A comparison message says one category is bigger than another. Length on a common baseline carries that better than anything else the eye does, so the bar family wins here and keeps winning.
The only real decision is orientation, and your labels make it for you. Columns run vertically and their labels sit underneath in a space as wide as the column. Bars run horizontally and their labels sit to the left in a space as wide as you choose to make it. The rule that follows is arithmetic. Count the characters in your longest category name, work out the horizontal space one column gets, and if the name doesn't fit, use bars.
Worked example: six channel labels on a widescreen slide
Let's work one. A 16:9 widescreen slide is 13.333 inches across in the current desktop defaults of the mainstream presentation tools, as checked in September 2026. Verify that against your own installation, since the default follows the template and the release. Half-inch margins leave 12.333 inches of plot. Six columns divide that into 2.0555 inches each. At roughly nine characters to the inch in 14-point type, each column label has room for about 18.5 characters. Now take six real channel names, the longest of which is "Business correspondent network" at 30 characters. That label needs 1.62 times the space available. The software won't tell you this. It'll rotate the labels 45 degrees, or shrink them to 8-point, or truncate them with an ellipsis. Each of those hides a decision you should have made on purpose.
Rotated labels are worse than they look. Your reader tilts their head, or their eye travels diagonally, and the reading rate drops sharply on a chart that was meant to take three seconds. Bars solve it completely. The label runs horizontally in the left margin, at full size, at whatever length it needs, and the chart gets easier to read at no cost.
Advertisement
What about more than twelve categories, and sorting?
Beyond about twelve categories, both forms start to fail for a different reason. Twenty bars of similar length stop being a comparison and turn into a texture, and the reader stops resolving individual values. The dot plot repairs this. One dot per category on a common horizontal scale, categories sorted, a thin leader line if the eye needs help travelling. It uses a fraction of the ink and reads accurately at forty categories. It also handles negative values without any special case. Bars don't.
Sorting isn't optional, and it isn't formatting. The sorting section makes the full argument. The short version is that the order of categories is a claim about what matters, and the default order your software chose is a claim you did not make.