Most bad charts are the right data in the wrong shape.
Before any tool opens, write one sentence that finishes “I want the reader to see that...”. “Our Fremantle site overtook Subiaco in June.” “Three of the five product lines lose money.” “Enquiries pile up on Monday mornings.” Once the sentence exists, it points at the chart.
Say you run three cafes and have a table of monthly takings. The question “which site is biggest” wants bars. “Is Fremantle catching up” wants lines across months. “Where does a dollar of takings come from” wants a pie, but only if you have a handful of sources.
That mapping is worth memorising, because an AI chart maker will draw whatever you hand it. Ask for a pie of twelve products and you get twelve slices of nearly identical size, with a legend you have to cross-reference, and the tool will not warn you.
The UK Government Analysis Function’s chart guidance gives a useful set of limits. It says pie charts should only be used with five categories or fewer, and that when differences between categories are small, a bar chart is better than a pie. It also advises aiming for a maximum of four lines on a line chart.
Here is how the editor’s 11 chart types line up with common questions.
| Your question | Chart in the editor | Watch out for |
|---|---|---|
| Which is biggest, which is smallest | Bar (column) or Bar (horizontal) | Horizontal bars suit long category names |
| How did it change over time | Line, or Area | Keep the series count low so the lines stay readable |
| What share does each part hold | Pie or Doughnut | Five slices or fewer, with clearly different sizes |
| How do two measures relate | Scatter | A pasted CSV gives labels plus number columns, so check the preview shows the points you expect |
| What does one item look like across several measures | Radar, Polar | Hard to read exactly, so use sparingly |
| Which cells of a grid are high or low | Heatmap | Needs rows and columns that mean something |
| How does a total break into many parts | Treemap | Good when a pie would need a dozen slices |
A bar chart also has a rule line charts don’t. The same guidance says the numerical axis should never be broken on a bar chart, because bar length is the message. It accepts a broken y-axis on a line chart when necessary. The editor labels axes but has no control for setting an axis minimum, so look at the preview before you export and make sure the bar lengths are not misleading.
There are two routes. You can ask in chat, for example “chart these monthly takings for the three cafes as lines”, and the worker opens the Visualisation editor with the chart drawn. Or you can open the editor and use Data, which opens a box for pasting JSON or CSV.
The CSV reader is simple and worth knowing. The first row is the header. The first column becomes the labels along the bottom, and every other column becomes a series. For the cafe example that is a month column followed by one column per site. The box shows a preview line as you paste, with the count of labels, series and points, so you can see at once if a column went missing.
Three habits keep the paste clean. Leave thousands separators out, so write 12400 and not 12,400, because the reader splits on commas. Keep every value in the number columns numeric, because anything it can’t read as a number is plotted as zero rather than flagged. And trim a long table before pasting, since a chart of 300 rows is a different job from a chart of 12.
JSON works too, either as labels plus datasets, as a list of label and value pairs, or as a simple object of names and numbers. A small object like that is the quickest way to chart a single series by hand.
A chart gets copied into a report, a Slack message or a slide, and it travels without your commentary. So it needs its own words. Open Axes in the toolbar and set three things, the chart title, the x-axis label and the y-axis label. Put units in the y label (“Takings, A$ thousands”), not only in a caption.
The Analysis Function calls it best practice to give a chart two titles, a headline title and a formal statistical subtitle. In the editor the chart title takes the headline. The Axes box has no subtitle field, so put the description of the data (what it is, where it is from, for when) in the page or slide text beside the exported image.
The same guidance advises avoiding legends on line charts unless necessary. The editor has a legend toggle, on or off, and no direct-label control. If you turn the legend off, make sure the series are named in your title or your surrounding text, or the reader has nothing to decode the colours with.
The editor starts with a default set of distinct hues. If you change the primary colour through Colours, the other series colours are derived from it by stepping around the colour wheel, and the editor offers no per-series colour picker, so set your brand colour and then look at what the other series became.
W3C’s guidance on use of colour (WCAG 2.2, Understanding SC 1.4.1, updated September 2025) says colour should not be the only visual means of conveying information, so people with colour vision deficiencies can still get it. For a chart that means being able to tell series apart some other way as well. Two practical moves are fewer series, and putting the series name in the title or the text beside the image.
Check your finished chart by converting a screenshot to greyscale, or by squinting. If two slices or lines merge, drop one or merge categories. General information, not legal advice. Whether any accessibility standard applies to your publication depends on who you are and who you publish for.
The toolbar has undo and redo, a chart type switcher, Data, Colours, Legend and Axes. Switching type is the cheapest experiment in the whole process. Draw the same numbers as bars, then lines, then a heatmap, and keep whichever one makes the sentence from the start of this post obvious.
Regenerate is the AI button. It sends a request to the chat composer to regenerate the chart, so use it when you want a fresh take on the shape rather than a tweak. For anything small, such as a typo in a label, an axis title or a series that should be a different colour, the hand controls are faster.
Export offers PNG, SVG, PDF and JSON. PNG suits a slide or an email. SVG stays sharp at any size, which is better for print and for a web page. PDF puts the chart on a page with its title. JSON saves the chart’s specification, with its type, data, title, axis labels and colour, so you can keep it or hand it to someone to rebuild. Open in Code turns the chart into a small standalone web page you can edit in the Code editor.
An exported image has no text alternative of its own. Wherever you place it, add alt text that states the finding, such as the Fremantle sentence from the top of this post, and put the same sentence in the body copy.
Copy the columns out as CSV text and paste them into Data. The first column is read as labels, so put your categories or dates there.
Not as separate chart types in the editor today. The list is the 11 types in the table above.
Yes, you paste the numbers, set the labels and export an image. A chart that must update itself every week is covered in the post on replacing a spreadsheet with a tool.
Paste the numbers, set the labels and export an image.