Charts that argue, and charts that just sit there

A chart on a slide has one job: make a single claim visible in about three seconds. That means the title states the finding rather than naming the axes, one series carries colour while everything else goes grey, and values sit next to the marks instead of in a legend. Most slide charts fail this because they were built as evidence with no claim attached — the room is handed "Revenue by quarter" and left to work out the point on its own, and by the time anyone has, the speaker is two slides further on.

A chart in a document and a chart on a slide are different objects with the same name. A document chart is explored: the reader controls the clock and can go back. A slide chart is glanced at while somebody talks over it. Everything below follows from that difference.

What should a chart title on a slide actually say?

It should state the finding. "Revenue grew 40% after we changed pricing" is a title. "Revenue by quarter" is a filing label — it names the drawer the chart came out of and says nothing about why it is on screen. The test is whether the title contains a verb doing work. Grew, fell, overtook, stalled, doubled. If the only verb available is an implied "is", you have named the data rather than reported it.

The objection is that a stated finding is editorialising, and a neutral title lets the audience judge for themselves. That gets the sequence backwards. They will reach a conclusion either way; a neutral title only means they reach it slowly, privately, and possibly not the one you were about to argue.

In the editor this is enforced by the shape of the slide rather than by advice. A slide title is a slot like any other, so it carries the character budget the layout was drawn for: chart-hero gives the title 40 characters and the note beneath it 220. Forty is not much, and that is the useful part. "Revenue grew 40% after we changed pricing" is 41 characters and will not fit. "Pricing change lifted revenue 40%" is 33 characters of pure claim. The caveats go in the note.

Why does one coloured bar beat a chart in full colour?

Because colour is a pointer, not decoration, and a chart with six colours in it points at six things, which is the same as pointing at nothing. Grey out every bar except the one your title is about and the reader's eye lands on the evidence before they finish the sentence above it. That is most of your three seconds recovered for the price of one decision.

The rainbow habit comes from spreadsheet defaults, which assign a colour per series because they have no idea which series matters. You do. The editor's charts draw themselves in the deck's brand palette automatically — there is no per-series colour picker to reach for. What there is instead is a sparse override: set the colour of one slot and every other slot stays on the palette. One entry in an array, and the rest of the chart goes quiet.

The related mistake is colouring by a category that carries no meaning. Q1 blue, Q2 orange, Q3 green tells the reader quarters are things of different kinds. They are the same thing at four times.

Are legends really that bad, or is that a design opinion?

A legend is a lookup table, and the reader pays for every lookup. Read the swatch, carry the colour across to the plot, find a mark that matches, hold the name in memory, go back for the next. Two series is four lookups if they glance twice, and on a slide held for ninety seconds that cost is charged again and again.

Direct labels remove the lookup. Put the series name at the end of its own line, in that line's colour, and the chart describes itself. The editor's presets do this where the geometry allows: the horizontal-bar preset ships with the legend off and values on, because the categories are already written down the axis and a legend would name things that are named.

A legend is defensible in two situations: the marks are too small to label, or the series interleave so tightly a label would land on another line. Both are signals to reconsider the chart, not the legend.

Is cutting the y-axis at something other than zero dishonest?

It depends on the mark, and the blanket "always start at zero" is wrong often enough to be worth replacing with the real rule. In a bar chart, truncation is a lie. The bar encodes its value as a length measured from zero — that is the whole visual contract — so a bar drawn from 90 instead of 0 makes 94 look roughly twice 92. The lengths no longer hold the ratio the numbers hold, and no axis label rescues it, because nobody reads the axis before they read the bars.

In a line chart the contract is different. A line encodes change as slope, and the reader compares points to each other rather than to the origin. Zero is often not on the scale of the question at all: a satisfaction score moving 4.1 to 4.4, a p95 latency going 180ms to 240ms, retention sliding 91% to 88%. Force those onto a zero baseline and you draw a flat line that tells the room nothing happened — its own kind of lie, and in practice the more common one.

So the rule is about the question, not the axis:

• If the question is "how big is this compared to that", the baseline must be zero. Bars, columns, stacked anything, waterfall totals.
• If the question is "which way is this moving, and how fast", a truncated axis is fine and often necessary. Lines, and areas only where the fill is decorative.
• Either way, put the size of the move in the title. A reader told the number cannot be fooled by the steepness.

How many segments can a donut hold before it stops working?

Three. Angle and arc length are the least accurately read of the common encodings — a reader ranking a 22% slice against an 18% slice from the arcs alone is guessing. Past three the arcs thin out, the labels stop fitting inside them, you add leader lines, the leader lines cross, and eventually you add a legend: the tax from two sections ago, paid on top of an encoding that was already weak.

The editor's donut preset ships with four categories and advises keeping to five. Five is the point at which it still renders cleanly. Three is the point at which it still argues. Those are different ceilings, and it is worth knowing which one you are near.

A donut earns its place in one case: when the claim is one segment against everything else. "Enterprise is 46% of revenue" is two arcs and a number, and the shape carries it instantly. The moment the claim becomes a ranking — biggest, second, third — switch to horizontal bars sorted by value: ranked categories, long labels down the left, values printed on the bars, no legend.

What do you do with a chart that supports no claim?

Delete it. This is the hardest instruction here, because the chart exists — somebody spent an afternoon on the query and the numbers are correct, so removing it feels like removing work. But a chart supporting no claim spends the audience's attention and returns nothing, and it makes the charts on either side of it seem less urgent by association.

If you cannot write the title as a sentence with a verb in it, you do not have a chart. You have a table with ambition.

Apply that test first. If the sentence comes out, you also know which bar to colour and which axis rule applies. If it does not, there are two honest outcomes: find the claim actually in the data, which sometimes means a completely different cut of it, or cut the slide and put the number in a sentence somewhere else.

The footnote test catches the near-misses. A source line is fine; the presets ship a placeholder source so the field is visibly unfilled rather than quietly absent. The footnote that condemns a chart is the one explaining the chart to itself. "Excludes Q2 owing to a reporting change." "Left axis is revenue, right axis is margin." Each admits that the picture, read as drawn, says something other than the truth. Dual axes are the worst offender: the author picks both scales, so the author picks where the lines cross.

Charts are edited in place here, with no spreadsheet step, and that changes the economics. When a fix costs thirty seconds rather than twenty minutes, "it took ages to make" stops being an argument for keeping it. The only expensive part left is deciding what you are claiming.