Direct answer: after removing script/style, the 14 homepages contain 525 img tags, of which 210 have non-empty alt (40.0%). Those 210 values contain only 158 distinct texts; 52 (24.8%) repeat within their page, 63 (30.0%) look like filenames (containing .svg/.gif or a lower-case slug), 14 are shorter than 4 characters and 2 are generic labels such as "image/product". The median length is 12 characters.
This is slightly above the 38.0% alt coverage in study 97, but the denominators differ: study 97 did not remove scripts (563 img), while this study removes them first (525 img). The bigger issue is not "no alt" but "alt that looks like a filename or repeats".
Method: remove scripts first, then inspect alt text
The sample is the same as studies 95–101: 14 identifiable homepages from the 2024 Su–Xi–Chang first-set equipment list (16 companies, excluding E21 HTTP 502 and E20 1,329 bytes). Script, style and comments are removed first, then each img's alt is read; whitespace is collapsed and only non-empty values are counted. Distinct values use exact string matching. "Filename-like" means an image extension or a lower-case/number/underscore slug of at least 8 characters; "generic" means the Chinese words for image, picture, logo, banner, product, homepage, icon, image, img, enterprise, company, advertisement or promotional image.
| Item | Rule | Boundary |
|---|---|---|
| Non-empty alt | Length >0 after trimming | Empty/missing not counted |
| Distinct | Exact string differs | No synonym/translation merge |
| Repeat | Second and later on the same page | Cross-page repeats separate |
| Filename-like | Contains .svg/.gif etc. or a plain slug | Not proof it is a filename |
Result: 63 filename-like, 52 repeated
| Item | Count | Of 210 |
|---|---|---|
| Non-empty alt | 210 | 100% |
| Distinct texts | 158 | 75.2% |
| Repeated | 52 | 24.8% |
| Filename-like | 63 | 30.0% |
| Shorter than 4 chars | 14 | 6.7% |
| Generic labels | 2 | 1.0% |
Median length is 12 characters, minimum 1 and maximum 116. Many of the 14 short values are markers such as "图" or "logo"; the 63 filename-like values include .svg/.gif names and plain lower-case slugs. They pass an "has alt" check but do not describe the image for a screen reader or image search.
Repeats concentrate on company and brand names
Within the 52 repeats, the largest groups are a company's full name repeated 20 times on one page and a brand name repeated 8 times; there are also 4 template-like "bottom_20_ewm" values and 3 "logo.svg"/"LOGO1.svg" filename values. A company name is reasonable for a logo, but 20 uses on one page suggests alt is used as a default filler rather than a per-image description. To avoid naming sites, the public data contains counts and categories only, not the raw alt text.
Compared with article-page alt
| Metric | 14 homepages | 36 article pages (study 57) |
|---|---|---|
| Non-empty alt | 210 | 395 |
| Distinct texts | 158 (75.2%) | 129 (32.7%) |
| Repeated | 52 (24.8%) | 312 (79.0%) |
| Median length | 12 chars | 4 chars |
| Alt coverage | 40.0% | 76.8% |
Homepage alt is longer and less repetitive, but coverage is lower (40.0% vs 76.8%) and 30.0% is filename-like. Article pages have higher coverage but heavy repetition and very short copy. The two page types need different fixes.
What this means for acceptance
- Check coverage first, then copy. 40.0% non-empty means most images lack a usable description; fix missing alt before polishing existing text.
- Ban filenames as alt. 63 (30.0%) look like filenames; use "what the image shows + purpose" instead.
- Do not fill every image with the company name. Repeating a company name 20 times gives a screen reader no distinction; a logo may use the brand, but content images need their own text.
- Review short alt too. 14 values are under 4 characters, usually too little to describe an image.
- "Has alt" is not a pass. An empty alt may be decorative, but a filename or generic label is not a useful description.
Limits
- 14-page pilot: the same homepages as studies 95–101; not an industry estimate.
- Form only: image role, accuracy and possible harm are not judged.
- Exact matching: synonyms, translations and case differences are not merged; repeats may be undercounted.
- Scripts/styles removed: 525 img differs from the 563 in study 97, which did not remove scripts.
- No raw text: only counts and categories are published, to avoid naming sites.
- Single snapshot: fetched 2026-11-07; no redesign recheck.
Data and reproduction
Anonymised data: equipment-homepage-alt-copy-2026.csv (14 rows × 9 columns: img tags, non-empty alt, distinct, repeated, median length, under-4-char, generic, filename-like). In-repo copy: docs/research/2026-11-homepage-alt-copy.csv. No raw alt text is published.
Parser: scripts/homepage_alt_copy_probe.py; tests: scripts/test_homepage_alt_copy_probe.py (no network, 6 tests: non-empty/whitespace, script template exclusion, single quotes, counts, short copy, empty page). Run:
python -m unittest scripts.test_homepage_alt_copy_probe
Related: the homepage image-delivery pilot, the article alt-copy pilot, and the homepage image-dimensions pilot.
Sources and definitions
- Sample: 14 homepage caches from study 95 (fetched 2026-11-07; 16 companies, excluding E21 502 and E20 1,329 bytes).
- alt: read from img after removing script/style/comments; non-empty values compared exactly; raw text not published.
- Measured 2026-11-07 with no network requests.