Direct answer: The 36 article pages contain 395 non-empty alt values, but only 129 distinct texts (32.7%); 312 occurrences (79.0%) repeat a text used elsewhere. The most repeated is the placeholder “图片名称” (47 times), followed by one brand name (25), “二维码” (20) and “电话” (11).
The median alt length is 4 characters, and 72 (18.2%) are 2 characters or fewer; only one is filename-like by extension and three match the fixed generic list. Coverage (76.8%) looks acceptable, but the copy is highly repetitive and short.
Method: reuse the article pages and the existing alt-quality rules
The study reads the 36 article page caches from study 45 (2026-11-03 08:40–08:41, 10 sites) with zero network requests. It extracts non-empty alt values, counts distinct texts, repeats, length and short values per page, then merges all pages for a global dedupe.
| Item | Rule | Boundary |
|---|---|---|
| Repeated occurrence | The same text appears >1 time across all 395 | Includes cross-page and cross-site repeats |
| Short text | ≤2 characters | Semantic sufficiency not judged |
| Filename-like | Ends with .jpg/.png and similar | IMG_ prefixes without extension are not counted |
| Generic | Exact match to a fixed list | A lower bound |
| Page level | All images on the page, including nav and footer | Body not separated |
Result: only a third of the texts are unique
| Measure | Count | Share of 395 |
|---|---|---|
| Non-empty alt | 395 | 100% |
| Distinct texts | 129 | 32.7% |
| Repeated occurrences | 312 | 79.0% |
| ≤2 characters | 72 | 18.2% |
| Filename as alt | 1 | 0.3% |
| Fixed-list generic | 3 | 0.8% |
| Median length | 4 | — |
| Repeated text (anonymised) | Occurrences |
|---|---|
| Placeholder “图片名称” | 47 |
| Brand name A | 25 |
| “二维码” | 20 |
| Brand name B | 12 |
| “电话” | 11 |
Per-page distinct counts sum to 279; after merging pages only 129 remain, showing heavy reuse across pages and sites.
These labels describe an image type, brand or contact method rather than the image's content; the same text on different images gives a screen-reader user no way to tell them apart.
By site: one site uses only five texts
| Site | Non-empty alt | Distinct | Repeated | Median length | ≤2 chars |
|---|---|---|---|---|---|
| E03 | 87 | 38 | 58 | 5 | 25 |
| E06 | 51 | 10 | 50 | 4 | 1 |
| E09 | 50 | 5 | 50 | 2 | 30 |
| E14 | 108 | 48 | 64 | 4 | 0 |
| E16 | 3 | 1 | 3 | 4 | 0 |
| E19 | 2 | 2 | 0 | 23 | 0 |
| E22 | 38 | 6 | 38 | 3 | 10 |
| E23 | 51 | 21 | 44 | 13 | 6 |
| E26 | 4 | 1 | 4 | 12 | 0 |
| E28 | 1 | 1 | 0 | 7 | 0 |
E09 has 50 alt values but only five texts, 30 of them two characters or fewer; E06 has 51 values and 10 texts; E22 has 38 values and six texts. E14 is the most varied with 48 texts out of 108, yet still has 64 repeated occurrences.
Compared with homepage alt quality
| Measure | Homepages (study 40) | Articles (this study) |
|---|---|---|
| Non-empty alt | 502 | 395 |
| Distinct texts | 353 (70.3%) | 129 (32.7%) |
| Median length | 6 | 4 |
| Repeated occurrences | 190 | 312 |
| ≤2 characters | 39 | 72 |
Article alt coverage is not low, but uniqueness is far below the homepage sample: only about one third of the texts are distinct, and nearly 80% of occurrences repeat.
What this means for acceptance
- Coverage is not copy quality. 76.8% have alt, but only 32.7% are distinct texts.
- Replace placeholders first. 47 images share “图片名称”, a template placeholder.
- Brand, QR and phone are not image descriptions. They cannot say what the image shows.
- Expand short values. 72 are two characters or fewer and the median is only 4.
- Write per-image descriptions. Applying one text in a template produces mass repetition.
Limits
- Strict filename rule: only text with an extension counts, so one is a lower bound.
- Limited generic list: only fixed words match, so it is a lower bound.
- No semantic judgement: repetition is not automatically wrong, but it cannot distinguish images.
- Page level: navigation and footer images are included; body is not separated.
- 10 sites, 36 pages: a pilot, no industry claim.
Data and reproduction
Anonymised data: equipment-article-alt-quality-2026.csv (36 rows × 11 columns: alt count, distinct, repeated, median length, short, filename, generic and top-count). The repository keeps docs/research/2026-11-article-alt-quality.csv.
Parser: scripts/alt_quality_probe.py; unit tests: scripts/test_alt_quality_probe.py (no network). Run:
python -m unittest scripts.test_alt_quality_probe
Related: the article image alt pilot, the homepage image alt coverage, and the article image attributes pilot., and the alt text quality checker, and the article anchor-text pilot
Sources and definitions
- Article cache: 36 pages from 10 sites in study 45 (2026-11-03 08:40–08:41).
- Alt: non-empty img alt values, trimmed.
- Repeat: texts appearing >1 time after merging all 395 values.
- Scan: 2026-11-05, zero network requests.