Direct answer: The 25 cached homepages contain 978 images, of which 502 have non-empty alt (51.3%). But having alt is not the same as writing it well: those 502 texts contain only 353 distinct strings, the median length is 6 characters, 49 are filenames (such as LOGO1.svg, all from E02), 39 are two characters or fewer, and 190 images share a repeated string. E03 has 56 images using the placeholder “图片名称”.
Study 38 counted whether alt exists; this study stays on the same pages and looks at what it says: a description, a filename or a template placeholder. Still zero requests, cached HTML only.
Method: three coverage states, four quality checks
Pages come from the study-4 homepage cache. Coverage keeps the non-empty / empty / missing split; quality checks four things: filename-like text (ending in .jpg/.png/.svg and similar), very short text (≤2 characters), very long text (>60 characters) and repeated text (the same string used by multiple images). Semantic accuracy is not judged.
| Item | Rule | Boundary |
|---|---|---|
| Filename-like | alt ends with a common image extension | May be a CMS default |
| Very short | Length ≤2 characters | Cannot describe the image |
| Very long | Length >60 characters | Possible keyword stuffing |
| Repeated | Same string used by ≥2 images | Company name, placeholder or UI word |
Coverage: 502 of 978 have alt
| Item | Count | Share |
|---|---|---|
| Images | 978 | 100% |
| Non-empty alt | 502 | 51.3% |
| Empty alt | 241 | 24.6% |
| Missing alt | 235 | 24.0% |
Text: 353 distinct strings, 49 filenames
| Metric | Count | Note |
|---|---|---|
| Non-empty alt strings | 502 | 353 distinct |
| Median length | 6 chars | Most are short nouns |
| Filename-like | 49 | All from E02 (LOGO1.svg, AboutKATOP.svg, …) |
| ≤2 characters | 39 | Such as “图” or a short brand |
| >60 characters | 10 | Possible keyword stuffing |
| Generic words | 3 | logo / icon |
| Images with repeated text | 190 | 37.8% of the 502 |
| Most-reused string | 56 images | E03's “图片名称” |
Four concrete examples
- E02: 49 alt strings are filenames.
LOGO1.svg,AboutKATOP.svg,Milestones.svg— as if the CMS copied the filename into alt. - E03: 56 images share “图片名称”. A template placeholder, which is effectively no description; the same site has 12 normal strings.
- E24: 4 images share “Wps图片(1)”. A trace of pasting from a document tool, unrelated to the page.
- E14: 5 images share “查看更多”. That is a UI word, not an image description.
Two extremes: four complete, four empty
| Type | Sites | Note |
|---|---|---|
| All non-empty alt | E06, E09, E14, E23 | 47, 23, 26 and 53 images all covered |
| None | E01, E11, E27, E28 | 14–40 images, zero non-empty alt |
| No images | E18, E20 | Excluded from the rate |
What this means for acceptance
- Check coverage, then text. Of 502 strings, 190 repeat and 49 are filenames; coverage alone overstates readability.
- List placeholders separately. “图片名称” and “Wps图片(1)” should count as unfilled.
- Filename alt is a common default. When a CMS copies filenames, someone has to override them.
- Judge repeated text by purpose. A company name on a logo is fine; the same string on 20 product images is not.
- Semantic accuracy still needs a human. This study checks form, not whether the text fits the image.
Limits
- Cached, zero-request: pages are frozen at 2026-10-31 to 11-01.
- Form rules: filename, length and repetition are proxies, not semantic quality.
- No JavaScript: lazy-loaded and script-inserted images are excluded.
- No purpose classification: decorative and content images follow the same rule.
- 25-site pilot: no industry claim.
Data and reproduction
Coverage data: equipment-image-alt-pilot-2026.csv (25 rows). Text data: equipment-alt-quality-2026.csv (25 rows: total strings, distinct, median length, filename/short/long/generic counts, repeated images and most-reused count). The repository keeps docs/research/2026-11-image-alt-pilot.csv and 2026-11-alt-quality.csv.
Parsers: scripts/img_alt_probe.py (coverage, 10 tests) and scripts/alt_quality_probe.py (text, 9 tests). Run:
python -m unittest scripts.test_img_alt_probe scripts.test_alt_quality_probe
Related: the article-page image alt pilot (514 images), the homepage privacy and analytics pilot, the title-date distance pilot, and the homepage semantics pilot (main on 3/25, 12 without H1).
Sources and definitions
- Homepage cache: 25 homepages from study 4 (2026-10-31 to 11-01).
- Alt states and text rules: non-empty/empty/missing; filename, length, generic words, repetition.
- Scan: 2026-11-03, zero network requests.