Direct answer: The 25 cached homepages contain 2,739 links, of which 2,339 have visible anchor text. Only 127 (5.4%) match an explicit generic list (more, view details, click here and similar), led by “查看更多” (63) and “查看详情” (29). But 81 anchor texts each point to two or more different URLs, covering 397 links (17.0%), and 283 URLs carry at least two different labels on the same site.
The median anchor length is just 4 characters, and 165 (7.1%) are 2 characters or shorter. The most common navigation and CTA labels are “联系我们” (54) and “首页” (22), reported separately rather than as generic.
Method: fixed list plus text dedupe
The study reuses the 25 cached homepages from study 4 with zero network requests. It counts only links with an href whose visible text is non-empty after tags are removed; the 391 textless links from study 44 stay out of the denominator, and 9 anchors containing only punctuation such as “>” are also excluded.
Anchor text is normalised by removing tags, unescaping entities, collapsing whitespace, trimming edge punctuation and lowercasing for comparison. The generic group is a fixed list (查看更多, 查看详情, 更多, 了解更多, 详情, 点击这里, read more, learn more and similar); navigation and CTA labels (联系我们, 首页, 返回, 立即咨询 and similar) are counted separately so normal menus are not treated as generic.
| Measure | Definition | Boundary |
|---|---|---|
| Generic anchor | Matches the fixed list | Phrases outside it are not counted; a lower bound |
| Shared label | Same normalised text points to ≥2 hrefs | Compared within one site |
| Several labels per URL | Same href has ≥2 different texts | Header/footer variants may be intentional |
| Anchor length | Python len in characters | No word-count conversion for mixed text |
Result: few generic words, many shared labels
| Generic anchor | Count | Share of 2,339 |
|---|---|---|
| 查看更多 | 63 | 2.7% |
| 查看详情 | 29 | 1.2% |
| 更多 | 16 | 0.7% |
| 了解更多 | 15 | 0.6% |
| 详情 | 4 | 0.2% |
| Total | 127 | 5.4% |
Generic anchors appear on 9 sites; the other 13 sites with links have none from the list. E29 has 29 (29.0% of its 100 text links), E03 28 (17.6%), E26 23 (16.0%) and E17 17 (11.9%).
397 links share their label with another target
| Item | Count | Note |
|---|---|---|
| Labels pointing to ≥2 targets | 81 | Same text, different hrefs |
| Links covered by those labels | 397 | 17.0% of 2,339 |
| URLs with ≥2 labels | 283 | Same href, different texts |
| Distinct anchor texts | 1,341 | Deduped within each site |
A shared label is not automatically wrong: “产品中心” may point to a section page in the header and to another entry in the footer. But it removes the reader's ability to predict the destination from the text. The 283 multi-label URLs may likewise come from responsive menus or footer shorthand; this study records the pattern only.
Anchors are short: median 4 characters
- ≤2 characters: 165 (7.1%) on 20 sites; the most common is “+” (33), followed by one- and two-character navigation words.
- Median length: 4 characters, showing that many links still use labels such as “产品”, “首页” or “更多”.
- Navigation/CTA: 81 (3.5%), including “联系我们” 54, “首页” 22, “返回” 4 and “立即咨询” 1.
Short is not bad for menus. The problem is using “更多” or “详情” for body and product entry links, where the reader cannot predict what comes next.
What this means for acceptance
- A generic list is a lower-bound check. 5.4% is the explicitly matched share, not proof that the rest is descriptive.
- Check shared labels. 81 texts point to several URLs and cover 17.0% of text links; fix body and product entries first.
- Unify labels per URL. 283 URLs have several texts; at least align header, footer and listing pages.
- Separate menu labels from body anchors. “首页” and “返回” can be short; product and article entries should describe the destination.
- Read alongside accessible names. Links whose text is empty but whose image alt names them were counted in study 44; this study looks at text only.
Limits
- Fixed list: covers only the listed phrases, so it is a lower bound rather than a full quality score.
- Within-site dedupe: identical text on different sites is counted separately; shared labels are not compared across sites.
- Several labels may be design: responsive menus and header/footer shorthand create them; this study does not judge intent.
- Static HTML: no JavaScript, no CSS-hidden text.
- 25-site pilot: no industry claim.
Data and reproduction
Anonymised data: equipment-anchor-texts-2026.csv (25 rows × 11 columns: links, text links, generic, navigation/CTA, short texts, distinct texts, shared labels, multi-label URLs, median length). The repository keeps docs/research/2026-11-anchor-texts.csv.
Parser: scripts/anchor_text_probe.py; unit tests: scripts/test_anchor_text_probe.py (7 tests, no network). Run:
python -m unittest scripts.test_anchor_text_probe
Related: the homepage link accessible names pilot, the homepage link structure pilot, the article page link structure pilot, and the article head metadata pilot.
Sources and definitions
- Homepage cache: 25 homepages from study 4 (2026-10-31 to 11-01).
- Text rules: remove tags, unescape, collapse whitespace, trim edge punctuation, compare lowercased.
- Generic and navigation lists: as in the method table; phrases outside them are not counted as generic.
- Scan: 2026-11-03, zero network requests.