A scenario prompt for a line chart with a real anomaly in the data, and the judgment call of whether the alt text should flag that the spike isn't what it looks like.
Works with Claude / GPT830 uses★ 4.4
The prompt
media-chart-data-viz-alt-text
We have a line chart in a blog post showing monthly website traffic over the past
12 months. Traffic is flat around 40,000-45,000 sessions most months, but there's a
sharp spike to 210,000 sessions in one month, which we later confirmed was bot
traffic from a scraper, not real growth, and we've noted that in the surrounding
post text.
[replace with your own chart data and any known data-quality caveats]
Write:
1. Alt text describing the overall trend (flat, with one anomaly) and naming the
approximate scale of the anomaly, without listing all 12 monthly values, which
would be too long and not useful read aloud.
2. A decision on whether the alt text itself should mention that the spike was bot
traffic, or whether that caveat belongs only in the surrounding paragraph text.
Consider that a screen reader user who only hears the alt text (if they skip
past the paragraph) would otherwise come away thinking real traffic tripled.
3. A caption (visible) that can carry the numeric detail the short alt text left
out, since the caption doesn't have the same length pressure as the alt
attribute.
The instinct is to describe only what’s visually in the chart. But if you know the spike is a data-quality artifact and a sighted reader will see that caveat in the surrounding text, a screen reader user deserves the same correction, not just a description of a shape that happens to be misleading.