Direct answer: The 36 article pages contain 3,583 links, 3,203 of them with text. After merging all pages only 502 distinct labels remain (15.7%), and 94.2% of occurrences repeat a label used elsewhere. Most repetition comes from site navigation repeated on every page: “联系我们” appears 98 times, “公司新闻” 70 and “首页” 62.
Within a page the picture is milder: the median share of distinct labels is 57.9% (36.9%–83.7%). Also, 259 link targets carry two or more labels on the same page, 49 of 281 image-only links (17.4%) have no non-empty alt, and 159 links point to javascript: or #.
Method: reuse the study-45 article cache
The study parses the 36 article pages cached on 2026-11-03 08:40–08:41 (10 sites, HTTP 200, HTML ≥5 KB) with zero network requests. It extracts every <a> with href, strips tags and entities to get visible text, then counts length, repetition, fixed generic words and label-to-target mapping. Five E20 pages of about 1.4 KB are excluded by the size threshold.
| Item | Rule | Boundary |
|---|---|---|
| Text link | Anchor text non-empty after trimming | Buttons and JS routing excluded |
| Repeated | Appears >1 time after merging 36 pages | Includes navigation and footer reuse |
| Generic | Exact match to a fixed list | A lower bound |
| Short / long | ≤2 chars / >40 chars | Sufficiency not judged |
| Same target | Same path and query, fragment ignored | Relative links resolved first |
| Image link | No text but contains img | Needs a non-empty alt for a name |
| Placeholder | javascript:, # or empty href | Click behaviour not judged |
Result: over half distinct per page, 15.7% across pages
| Measure | Count | Share of 3,203 |
|---|---|---|
| Text links | 3,203 | 100% |
| Distinct across pages | 502 | 15.7% |
| Repeated occurrences | 3,016 | 94.2% |
| Per-page distinct sum | 1,780 | 55.6% |
| Fixed-list generic | 170 | 5.3% |
| ≤2 characters | 282 | 8.8% |
| >40 characters | 67 | 2.1% |
| Median length | 4 | — |
The per-page distinct sum is 1,780 (55.6%), close to the 57.3% measured on homepages in study 46; merged, only 502 labels remain. That gap measures template reuse, not the quality of a single page's copy.
The most repeated labels are navigation
| Anchor text | Occurrences | Note |
|---|---|---|
| 联系我们 | 98 | Usually header, footer or a floating button |
| 公司新闻 | 70 | Section navigation |
| 首页 | 62 | Logo and navigation |
| 发展历程 | 46 | About submenu |
| 企业文化 | 43 | About submenu |
| 荣誉资质 | 42 | About submenu |
| 公司简介 | 39 | About submenu |
| 关于我们 | 38 | Top-level navigation |
The top eight labels account for 438 occurrences, 13.7% of all text links. They name sections and point to stable URLs; they are not errors for readers, but they depress the cross-page distinct ratio.
Anchor texts are quoted verbatim from the sampled pages and are not translated; the sites publish in Chinese.
Generic words are 5.3%, but “see here” is rare
| Matched text | Count |
|---|---|
| 联系我们 | 98 |
| 首页 | 62 |
| 返回 | 4 |
| 下一页 | 3 |
| 查看详情 | 2 |
| 查看更多 | 1 |
Of the 170 matches, 160 are navigation words; only three are “查看详情/查看更多” and none is “阅读更多”. The fixed list is a lower bound, and short labels that miss it still need human review.
Image links, empty text and placeholders
| Type | Count | Share of 3,583 links | Note |
|---|---|---|---|
| Image-only links | 281 | 7.8% | 49 have no non-empty alt |
| Fully empty text | 99 | 2.8% | No text and no image |
javascript: | 116 | 3.2% | Handled by script |
# or empty href | 43 | 1.2% | No real target |
| Links to the same page | 44 | 1.2% | Includes “back to top” |
The 49 unnamed image links cluster in two sites: E16 has 36 and E22 has 10. A screen-reader user cannot tell where those links go. The 159 placeholders give keyboards and crawlers no real address; 43 are a bare #.
By site
| Site | Pages | Links | Text links | Distinct | Generic | Empty | Image links | Placeholder |
|---|---|---|---|---|---|---|---|---|
| E03 | 5 | 574 | 507 | 89 | 15 | 20 | 47 | 36 |
| E06 | 5 | 543 | 533 | 55 | 45 | 5 | 5 | 5 |
| E09 | 5 | 365 | 330 | 37 | 25 | 15 | 20 | 45 |
| E14 | 5 | 530 | 418 | 90 | 25 | 10 | 102 | 14 |
| E16 | 3 | 203 | 143 | 61 | 9 | 21 | 39 | 27 |
| E19 | 1 | 68 | 62 | 42 | 6 | 4 | 2 | 5 |
| E22 | 5 | 655 | 622 | 69 | 20 | 10 | 23 | 6 |
| E23 | 5 | 473 | 422 | 100 | 17 | 14 | 37 | 15 |
| E26 | 1 | 109 | 103 | 47 | 5 | 0 | 6 | 2 |
| E28 | 1 | 63 | 63 | 48 | 3 | 0 | 0 | 4 |
E06 has the most text links (533) but only 55 labels and 45 generic hits (8.4%), the most templated site in the sample; E09 has 330 text links and 37 labels plus 45 placeholders; E14 has 102 image links, the most here. E23 is the most varied with 100 labels.
Compared with homepage anchor text
| Measure | Homepages (study 46, 25 pages) | Articles (this study, 36 pages) |
|---|---|---|
| Links | 2,739 | 3,583 |
| Text links | 2,339 | 3,203 |
| Per-page distinct sum | 1,341 (57.3%) | 1,780 (55.6%) |
| Fixed-list generic | 127 (5.4%) | 170 (5.3%) |
| ≤2 characters | 165 (7.1%) | 282 (8.8%) |
| Median length | 4 | 4 |
Per-page text variety is similar on both page types; article pages carry more links because their navigation, previous/next, share and contact blocks are longer, and short labels are slightly more common.
What this means for acceptance
- Separate repetition from error. Navigation reuse is not a defect, but it hides body-link quality; acceptance should look at navigation and body separately.
- One target, one label. 259 targets carry two or more labels on the same page, so readers cannot tell whether they lead to the same place.
- Name every image link. 49 image-only links have no non-empty alt, which leaves them without an accessible name.
- Remove target-less placeholders. 159
javascript:or#links should become real URLs or button semantics. - Write body links that name their target. Generic words are 5.3% overall and only three body-level “view details” cases appeared, so structure and accessible names matter more here.
Limits
- Navigation included: page-level counts take in header and footer, so cross-page repetition is inflated.
- Fixed list: only exact matches count, so 5.3% is a lower bound.
- No semantic judgement: repetition is not automatically wrong; the study records length, repetition and mapping only.
- Snapshot: one cache from 2026-11-03, with no before/after redesign comparison.
- 10 sites, 36 pages: a pilot, no industry claim.
Data and reproduction
Anonymised data: equipment-article-anchor-text-2026.csv (36 rows × 18 columns: links, text links, empty text, image links and missing alt, generic, short/long, distinct, median length, one label to many targets, one target with many labels, placeholders, self links and the top label). The repository keeps docs/research/2026-11-article-anchor-text.csv.
Parser: scripts/article_anchor_probe.py; unit tests: scripts/test_article_anchor_probe.py (no network). Run:
python -m unittest scripts.test_article_anchor_probe
Related: the homepage anchor-text pilot, the article link-structure pilot, the link accessible-name pilot, and the article alt-copy pilot., and the article inquiry-path pilot
Sources and definitions
- Article cache: 36 pages from 10 sites in study 45 (2026-11-03 08:40–08:41).
- Anchor text: visible text of
<a href>, tags stripped, entities decoded, whitespace collapsed. - Same target: relative links resolved, then path and query compared with the fragment ignored.
- Scan: 2026-11-05, zero network requests.