Direct answer: Of the 391 links without visible text counted in study 43, 254 (65.0%) still have an accessible name once aria-label, the inner image alt and title are merged: 232 from image alt, 17 from title and 5 from aria-label. The remaining 137 (35.0%) have no name at all in the static HTML, across 20 sites.
Another number: the 2,739 links contain 567 inner images, 274 of them with an empty alt (48.3%). An empty alt means decorative and gives no name; if that image is the only content of the link, the link has no name.
Method: take the name by priority
The same 25 cached homepages as study 43, still with zero network requests. For every <a> with href, the first non-empty source is taken in this simplified accessible-name order:
| Priority | Source | Handling |
|---|---|---|
| 1 | aria-label | Non-empty attribute is used |
| 2 | aria-labelledby | Reference recorded, id not resolved |
| 3 | Visible text | Non-empty after tags removed |
| 4 | Inner image alt | First non-empty alt |
| 5 | title | Non-empty attribute |
“No visible text” keeps the study 43 definition: empty after inner tags are removed. An empty alt is decorative and gives no name. No JavaScript runs and CSS-hidden text is not computed.
Result: 391 textless links, 254 named
| Name source | Count | Share of 391 |
|---|---|---|
| Inner image alt | 232 | 59.3% |
title | 17 | 4.3% |
aria-label | 5 | 1.3% |
| No name at all | 137 | 35.0% |
| Total | 391 | 100% |
In other words, 65.0% of the 391 “textless” links are not actually unnamed. The difference is the layer counted: study 43 counted visible text, this one counts accessible name.
Across all links, the 137 unnamed ones are 5.0% of 2,739 and appear on 20 of the 22 sites that have links.
137 unnamed links across 20 sites
| Site | Unnamed links | Note |
|---|---|---|
| E17 | 20 | Also has 5 aria-label links |
| E16 | 19 | 13 inner images, 12 with empty alt |
| E13 | 18 | 35 named by alt |
| E26 | 12 | 22 named by alt |
| E11 | 8 | All 13 inner images have empty alt |
| Other 15 sites | 60 | — |
| Total | 137 | 20 sites |
The top five sites hold 77 (56.2%). E27 and E30 name all 7 of their textless links (E27 by title, E30 by alt) and are the only two link-bearing sites with zero unnamed links.
Images inside links: 567, 274 with empty alt
| Item | Count | Note |
|---|---|---|
| Images inside links | 567 | Across 22 sites with links |
| Empty alt | 274 | 48.3%, treated as decorative |
| Links named by alt | 232 | Across 17 sites |
An empty alt is not a mistake by itself: decorative images should have one. The problem is a link whose only content is such an image, because it then has no name for a screen reader. The 137 unnamed links are the indirect count of that case.
Only 5 aria-labels, 17 titles
aria-label: 5, all on E17, and none has visible text; no case overrides visible text.aria-labelledby: 0.title: 17 across 5 sites; E12 6, E27 5, E13 3, E04 2 and E29 1.
Title can supply a name, but whether it is shown on touch and keyboard depends on the browser; it is usually a supplement, not a sole name.
Compared with study 43
| Measure | Result | Share of 2,739 |
|---|---|---|
| Study 43: no visible text | 391 | 14.3% |
| Study 44: textless but named | 254 | 9.3% |
| Study 44: no name at all | 137 | 5.0% |
The two studies do not conflict: one counts “no visible word”, the other “no readable name”. Acceptance should check the latter first, then look back at the former.
What this means for acceptance
- Do not fail a link just for having no text. 65.0% of the 391 are named by alt, title or aria-label.
- Merge alt before concluding. Image alt supplies 232 names, the largest layer.
- Check empty-alt links separately. Only 274 empty-alt images that sit in an image-only link cause a missing name.
- Do not rely on title alone. The 17 title-only links should get visible text or an aria-label.
- aria-label is rare. Just 5, all on one site; when needed it is the cheapest fix.
Limits
- Static approximation: no JavaScript, no CSS-hidden text; names added later by script are invisible.
- Not full accname: role=presentation, an image's own aria-label and aria-labelledby id resolution are not implemented.
- Title counted as a name: weak practice, reported separately so readers can exclude it.
- Regex parsing: nested anchors and commented links are not handled.
- 25-site pilot: no industry claim.
Data and reproduction
Anonymised data: equipment-link-names-2026.csv (25 rows × 15 columns: links, visible text, aria-label, image alt, title, labelledby, unnamed, no visible text, per-source split, inner images and empty-alts). The repository keeps docs/research/2026-11-link-names.csv.
Parser: scripts/link_name_probe.py; unit tests: scripts/test_link_name_probe.py (18 tests, no network). Run:
python -m unittest scripts.test_link_name_probe
Related: the homepage link structure pilot (study 43), the homepage semantics pilot, the homepage image alt coverage and text study, and the article page link structure pilot.
Sources and definitions
- Homepage cache: 25 homepages from study 4 (2026-10-31 to 11-01).
- Name order: aria-label → aria-labelledby (recorded only) → visible text → image alt → title.
- Empty alt is decorative; the no-visible-text definition matches study 43.
- Scan: 2026-11-03, zero network requests.