Distinctiveness of labels affects choice more than precision

What it is

Two labels can both be defensible in a dictionary and still be useless for choosing if they cannot be told apart in the layer. “Solutions” and “Services” may each describe what follows and still fail as an elimination pair. Distinctiveness outranks accuracy: a label’s first job is to create an opposable difference among the options on this layer, not to survive a solitary definition check. A slightly coarse word that can eliminate its neighbour beats two precise words that are true together.

Accuracy still matters. It stops scent from pointing at the wrong object. At the moment of “which one,” though, people compare labels with each other, not each label with an encyclopaedia.

Why it happens

Choice is a contrast process. Several words sit in view, and working memory subtracts them pairwise: which is more like what I want. When the difference is zero, accuracy cannot help, because both words are “right” for the target. Gaze oscillates, or the left-hand item is clicked from positional habit. Information scent is relative: the same word can become distinctive beside a different neighbour—“Account” is clear next to “Help” and unclear next to “Personal centre.”

So labels cannot be optimized one at a time toward “more precise.” They have to form minimally separable pairs inside a fixed neighbour set. Splitting one confusable pair often cuts misclicks more than making every word more official.

Studying it

Use a confusion matrix, not Likert ratings of “is this word accurate.”

• Paradigms: pairwise judgments on same-layer labels (do they point at the same kind of thing); tree tests or first clicks with errors coded as confusable pairs; then a rename experiment that changes distinctiveness while holding extension roughly constant.
• Independent variables: semantic distance among same-layer labels, presence of synonym pairs, accurate-but-abstract versus coarse-but-opposed.
• Dependent variables: share of errors on confusable pairs, decision time, gaze trips between two labels.
• Methodological note: presented alone, every label can look accurate; presented side by side, inseparability appears. Measure in the real neighbourhood. Experts hear a large difference because they know backend module boundaries; users do not have that diagram.

Where it stops holding

Statutory names cannot be changed for distinctiveness (drug names, standard numbers); difference has to come from neighbour choice and added qualifiers (form, indication), not from inventing a friendlier false name. Imperative micro-labels (Save, Delete, Export) are already highly opposed; nouning them for “accuracy” weakens contrast. When a brand is used as a label, distinctiveness comes from whether the brand is already used as a class, not from literal accuracy.

Applying it

• Pair every label on a layer and mark pairs whose difference cannot be stated. Change those pairs first, not the words that “don’t sound professional.”
• Prefer an opposition axis when renaming (personal/business, buy/sell, in progress/done). Do not swap in another synonym.
• Accurate but abstract words may remain as on-page titles; entrance labels need words that eliminate their neighbours.
• Verify by covering page content and leaving only that layer of labels. Ask someone uninvolved the difference between any two. “Pretty much the same” marks where accuracy is already enough and distinctiveness is not.

Related

• Within the group: G1.04.1 Labels must use the user’s words, not internal jargon · G1.04.2 One concept gets one label across the site
• Adjacent: G1.03 Taxonomy and hierarchy · G2.08 Predictability of navigation items · T2.02 Predictability of titles and labels
• Search terms: label distinctiveness · information scent · confusable category names