not the shape most people would have grabbed. Parts of a whole usually reads better as a bar than a pie, and a single number is not a chart at all. Those two swaps alone fix most of the charts that end up in reports nobody understands.

Here is the same idea as a quick flow you can run in your head before you prompt for anything.

Comparing categories: use a bar chart

When the reader’s question is “which one is biggest,” bars win, because the human eye compares lengths against a shared baseline far more accurately than it compares angles, areas, or colors. Keep the baseline at zero, sort the bars by size unless the categories have a natural order, and if one bar is the point, give it your accent color and let the rest recede.

Horizontal bars, like these, are the safest default when your category names are words rather than dates, because the labels sit flat and stay readable no matter how long they get. Vertical bars are fine for short labels or time buckets.

Change over time: use a line chart

The moment your horizontal axis is time and you want to show a trend, use a line. The slope does the work: the reader sees the direction, the speed, and the wobble in one glance. Bars can show time too, but they are better when you have a handful of discrete periods to compare than when you want a continuous story.

That label is not decoration. A line chart with an annotation on the one moment that matters tells a story. The same line with no marker is just a shape going up. More on that below.

Parts of a whole: usually a bar, sometimes a pie

This is the row people get wrong most often. A pie chart works for exactly one situation: two or three slices of one whole, with sizes different enough that you can tell them apart at a glance. Add a fourth slice, or make two slices close in size, and the eye gives up, because comparing wedge angles is genuinely hard. The tell is simple: if you have to print the percentages on the slices so people can read the chart, the chart failed and a bar would have worked.

Same numbers, both times. The bar chart on the right answers “which is bigger” instantly. The pie on the left makes you squint. When in doubt, bar it.

Relationships and distributions: scatter and histogram

Two less common jobs are worth knowing because nothing else does them well.

Use a scatter plot when you want to show whether two numbers move together: ad spend versus signups, price versus rating, hours slept versus focus score. Each dot is one observation, and the shape of the cloud tells you whether there is a relationship and how tight it is. No other chart shows correlation as directly.

Use a histogram when you have many values of one number and want to see the shape of them: how long support calls last, how big orders are, what ages your users are. It is not the same as a bar chart. A bar chart compares named categories. A histogram buckets a continuous range and shows you what is typical and what is rare.

When the answer is “not a chart”

The most over-charted thing in the world is a single important number. Revenue this month. Active users. Conversion rate. You do not need a one-bar bar chart for that. You need a stat tile: the number, big and bold, with a label and maybe a small change indicator. It reads in a fraction of a second, which is exactly what a headline number is for.

Save your charts for the numbers that need context to make sense. Give the headline figures the stat-tile treatment and your dashboard gets calmer and clearer at the same time.

The charts to avoid

A short list of shapes that look tempting and almost always backfire:

• Dual-axis charts. Two different y-scales on one chart, revenue on the left, conversion rate on the right. The crossing point looks meaningful but is an artifact of where you set the two scales. Use two charts instead, or index both to a common base.
• 3D anything. The perspective distorts the very lengths and areas the chart exists to compare. There is no data reason to tilt a bar chart.
• Pie charts with many slices. Covered above. Four or more, or slices of similar size, means bar.
• Truncated axes on bar charts. Starting the y-axis at 90 instead of 0 turns a 2% difference into a cliff. It is the fastest way to lie with a true number. Start bars at zero and say so if you ever have a real reason not to.

How to ask AI for the right chart

Once you know which chart you want, the prompt is short and specific. Name the type, hand over the data, pin the axis behavior, and say the one thing the chart should make obvious.

Create a self-contained HTML page with a horizontal bar chart of this data,
sorted largest to smallest, y-axis starting at zero, using a single green
accent with the largest bar emphasized and the rest muted. Label each bar
with its value. Make it responsive. Here is the data:
[paste your rows]

Swap the first line for the chart you actually need:

...with a line chart, time on the x-axis, y-axis starting at zero, and a
labeled marker on the point where the trend changes...

...with a scatter plot of ad spend (x) versus signups (y), one dot per
campaign, and a light trend line...

The pattern is always the same: chart type, data, axis rule, and the takeaway. That last part matters most, because it turns a generic chart into one that says something. For more on writing prompts that produce clean, publishable output, see AI prompts that produce a page worth sharing, and for the mechanics of SVG versus Chart.js versus D3, see how to create charts and graphs with AI.

Make the chart say something, not just show something

Picking the right chart is half the job. The other half is aiming it. Three habits do most of the work:

• Write the title as the takeaway. “Revenue by channel” is a label. “Search drives half of revenue” is a headline. If you know the point, put it in the title.
• Highlight the one thing. Give the important bar or point your accent color and let everything else go quiet. A chart where everything is emphasized emphasizes nothing.
• Annotate the moment. A single marker on “launch week” or “price change” turns a trend line into a story, the way the signups chart above did.

These three moves are what separate a chart that decorates a report from one that makes a point. When you assemble several of them, the next question is layout, and how to make a dashboard with AI covers arranging charts and stat tiles into something scannable, while how to make reports and dashboards with AI covers the wider document.

Get the chart in front of people, working

Once you have the right chart, it is a self-contained HTML page: the hover tooltips work, it resizes on any screen, and the data reads clearly. The problem is that it is sitting in your chat window or on your laptop, and the people who need it are somewhere else. A screenshot flattens it and kills the tooltips.

That last step is what VisiblePage handles. Paste the HTML the AI produced, or drop a ZIP for a multi-file dashboard, and you get a live URL immediately. The interactivity survives because you are sharing the real page, not a picture of it, and it serves fast from the edge so it opens quickly for whoever you send it to. Set the page public, private, or password-protected for a client, put it on a custom domain, and when the numbers change, re-publish to the same link so everyone always sees the current version instead of hunting for the latest attachment. To be clear about the split: the AI draws the chart, and VisiblePage publishes that finished HTML as a link. If the chart lives inside a larger document, turning a spreadsheet into a report with AI shows the full path from raw data to a shared page.

Try it now: Publish your AI chart as a live link with VisiblePage and send one interactive URL that stays current.

Frequently asked questions

Which chart should I use for my data?

Start with the question your data answers, not the shape. Comparing categories calls for a bar chart. Change over time calls for a line chart. A relationship between two numbers calls for a scatter plot. Parts of a whole with a few slices can use a pie, but a bar chart is usually clearer. A single headline number is not a chart at all, it is a big-number stat tile. Decide the job first and the chart type names itself.

When should you use a pie chart?

Only when you are showing parts of one whole and there are just two or three slices with obviously different sizes. The moment you have four or more slices, or the slices are close in size, the human eye cannot compare angles reliably and a horizontal bar chart reads better. If you catch yourself adding percentage labels so people can tell the slices apart, that is the chart telling you it should have been a bar.

What chart shows change over time?

A line chart. Time goes on the horizontal axis, your value on the vertical axis, and the line makes the trend, the direction, the speed, the dips, immediately visible. Use a bar chart for time only when you have a few discrete periods and want to compare their totals rather than show a continuous trend.

How do I tell AI which chart to make?

Name the exact chart type, the data, the axis behavior, and the one thing you want it to show. For example: make a self-contained HTML line chart of this monthly revenue, y-axis starting at zero, and highlight the point where growth accelerates. Being explicit about the type and the takeaway is what separates a chart you can publish from one you have to redo.

What is the most common charting mistake?

Two tie for first. Picking the chart shape before deciding what it should say, which leads to pie charts with ten slices and dual-axis charts nobody can read. And truncating the axis so a small change looks dramatic. Decide the message first, keep bar-chart baselines at zero, and most bad charts never happen.

The short version

The answer to “which chart should you use” is never about taste. It is about the question your data answers. Comparisons are bars, trends are lines, relationships are scatters, one number is a stat tile, and parts of a whole are usually bars too. Decide the job, name the chart to your AI along with the takeaway you want it to show, check the axis tells the truth, then publish the finished HTML as a live link so people can actually read it.

Ready? Turn your AI chart into a live link with VisiblePage and your first page is free.