With several things happening in one chart, an annotation is what tells readers which one matters

What it is

A data chart usually contains multiple attendable features at once: trends, outliers, crossovers, plateaus. Readers do not know which feature matters, nor whether the one they noticed is the one the author intended. Annotations (an arrow to a point labeled "v2.0 launched March 15," a circle around an anomalous segment, "target 5%" on a threshold line) make explicit "where to look and how to understand what you see." An annotation is an explicit contract between author and reader about what deserves attention—it transfers the interpretive work from the reader to the author.

Why it happens

Annotations work through directed visual attention: annotated elements (arrows, color marks, text labels) are salient stimuli that capture attention first, steering the reader's initial scan to the annotated position and building understanding outward from there. This contrasts with the unannotated chart, where attention defaults to the most visually salient feature (the tallest bar, the boldest line)—which may have nothing to do with the author's narrative: the tallest bar might be an artifact of the axis minimum, while the real story sits in an inconspicuous gray line at the lower right. Annotations do more than "look here"—they also say "read it this way": labeling a key point "new feature shipped" tells the reader why the point matters and what it should be attributed to. The cost is visual noise: every annotation competes with data elements for attention, and too many (every line labeled, every point arrowed) destroys the reader's sense of priority, reverting to free scanning—annotation value comes from scarcity; three to five annotations targeting key features beat comprehensive coverage every time.

Where it stops holding

Annotation placement and density face design constraints: annotation text must not occlude the data elements themselves (text covering the bar the arrow points at is the most common design error), and with too many annotations readers do not know which to read first. Suitability varies by context: charts in formal reports can carry denser annotations (readers have focused reading time), while small charts embedded in dashboards can carry only one or two—or rely entirely on tooltips for annotation content. There is also an interaction boundary—in interactive charts, annotations can be progressive: show only the one or two most critical by default, revealing the rest on hover or click, preserving first-glance simplicity while keeping details reachable.

Applying it

• Explanatory charts carry at least one annotation by default, pointing at the key feature supporting the title's conclusion.
• Follow a "point + explain" two-layer structure: a visual marker (arrow/circle) locates the position, text explains the meaning (event, cause, judgment).
• Cap annotations at 5; beyond that, filter by importance and move the rest into tooltips or footnotes.
• Verification: show the chart to a first-time reader for 15 seconds and ask "what does this chart most want you to notice?" If their answer does not match the annotated feature, the guidance has failed.

Related

• Same group: U9.03.2 Annotations carry causal and contextual explanation · U9.03.3 Unannotated charts are interpreted idiosyncratically
• Nearby: U9.02.2 Conclusion-style titles guide correct interpretation · U9.01.3 The two differ in interactivity and annotation density
• Search terms: chart annotation · visual guidance · callout