Writing Alt Text for Scientific Figures: Charts, Graphs, and Diagrams That Actually Work
Most alt text on scientific figures isn’t missing — it’s useless. “Bar chart of results,” “Figure 3,” “graph showing data.” It satisfies a checklist and tells a blind reader nothing. The problem is almost never effort; it’s aim.
Here is the principle that fixes most of it, in one line: describe the data, not the picture. A sighted reader doesn’t take “a blue bar chart with gridlines” from a figure — they take a finding. Your job is to give the reader who can’t see it that same finding, and the means to explore the numbers behind it. Everything else about writing alt text for charts, graphs and diagrams follows from that.
“Alt text” for a complex figure is really three layers
WCAG 1.1.1 requires a text alternative for every informative image. For a photo, one line does the job. For a scientific figure, one line physically cannot carry the information — so the working model is layered:
- Short alt text — a concise summary of the figure’s type and its single most important message.
- A long description — the essential information the figure conveys, very often best delivered as a data table rather than prose.
- The data itself, where possible — so a screen-reader user can navigate the numbers independently instead of being read a paragraph at them.
Getting the split right is the whole craft: enough in the alt text to orient the reader, with the detail moved somewhere they can explore at their own pace.
The difference in one example
Take a revenue chart. Here is alt text that passes a checker and fails a reader:
“Bar chart of company revenue.”
And here is alt text that actually works:
“Bar chart: company revenue rose from $5M in 2021 to $12M in 2023, a 140% increase.”
The second one took no more effort — it just aimed at the finding instead of the decoration. Note what it leaves out: the colours, the gridlines, the 3-D bevel, the legend position. None of that is information; it is styling, and styling is not what the reader came for.
A reliable formula for the short text: [chart type] of [what], showing [the key message], with [axes and units]. Fill those four slots and you have a working description almost every time.
For anything with more than a handful of points, the long description is best delivered as a table the reader can navigate. The same revenue chart becomes:
| Year | Revenue |
|---|---|
| 2021 | $5M |
| 2022 | $8M |
| 2023 | $12M |
Now a screen-reader user can move through the numbers cell by cell — exactly what a sighted reader does with their eyes.
Match the description to the figure type
- Charts and graphs. Alt text states the chart type, the key trend or finding, and the axes and units. The long description carries the numbers — and for anything beyond a few data points, a data table is far better than prose, because the reader can move through it cell by cell. Don’t narrate every point; give the shape of the data and then the data.
- Diagrams, flowcharts and schematics. Describe the process, order and relationships, not the shapes. A flowchart or an organisation chart is usually clearer to a screen-reader user as a nested bulleted list than as a sentence trying to trace arrows in words.
- Chemical structures and equations. These are data, not pictures. “Two fused hexagons” helps no one — they should be encoded in the proper accessible notation (MathML for equations, structured chemistry markup for molecules), so the meaning survives.
- Maps. Describe the pattern the map exists to show (“incidence is concentrated in the south-east”), then provide the underlying figures. A map is a visualisation of data; give the data.
- Multi-panel or grouped figures. If the grouping means something, state that meaning first, then describe each panel in turn so the reader understands how the parts relate.
Context changes everything
The single most overlooked rule: the same figure is described differently depending on what it is being used to show. A scatter plot used to make a point about one outlier should foreground that outlier; the same plot used to show an overall correlation should foreground the trend. Alt text isn’t a neutral inventory of the image — it is a description written for the argument the figure supports here.
Which leads to the second half of the rule: don’t duplicate the caption or the body text. Alt text and captions should complement each other, not repeat. And if the surrounding text already fully explains the figure, you may not need a separate long description at all — though the image still needs alt text so a reader knows what is there and that nothing was withheld.
Small rules that keep you out of trouble
- Don’t open with “image of” or “photo of” — the screen reader already announces it’s an image. Naming the chart type, though, is useful.
- Keep the alt text short and put the detail in the long description; if a description is coming, say so in the alt text.
- If an image is genuinely decorative, give it null alt text (an empty
alt="") so it is skipped — and so the omission reads as a decision, not an oversight. Scientific figures almost never qualify. - Never leave the alt text as the file name. “fig3_final_v2.png” is not a description.
- Write the description in the same language as the content.
How it’s actually delivered
The short text goes in the alt attribute. The long description can live in visible text near the figure, in a data table, inside a <figure>/<figcaption>, or be linked and associated programmatically (modern mechanisms such as aria-details or a <details> element; the old longdesc attribute is deprecated). In accessible EPUB and in journal XML, the point is the same: the link between the figure and its description must be machine-readable, so assistive technology can find the description, not just a sighted editor.
Why good scientific alt text needs a specialist
The hard part of this work is not the markup — it is judgement. Deciding which trend is the point, which numbers matter, how to render a reaction scheme or a nested process, what to leave out: all of that requires someone who understands the science, not just the syntax. It is exactly why automated and AI-generated alt text falls down on scientific figures — a model can tell you a chart is blue and has bars; it cannot reliably tell you what the chart means in context. That understanding is the difference between a figure that is technically “described” and one a reader can actually use — and it is the part we don’t automate.
The bottom line. Accessible figures aren’t about adding more words — they’re about adding the right ones. Name the figure, give the finding, hand over the data, and let context decide the emphasis. Do that and a reader who can’t see the chart still leaves with exactly what it was drawn to show.
Frequently asked questions
How long should alt text be?
Short — a sentence or two that gives the figure type and its key message. Anything more detailed belongs in a long description, not the alt attribute.
Do I have to describe every data point?
No. Describe the shape of the data (the trend, the outlier, the comparison) and then provide the numbers themselves, ideally as a navigable data table.
What about purely decorative images?
Give them null alt text (alt="") so screen readers skip them. Genuine scientific figures are almost never decorative.
Can’t the caption serve as the description?
Alt text and captions should complement, not duplicate. If the body text already fully explains the figure you may not need a long description, but the image still needs alt text.
Can AI write alt text for scientific figures?
It can draft text for simple images, but for charts, diagrams and scientific content it describes appearance rather than meaning, so it needs expert review to be trustworthy.
This article reflects our practical approach to figure accessibility; for detailed, worked examples the DIAGRAM Center’s Image Description Guidelines are an excellent reference.