Form Label Coverage: A Metric That Counts Labels and Misses Meaning
Form label coverage, the share of inputs with a programmatically associated label, is the metric compliance dashboards lead with, because it is easy to compute and usually improves fast. A site can sit at 98 percent coverage and still have checkout forms that screen reader users cannot complete with confidence, because coverage counts the presence of a label. It says nothing about whether the label describes the field it is attached to.
Criterion 1.3.1 requires the relationship to exist. Criterion 3.3.2 requires labels or instructions where input is needed. Neither is satisfied in spirit by a label that reads Field 1, or by the word Search attached to a postcode box because a template reused the wrong string. The scanner sees an association. The user hears nonsense, correctly, at the exact moment they are typing payment details.
The failure modes coverage cannot see
Wrong labels are the first: copy pasted templates where the accessible name describes a different field. The second is placeholder as label, counted as covered by some tools when a real label element is absent, and gone the moment the user starts typing, taking the field's identity with it. The third is the visible label and the accessible name disagreeing, which breaks voice control users under 2.5.3 Label in Name: they speak the words they can see, and the field, named differently in code, does not respond.
Each of these passes a coverage check. Each of them fails a person. A metric programme that reports coverage without a label quality sample is measuring the paperwork of labelling rather than the experience of being labelled for.
Add a sampled quality audit, small and regular
Take twenty labelled fields a month from your highest traffic forms and score three things by hand, with a screen reader where possible: does the accessible name match the visible label, does it describe this field and no other, and do format instructions travel with the field rather than sitting in a paragraph above the form. Twenty fields is an afternoon. The defect rate in that sample is the number that predicts complaints.
Track it as its own line, label quality defect rate, next to coverage. Coverage will sit near 100 and barely move, which is fine, it becomes a hygiene check. The quality rate is the one that should trend down quarter by quarter, and it is the one that will catch the template regression before a user reports it in the worst way.
Fix at the component, not the page
Label defects cluster in components: one bad checkout input pattern replicated across address, payment, and account forms. That is good news disguised as scale. Fixing the component fixes every instance, and the quality sample will show it in the next cycle. Page by page whack a mole, by contrast, produces coverage reports that improve while the underlying pattern keeps shipping new instances of the same error.
Write the labelling rules into the component itself: visible label text and accessible name generated from the same source, never typed twice. Most label mismatches are born as a duplication, two strings that started identical and drifted. Remove the duplication and the defect class dies at the root.
Related reading: Keyboard Trap Count: The Metric That Should Always Read Zero, Alt Text Coverage Rate: A High Number That Can Still Mean Failure, Target Size Pass Rate: Catching Clickable Boxes Too Small to Use.
Frequently asked questions
What does form label coverage actually measure?
The percentage of inputs with a programmatically associated label. It confirms a relationship exists. It does not check that the label is correct, matches the visible text, or describes the field it is attached to.
Can a form pass label coverage and still be unusable?
Yes. Wrong labels, placeholder only labels, and accessible names that differ from visible labels all count as covered in many tools while failing screen reader and voice control users in practice.
What is the Label in Name criterion?
WCAG 2.5.3 requires the accessible name of a control to contain the visible label text, so voice control users can activate it by speaking what they see. A mismatched name passes coverage and fails these users completely.
How do I measure label quality without a huge audit?
Sample twenty fields monthly from high traffic forms and score name match, accuracy, and whether instructions travel with the field. Report that defect rate alongside coverage. It takes an afternoon and predicts real complaints.