Announcing just 42 tells a screen-reader user nothing about which series, month, or unit it belongs to

What it is

When a screen reader reaches a chart's data point, the quality of the announcement decides how much information the user receives. "42" is a nearly useless announcement—42 of what? Which series? "March, sales, 420 thousand yuan" is complete: identifier (which point: March), value (42), unit (thousand yuan)—all three present. Missing any one forces the user to fill gaps from memory or guesswork—and guesswork in a data context is the entrance to misreading.

Why it happens

The three-element mechanism maps to a data point's complete semantics: the identifier answers "which data is this" (time point, category name, series membership), the value answers "how much" (at reasonable precision), the unit answers "measured in what" (yuan, percent, people). The typical consequences of missing pieces: no identifier—the user hears a string of values but cannot align them to time or category, so comparison tasks fail; no unit—"42" is understood as 42 yuan when it means 42万元—an order-of-magnitude error is more dangerous than missing information because it produces confident wrongness. At the implementation level, announcement text should pass through a formatting layer rather than dumping raw values: numbers formatted to the metric's precision (never "42.00000001"), units composed into natural readings ("420 thousand yuan" rather than "420000"), identifiers spoken as business names rather than indexes ("March" not "data point 3"). Relative-relationship announcements are the three elements' enhancement: "March, 420 thousand yuan, up 12% from the previous month"—translating what visual users see directly (the line's rise) into words, compensating for what the graphic loses in the auditory channel.

Where it stops holding

Announcement length and information density must balance: full three elements plus relative relations on every point causes listening fatigue over long traversals; a workable layering announces the full context once on entry (series names, units), then keeps point-level announcements compact ("March, 420 thousand, up 12%," dropping the unit when context makes it unambiguous). Precision has a boundary too: announce at the graphic's displayed precision ("420 thousand" if that is what the chart shows), which matches expectations better than raw data precision and is shorter. In multilingual environments, unit and number-format localization (thousands separators, decimal marks) is the formatting layer's job, not each chart's.

Applying it

• Define announcement templates per chart type: "identifier + value + unit (+ relative relation)," implemented in one shared formatting layer.
• Use business-readable names (month names, category names) in announcements, never indexes or internal codes.
• Announce units and the full series list once on chart entry, allowing point-level announcements to be compact.
• Verification: traverse the chart listening only to the screen reader and check whether each point's "which, how much, in what unit" can be noted down; any missing element means the announcement is incomplete.

Related

• Same group: U11.02.1 Chart elements need a predictable keyboard traversal order · U11.02.2 Traversal should support switching between series and data-point levels · U11.02.4 Overall trends need a readable summary layer rather than point-by-point listening · U11.02.5 View changes caused by interaction must be announced
• Nearby: U11.02.4 Overall trends need a readable summary layer rather than point-by-point listening · U7.09.3 Data must carry a visible timestamp
• Search terms: accessible announcement · aria-live · screen reader chart