Direct answer: Across the same 25 homepages, static navigation shows product 22, about 20, news 22, contact 21; solution 13, service 14, case 13, jobs 11; downloads only 4 and search 8 (7 with a real input). A language entry appears on 16, but only 9 point at an English URL (/en or an en. subdomain); the other 7 are JS switches or same-page text. hreflang in the head appears on 1.
This round adds the navigation layer. Earlier ones covered robots, sitemap, contact, product pages, performance and entry protocol; none looked at where the homepage sends people. Data is again the cached HTML of the 25 homepages — zero requests.
Method
Parse anchor text and hrefs from the 25 cached homepages: a keyword match counts as an entry. No JavaScript, no semantic classification. Language entry has two levels: text matching English/英文/language switcher counts as "entry"; an href pointing at /en, en-us, an en. subdomain or ?lang=en counts as "English URL".
| Entry | Hits | Share of 25 |
|---|---|---|
| Product | 22 | 88% |
| About | 20 | 80% |
| News | 22 | 88% |
| Contact | 21 | 84% |
| Solution | 13 | 52% |
| Service | 14 | 56% |
| Case | 13 | 52% |
| Jobs | 11 | 44% |
| Downloads | 4 | 16% |
| Mentions search | 8 | 32% |
| Has a search input | 7 | 28% |
Result 1: the basics are there, depth is not
Product, about, news and contact all clear 80%, which makes "company—product—news—contact" the shared skeleton. The spread appears in solution, service and case (about half) and in downloads and search (under a third).
For a manufacturer, downloads (samples, specs, selection guides) are often what a buyer wants most. At 16% homepage coverage, most sites bury documents inside product pages or do not publish them at all.
Result 2: the three layers of "multilingual"
| Layer | Hits | Note |
|---|---|---|
| Language entry in navigation | 16 / 25 | English / 英文 / language switcher |
| Links to an English URL | 9 / 25 | Mostly an en. subdomain, some /en |
| hreflang in the head | 1 / 25 | The only site with both an entry and hreflang |
16 → 9 → 1 is the number to remember. The three signals have very different strength: navigation text is the weakest, a clickable address is stronger, hreflang is the strongest. Eight sites have both weaker layers and no hreflang — a visitor can reach the English site, but nothing tells a search engine the two pages are the same content in different languages.
Result 3: search and static reachability
Eight homepages mention search and seven contain an input, so those who wrote it mostly built it; the gap is coverage. Separately, three homepages have no parsable <a> at all (one has 11 characters of text), matching the earlier finding of four homepages without a contact entry. Their navigation lives in JavaScript or is missing.
Anchor counts range from 0 to 204 with a median of 102. That mixes header, footer, sidebar and body links; it is not the number of main menu items, only a sign of very different template sizes.
Five actionable conclusions
- Add a downloads entry. At 16%, it does not match what buyers look for.
- Search: finish it or skip it. Eight mention it and seven have an input; the real gap is coverage, not half-built widgets.
- Make the language entry clickable. Seven are JS-only or text-only; at least link to
/enor anen.subdomain. - If an English site exists, add hreflang. Sixteen entries, one hreflang; eight sites are missing the signal.
- Do not hide navigation in JS. Three homepages have no static links, invisible to any crawler that does not run JS.
Limits
- Static layer: JS menus, mobile drawers and floating navigation are out of scope.
- Keyword method: synonyms match differently; results may undercount.
- Anchor count is not IA quality.
- No click-through: an entry does not prove the target is reachable.
- 25 sites, a pilot.
Data and recalculation
Anonymised data: equipment-nav-links-2026.csv (25 rows, 15 fields). The repository keeps docs/research/2026-11-nav-links.csv. Parser: scripts/nav_links_probe.py; tests: scripts/test_nav_links_probe.py (13 tests, no network).
python -m unittest scripts.test_nav_links_probe
Related: the 30-site robots scan (hreflang 1/25), the contact-path pilot (four homepages without an entry) and the hreflang checker.
Sources and definitions
- Sample and cache: the 25 analysable homepages from the earlier scan, captured 2026-10-31 to 11-01; zero requests this round.
- hreflang count comes from
2026-11-equipment-robots-scan.csv. - Entries are keyword matches on anchor text or href, not a semantic classification.