Data visualization generator

A chart is an argument
about what matters.

So the generator makes it like one: the encoding chosen for how accurately people read it, the axis honest, the labels sized for a phone, and the finding written across the top instead of the topic.

Visualise your data

Accurate before pretty

Zero baselines on bars, consistent intervals on time axes, and no perspective on anything. A chart that flatters the number is not a design choice, it is an error with a nice palette.

Readable at the size it will be seen

Most charts are read on a phone. Labels are sized against the frame, not against the desktop preview, and a chart that would need eight tiny labels is summarised to five legible ones instead.

Legible to every reader

Series are separated by more than hue, text clears 4.5:1 against its background, and every figure is real selectable text — so a screen reader gets the numbers rather than the alt attribute’s best guess.

The one ranking worth memorising

How accurately people read each kind of chart

Visual encodings are not interchangeable. Ordered by how precisely a reader can judge a quantity from each — the higher up the list, the more of the comparison the chart does for them rather than leaving it to their guesswork.

  1. 1

    Position on a common scale

    Bars from a shared baseline, dots on one axis

    The most accurate judgement people make. If a comparison matters, put it here.

  2. 2

    Position on identical, unaligned scales

    Small multiples — the same chart repeated per category

    Nearly as good, and it scales to many series without a legend.

  3. 3

    Length

    Stacked segments, floating bars

    Good, but only for the segment that starts at the baseline. The ones above it are hard.

  4. 4

    Angle and slope

    Pie and donut slices, line steepness

    Fine for “about a half”, poor for “37% versus 41%”. Label the slices.

  5. 5

    Area

    Bubbles, treemaps, area charts

    People systematically underestimate area. Doubling a value looks like a 40% rise.

  6. 6

    Colour intensity and saturation

    Heatmaps, choropleths

    Good for pattern, bad for value. Always pair it with the number written out.

This ordering comes from graphical-perception research into how people judge quantities, and it is why bars beat pies for close comparisons and why bubble charts flatter big values. It is also why the generator reaches for position first.

Six questions to ask the chart before anyone else sees it

01Does the title say the finding, not the topic?
“Revenue grew fastest in the North” beats “Revenue by region”. The title is the only part everyone reads.
02Is the unit on the sheet?
Currency, percentage, per-thousand, and the period. A number without a unit is a decoration.
03Can a reader tell what it is not saying?
A chart of five regions in a company with nine invites a wrong conclusion. Say which are shown.
04Does colour mean one thing?
One hue, one meaning, across the whole sheet — and ideally across the whole report.
05Would it survive being screenshotted?
If the caption carries essential context, that context is lost the first time somebody crops it. Put it on the sheet.
06Is the source named?
Where the numbers came from, and when they were pulled. This is what makes a visualisation citable rather than merely shareable.

Scope and limits

What data can I bring?

A spreadsheet or CSV, a table pasted as text, figures embedded in a report or PDF, or a page of prose with the numbers written into it. There is no required schema.

Does it connect to my database or BI tool?

No. It works on data you bring to it — a file, a paste or a document. That keeps the scope honest: this is a tool for making an explanatory visual, not a live dashboard.

Explanatory or exploratory?

Explanatory. It is built for the point at which you already know what the data says and need somebody else to see it. For exploring a dataset for the first time, use an analysis tool and come back when you have a finding.

How does it handle colour-blind readers?

Series are distinguished by order, label and position as well as hue, text is held to a contrast ratio rather than a swatch, and the default palettes are chosen so adjacent series differ in lightness as well as in colour.

Can I export something editable?

Yes — SVG keeps every shape and every string editable in a vector tool, and stays sharp at any size, including a printed poster.

Is the output accessible?

Text in a piece is real text, so it can be selected, searched, translated and read aloud. Charts carry accessible labels, and a share link is a real page rather than an image.

You already found the story

This is the part where somebody else has to see it in three seconds.

Make a visualisation