Grouped bar chart maker

Every bar starts at zero

That single sentence is the entire argument for a grouped bar chart over a stacked one.

In a stacked bar, only the bottom band sits on the baseline; everything above floats. In a grouped bar — sometimes called a clustered bar — the bars sit side by side, each rising from zero. So every value in the chart can be compared with every other value, accurately, in any direction.

What you give up is the total. Nothing in a grouped bar shows you that Q4 was $6.8 million altogether; the reader would have to add three bars by eye. That is the trade, and it is the whole decision:

• The parts are the story → grouped
• The total is part of the story → stacked
• Both matter equally → two charts

The decision people skip: which variable groups

A grouped bar has two categorical variables — say region and year — and you must choose which one forms the groups and which forms the bars within each group. Most tools decide for you based on column order. The choice changes what the chart is about.

Group by region, bars are years: four clusters labelled North, South, East, West, each containing an FY25 bar and an FY26 bar. The easy comparison is within each cluster. This chart says: how did each region change?

Group by year, bars are regions: two clusters labelled FY25 and FY26, each containing four regional bars. The easy comparison is now within each year. This chart says: how did the regions rank against each other, and did the ranking shift?

Same numbers, two different arguments. Adjacent bars are always the easiest comparison, so put the things you want compared next to each other.

Say group by year or group by region to control it.

The ceiling: three series, seven categories

A grouped bar degrades faster than most charts as you add data, because total bars multiply: series × categories.

• Two series across six categories — twelve bars. Comfortable.
• Three series across seven categories — twenty-one bars. The practical ceiling.
• Four series across eight categories — thirty-two bars. Unreadable; the clusters merge into a wall.

Past the ceiling, the options are: drop to the two series that matter, switch to small multiples (one small chart per series, shared axes), or use a line chart if one of your variables is time.

Spacing, and why it matters more here

A grouped bar chart depends on the reader instantly seeing which bars belong together, and that is carried entirely by gap hierarchy:

• No gap, or a very small one, between bars inside a group — they read as one unit
• A clear gap between groups, wider than the bars themselves
• Consistent colour per series across every group — series identity is carried by colour, and it must never shift

Get the gaps wrong and a reader comparing the wrong pairs will draw the wrong conclusion without ever realising it. ChartGPT sets the hierarchy by default.

When to use a grouped bar

It works well for:

• This period against last across a handful of categories
• Actual against target or budget, side by side
• Two or three products compared across regions or segments
• Before and after, where each category has a pair
• Survey results split by two groups — responses by age band, say

Reach for something else when:

Your situation	Better chart	Why
The total matters	Stacked bar	Grouped bars hide totals
More than three series	Small multiples	Clusters become a wall of bars
One of the variables is time with many periods	Line chart	Slope reads better than clusters
One series across many categories	Bar chart	No need for grouping
Long category names	Horizontal bars	Cluster labels rotate and become unreadable
Parts of a single whole	Pie or donut	Grouping does not signal a whole

Two worked examples

1. Year on year by region

Data:

Region	FY25	FY26
North	42	58
South	51	54
East	47	71
West	68	66

Prompt: Grouped bar comparing FY25 and FY26 revenue by region, in $ thousand, group by region. The point is that East grew fastest.

What you get: four clusters, two bars each, same two colours throughout, all rising from zero. Because the grouping is by region, each pair sits together and the year-on-year change in each region is the easy read — which is what the title claims.

Note that West fell. In a stacked version that decline would be much harder to see.

2. Actual against target

Data:

Team	Target	Actual
Sales	100	118
Marketing	100	87
Support	100	104
Success	100	92

Prompt: Grouped bar of actual against target by team

What you get: four clusters with a target bar and an actual bar in each, the same two colours throughout. Over- and under-performance is visible as the gap between the pair in each cluster.

Four things that ruin a grouped bar

1. Too many bars. Covered above and the most common failure by far. Twenty-one bars is the limit.

2. Colour that shifts between groups. If FY25 is blue in the first cluster and green in the second, the chart is unreadable. Series colour must be fixed, and ChartGPT fixes it.

3. Equal gaps everywhere. With no gap hierarchy the eye cannot tell clusters apart, and the chart reads as one long undifferentiated row.

4. Grouping the wrong way round. Not a rendering fault but a framing one — putting the comparison you care about in non-adjacent bars makes the reader do work they should not have to.

Getting your data in

A grouped bar needs a wide table: the first column holds your categories, and each following column is one series.

Region	FY25	FY26
North	42	58

Column headers become series names in the legend, so keep them. Two columns of values gives you two bars per group, three gives three.

Blank cells are treated as zero. If a series genuinely did not exist for a category, zero is right; if the number is merely unknown, say so, because a zero-height bar looks like a real result.

The paste limit is 500 rows, on every plan — though a grouped bar becomes unreadable long before that.

How to make one

1. Paste a wide table — categories first, then one column per series.
2. Say grouped bar chart so it does not infer the type.
3. Say which variable groups — group by region or group by year.
4. Name your units so the axis and subtitle read correctly.
5. Adjust in words. Sort by FY26 descending. Swap the grouping. Keep calendar order.
6. Export — SVG for slides, PNG for a quick paste, PDF for print.

Related chart types

• Stacked bar — the same data stacked. Totals visible, individual parts harder to compare.
• Bar chart — one series across categories, and free on every plan.
• Line chart — when one of your variables is time with more than a few periods.

Exporting

SVG, PNG and PDF on every plan. SVG is worth using here — grouped charts have many thin shapes and narrow gaps, which is exactly what raster scaling smears.