A chart has one job: let somebody see a relationship faster than they could read it. Most charts fail at that, and almost always for the same three reasons. They picked the wrong shape, they decorated instead of explaining, or they never said what the point was.
The good news is that this is one of the most rule-governed areas of design. There are actual right answers, which after a week of arguing about typography is deeply refreshing.
Pick the shape from the question
Not from what looks nice. From what the reader is trying to find out.
- Comparing amounts between categories? Bar chart. Horizontal if the labels are long, which they usually are.
- Showing change over time? Line chart. Time goes left to right, always.
- Showing parts of a whole? A stacked bar, or just a table. Not a pie, unless there are two or three slices and the differences are obvious.
- Looking for a relationship between two measures? Scatter plot. This is the most underused chart there is.
- Precise values that people will look up? A table. Tables are not a failure of imagination, they are the right answer surprisingly often.
The pie chart problem
People are bad at comparing angles and areas, and good at comparing lengths. That is not an opinion, it has been measured repeatedly. A pie with seven slices is a puzzle; the same data as a sorted bar chart is instantly readable.
Pies survive for two or three segments where one is obviously dominant. Beyond that, and especially in three dimensions where perspective distorts the very angles you are meant to compare, they actively mislead.
The bar chart rule you cannot break
Bar charts must start at zero. The length of a bar is the value, so a truncated axis makes a five per cent difference look like a doubling.
Line charts are different, because there the message is the shape of the change rather than the absolute size, so a zoomed axis is legitimate. Just label it clearly enough that nobody is fooled.
If a difference is genuinely too small to see on a zero-based bar chart, that is information. The honest response is to say the difference is small, not to rescale until it looks big.
Take things away until it hurts
Every default chart arrives with a pile of things that carry no information: heavy gridlines, a border, a background fill, a legend for a single series, tick marks, a drop shadow, a gradient in the bars.
Remove all of it and see what breaks. Usually nothing does.
What survives: the data, the axis labels, one light gridline set if values need estimating, and a title. What you gain is that the data is now the loudest thing on the chart, which sounds obvious and is the exception rather than the rule.
Two specifics worth naming. Label directly instead of using a legend where you can. Putting the series name at the end of its line beats making the reader look back and forth to a color key. And sort your bars unless the order carries meaning like time or a natural sequence. Alphabetical order is almost never what anyone wants to know.
Color carries meaning, so spend it carefully
Use one color for everything, and a second color only for the thing you want people to look at. A chart where every bar is a different color is a chart where color means nothing.
Then a few practicalities. Around one in twelve men has some degree of color blindness, so red against green as your only distinction excludes a real slice of your audience. Vary lightness as well as hue, and the chart works in grayscale too, which is a good test for whether it works at all.
Keep to a maximum of about five or six distinguishable series. Beyond that, no palette helps and you need a different chart, usually several small ones side by side.
Say what the point is
This is the change that improves charts the most and it is not a design change at all.
Most chart titles describe the contents: “Monthly revenue 2026”. That tells the reader what they are looking at, not what they are supposed to see. Compare with: “Revenue doubled after the March relaunch”. Now the chart is evidence for a statement rather than a puzzle to solve.
If you cannot write that sentence, you may not need the chart.
And while you are there: that sentence is also the alt text. A chart with alt text saying “bar chart of monthly revenue” is useless to somebody who cannot see it. The finding is what they need, plus a link to the numbers if precision matters.
Make it look like it belongs
The reason so many charts look bolted on is that they come out of a different tool with a different personality: different typeface, different grays, different corner radius, a color palette from 2009.
When a chart arrives as editable layers, that stops being a problem. Set the type to your typeface, the accent to your accent color, the gridlines to your hairline gray, and it reads as part of the document rather than a screenshot of a spreadsheet.
Small thing, big effect on whether a report feels authoritative.
The test
Show it to somebody for five seconds, take it away, and ask what it said.
If they can tell you the finding, the chart works. If they describe the layout, or ask a clarifying question, it does not, and no amount of styling will fix that. Go back to the shape of the chart and the sentence in the title, because that is where the failure always is.
