Adding Text Labels Next to Icons: A/B Test Example

This test is worth running when visitors seem to struggle with what the icons mean: support tickets ask what an icon does, session recordings show people tapping icons by trial and error and backing out, or an important function goes almost unused. If the label helps, visitors reach the right function without decoding the icon, and misclicks drop. If the function sits somewhere hard to find or is not something visitors need, question its placement and purpose first, because a label will not make people use a function they are not looking for.

Use of the labeled function

Compare completion of the function represented by the selected icons among eligible assigned visitors.

Also track

• Action completion rateCompare completion of the action represented by the icon.
• Misclick rateMeasure clicks that do not complete the intended function or lead somewhere else.
• Function-related support requestsCompare questions about what the icon means using the same traffic denominator.

• LabelsAdd visible text labels beside selected icons.
• Icon setChoose the functions in scope before the test and keep the icons unchanged.
• PlacementUse one fixed label position.
• SpaceCheck space for nearby controls and content.
• Device and languageReview layouts where localized labels have different lengths.

1. 1Read any rise in use of the labeled function together with the misclick rate; if use and misclicks rise together, the label may not have made the meaning clearer. A change in function-related support requests adds context but does not decide the test on its own.
2. 2Do not count a hover-only tooltip as an added label; this test measures a persistent visible one. If tap targets shrank to fit the labels, do not treat the gain as a reason to ship. Keep accessible names available to screen-reader users in both arms instead of relying on the visual label, so the icon-only version still has a clear name.
3. 3Your result is bound to the language you measured and the selected icon set; re-test in that layout before moving labels to languages with much longer text or to narrow screens. If the target function is rarely used, enter its current usage rate in the sample size calculator before launch to learn the visitors per arm needed to catch a small difference.