A text label sharply raises recognition and first-use success
Aliases: labelled icon · icon caption · first-use success
What it is
The words beside an icon are not a caption for a picture that already speaks. They are the channel that makes the first attempt succeed. Without them, the user has to decode a drawing before deciding to tap; with them, the action is readable and the glyph drops back to a locating mark. First-use success tracks the words, not the line work. Experts later scan glyphs alone — that is a shortcut after learning, not proof that newcomers can do without type.
Why it happens
Reading a word is faster and more stable than solving a compressed picture. A word collapses candidates to one; a picture lights several associations at once. The label does not decorate. It turns an open guess into closed recognition: seeing "Share" removes the need to decide whether a node drawing means network, link, or forward. The glyph still earns its keep as a landing mark that separates this item from its neighbours during a scan. It no longer names. Removing the label hands naming back to a drawing that was never built to name.
Studying it
The contrast is blunt: the same functions, glyph-only versus glyph-plus-label, scored on correct first action and time to the right control — not on whether the set "looks nicer". Participants must be new to the interface; an internal review spends the learning effect before the test. A second pass after practice separates first-use from skilled performance. Label gains concentrate on the first few trials; the two conditions then converge. That convergence is not "the label was redundant". It is the shortcut appearing.
Independent variables: label present or absent, function name versus a descriptive phrase, label always visible versus revealed on hover. Dependent variables: first-try accuracy, time to first success, wrong taps. Hover-only labels measure a discovery cost, not accompanying reading.
Where it stops holding
- A screen-reader name is a different channel. It does not replace words that sighted users can scan on the surface.
- In a very tight toolbar labels will blow the layout. That is a space problem, not evidence that labels fail; if space is gone, cut functions or change the container rather than assume the glyph is enough.
- Experts may hide labels. That does not license a default with none.
- This is the payoff of labels as a naming channel, not which functions may go bare, and not how many columns a label occupies.
Applying it
- For navigation, uncommon toolbar actions, and anything on a new-user path, ship icon plus words. Treat a scannable name as part of the function, not as chrome.
- Do not substitute a hover tip for a standing label. Hover does not exist on touch, and on desktop it turns discovery into a second move.
- How to check: give the task to people who have never seen the product and score whether they hit the target control on the first try. If three redraws of the glyph leave that rate unchanged, stop redrawing and add words.