Direct answer: Across 10 sites and 36 article pages there are 3,583 links: 3,000 internal (83.7%), 270 external (7.5%), 162 tel:, 116 javascript: and 35 mailto:. The 270 external links point to only 63 unique URLs (23.3%), a 76.7% repeat rate, on 15 external hosts. The median page has 4 external links, and 22 pages have at most 5.

Attributes: 26 external links carry nofollow (9.6%), 199 use target="_blank" (73.7%), and sponsored and ugc are both 0. Another 380 links have no visible text. Five URLs from one site return only a ~1.4 KB JavaScript shell; they are listed separately and excluded from the 36 pages.

36
usable pages
3,583
links
270
external
63
unique external URLs
15
external hosts
4
median external per page

Method: page level, not body only

The study reuses the article sample from study 28 (sitemaps of 12 sites, 2–5 URLs each, 46 total) and refetches it serially on 2026-11-03 between 08:40 and 08:41, with at least 0.6 seconds between requests. 41 URLs return 200; five of them are ~1.4 KB shells from one site, leaving 36 pages across 10 sites.

ItemRuleBoundary
Internal / externalCompared with the final page host, including relative pathsSubdomains count as internal; protocol-relative URLs are parsed
External dedupeEach distinct href counts onceThe same URL on many pages still counts each time
relnofollow / sponsored / ugc counted separatelyNo judgement of whether the combination is right
Link densityLinks ÷ visible characters × 1,000Visible text excludes script/style and strips tags
Shell pageResponse < 5 KBAn operational threshold, not a standard

Key definition: this is a page-level count including navigation, footer, share and filing links, not a body-only citation count. The same template repeats across pages and inflates raw external counts, so the next section looks at deduplication first.

Result: 83.7% internal, 270 external

TypeCountShare of 3,583Pages (36)
Internal3,00083.7%36
External2707.5%36
tel:1624.5%34
javascript:1163.2%35
mailto:351.0%25
In-page anchor00%0

All 36 pages have at least one external link; counts range from 1 to 61, with a median of 4, and 22 pages have at most 5. The two heaviest pages (61 and 33) repeat several recommendation modules in one template.

270 external links, only 63 unique URLs

MeasureCountNote
External links270Across 36 pages
Unique external URLs6323.3%; repeat rate 76.7%
External hosts15Includes one malformed http://tel:… address
Median external per page4Range 1–61
Median unique external per page3—

Within-site dedupe gives 74 URLs; merging across sites leaves 63, because some external links (such as a filing address) repeat on several sites. One external host appears 120 times, from a recommendation module repeated under every article.

SitePagesExternalUnique URLsRepeat
E035773159.7%
E06520385.0%
E09515380.0%
E145721875.0%
E1639366.7%
E191110%
E22535488.6%
E23534779.4%
E2613233.3%
E2814250.0%

Attributes: little nofollow, many new windows

AttributeAll linksExternal only
rel contains nofollow173 (4.8%)26 (9.6%)
target="_blank"295 (8.2%)199 (73.7%)
rel contains sponsored00
rel contains ugc00

73.7% of external links open in a new window, but only 9.6% carry nofollow; the study did not count rel="noopener", so it cannot judge safety. 21 pages have at least one nofollow link, concentrated in a few templates.

Empty anchors and link density

  • No visible text: 380 (10.6%), on 35 pages; mostly icons, images or share buttons. Whether they are truly unnamed needs the alt/aria merge from study 44.
  • Link density: a median of 54.6 links per 1,000 visible characters, up to 138.8. The high value comes from navigation and footer blocks repeated on every page, not from body text.

Five article URLs are only a JavaScript shell

Five URLs from one site return about 1.4 KB of HTML with almost no links or body text. They are not 404s, but for a crawler or a reader without JavaScript they are close to empty. Acceptance should confirm whether such URLs depend on client-side rendering.

What this means for acceptance

  1. Dedupe by URL before judging external links. The 270 links are only 63 addresses, a 76.7% repeat rate.
  2. Separate template links from citations. Page-level counts include navigation, footer and recommendation modules, not source citations.
  3. Check external attributes together. 73.7% open in a new window and only 9.6% carry nofollow; confirm noopener as well.
  4. Empty anchors still need a name. The 380 textless links should be merged with alt/aria as in study 44.
  5. List JavaScript shells separately. A 1.4 KB 200 response is not “content”; confirm server rendering or a fallback.

Limits

  • Page level, not body level: navigation, footer, share and external modules are included and are not source citations.
  • Single snapshot: one fetch on 2026-11-03 08:40–08:41, without a 12/24-hour recheck.
  • Regex parsing: no JavaScript; script-inserted links and share widgets are invisible.
  • Shell threshold: 5 KB is an operational definition; smaller does not always mean a shell, and the five were spot-checked.
  • Sample frame: 10 sites and 36 pages, a pilot, no industry claim.

Data and reproduction

Anonymised data: equipment-article-links-2026.csv (36 rows × 22 columns: link types, target/rel, external dedupe, visible characters, density, status and bytes). The repository keeps docs/research/2026-11-article-links.csv. Raw HTML and fetch records stay off-site and are not deployed.

Parser: scripts/article_link_probe.py; unit tests: scripts/test_article_link_probe.py (10 tests, no network). Run:

python -m unittest scripts.test_article_link_probe

Related: the homepage link structure pilot, the homepage link accessible names pilot, and the article visible date pilot, and the homepage anchor text pilot.

Sources and definitions

  1. Sample frame: article pages from study 28 (12 sites, 46 URLs); 41 returned 200 here.
  2. Fetch: 2026-11-03 08:40–08:41 Asia/Shanghai, serial, ≥0.6 s between requests.
  3. Counts: page-level href types, target/rel attributes, external URL dedupe and visible characters.
  4. Shell pages: five same-site URLs under 5 KB, listed separately outside the 36.
About the author

Lingyi Information Official (delivery team) works on requirements, proposals and acceptance for corporate websites, brand and smart-manufacturing projects. This article counts page-level link structure and attributes only; it does not judge target quality or whether citations are sufficient, and makes no claim about any site's rankings or traffic; the 10-site, 36-page sample is a pilot.

Dedupe external links first

Page-level counts include navigation and footers; after dedupe, 270 external links are 63 addresses and should be reported separately.

Request a recheck plan