Direct answer: in the visible text of 25 homepages, 13 show an ICP filing, 10 an address and six a phone. After fetching one or two contact/about pages, address rises to 14, phone to 12 and ICP to 14: four sites add an address, six add a phone and one adds ICP; 22 sites had an auditable inner page. Only four sites still carry ICP, address and phone on the homepage itself.

This study reads visible HTML text only, without running JavaScript or OCR, and does not query the filing registry; every result describes these 25 homepages in this one fetch.

25
homepages
13
visible ICP
14
ICP total
14
address total
12
phone total
22
auditable inner pages

Method: one fetch, visible text only

On 2026-11-08 between 07:42 and 07:43 we sent one GET to the final URL of each of the 25 homepages from study 4 (15-second timeout, serial, 0.6 s apart), removed script, style, comments and tags, and matched the visible text with regular expressions for the ICP filing number (province abbreviation + ICP备 + digits + 号), the public-security filing number, address, phone and business-licence/credit-code mentions. Only booleans and counts are published; filing numbers and site names are not.

ItemRuleBoundary
ICP filingVisible text contains “province + ICP备 + digits + 号”Number validity and entity match not checked
Public-security filingVisible “公网安备 + digits + 号”Domain coverage not checked
Address / phoneField labels “地址/Address”, “电话/Tel”Recency not checked
Licence / credit codeMentions “营业执照” or “统一社会信用代码”The code itself is not published

Result: 13 ICP filings, 3 public-security filings

SignalSitesShare of 25
ICP filing number1352.0%
Public-security filing312.0%
Address1040.0%
Phone624.0%
Business licence / credit code728.0%

All three public-security filings appear alongside ICP, but only three of the 13 ICP sites show one. ICP behaves like a launch requirement while the security filing covers fewer sites; neither proves the entity block is complete.

Combinations: only four sites show ICP, address and phone together

CombinationSitesNote
ICP + address + phone4Entity, contact and filing all visible
ICP + address (no phone)1—
ICP only (no address, no phone)8Filing number without visible contact
No ICP, but address or phone5Contact info without a visible filing number
None of the three7No ICP, address or phone on the homepage

The eight ICP-only sites show that the filing number and contact details are separate systems: the number often sits in the footer while the address and phone live on a contact page. When all three appear on the homepage, a buyer can verify the entity with no extra click; with only the filing number, one more page is needed.

Contact pages add four addresses and six phones

On 2026-11-08 between 09:40 and 09:42 we discovered same-origin contact/about pages from the 25 homepages (“contact/about/company/联系/关于”), fetched at most two per site (22 sites had one, three had no reachable candidate), re-counted address and phone in visible text and searched the raw HTML for ICP.

ScopeICPAddressPhone
Homepage visible text13106
Homepage + contact/about page141412
Gain from inner pages+1+4+6

Inner pages mainly add contact details: four sites add an address, six add a phone, and ICP adds only one. The filing number usually sits in a site-wide footer shared by the homepage and inner pages, while address and phone more often appear only on the contact page.

Of the eight “ICP-only” homepages, seven had an auditable inner page: three added an address and four added a phone (E08, E11 and E28 added both; E12 added only a phone). “Only a filing number on the homepage” therefore does not mean the site has no address or phone — only that they are not on the homepage.

Audit: of 12 sites without visible ICP, one has it in source or inner pages

We re-checked the 12 homepages whose visible text had no ICP: the homepage source was fetched again (including script/style), same-origin links matching “contact/about/company/联系/关于” were collected, up to three were fetched, and the raw HTML was searched for a filing number; when no candidate link existed, /contact.html and /about.html were tried. The run was on 2026-11-08 between 08:40 and 08:41.

Result: one site (E23) has ICP in its homepage source and on all three inner pages — the number exists but not in visible text; eight sites still show none in the source or in two to three inner pages; three sites could not be audited (E13 returned 502, and E16 and E20 had no reachable candidate page).

Audit resultSitesNote
Found in source or inner page1E23; homepage source and three inner pages
Still not found8Homepage source plus two or three inner pages
Could not audit3E13 returned 502; E16 and E20 had no candidate page

The two measures must be reported separately: 13 of 25 in visible text, 14 of 25 in HTML; the visible-text rule missed one of 12. The combination finding (only four sites carry ICP, address and phone together) is unchanged, but a homepage visibility check cannot replace a source check.

What this means for acceptance

  1. Check the filing number on the homepage. 13 of 25 show it; the other 12 may keep it in an image, script or inner page, so manual confirmation is needed before calling it missing.
  2. A filing number is not a complete entity block. Eight sites show no visible address or phone.
  3. Contact details shape the enquiry path. Six homepages show a phone and ten an address; read alongside the homepage enquiry-path pilot, the homepage is only one entry point.
  4. Licence and credit codes remain rare. Seven mention them and none shows a full 18-character code; verification should point to an about/qualification page.
  5. Write “visible” into acceptance. Keeping the filing number, address and phone in the footer or contact block saves the buyer a click; see the website development service checklist.
  6. Record visible and source separately. One of 12 “missing” ICP numbers was present in source; acceptance should state whether it checks visible text or HTML.
  7. Include the contact page in entity acceptance. A filing number on the homepage is not enough; the contact/about page should add address and phone — four sites added an address and six a phone.

Limits

  • Visible text only: no JavaScript execution and no OCR; a number rendered by an image, canvas or script is recorded as absent.
  • No registry query: validity, entity match, domain coverage and filing status are not verified.
  • One fetch: 2026-11-08 07:42–07:43; redesigns or WAF responses can change the result.
  • Regex scope: address and phone require field labels and do not recognise bare numbers or place names.
  • 25-site pilot: no industry claim and no compliance conclusion.
  • Limited audit: homepage source plus up to three same-origin inner pages; no OCR, no login and no registry query; three sites could not be audited.
  • Only two inner pages: discovery uses link text, so JavaScript menus or unusual labels can be missed; three sites had no reachable candidate and are not counted as gains.

Data and reproduction

Anonymised data: equipment-homepage-entity-signals-2026.csv (25 rows × 10 columns, homepage visible text). Audit data: equipment-homepage-entity-audit-2026.csv (12 rows × 7 columns, source/inner audit). Inner-page comparison: equipment-homepage-entity-inner-2026.csv (25 rows × 13 columns: homepage and one or two contact pages with ICP/address/phone and gain flags). The repository keeps the matching CSVs under docs/research/.

Parser: scripts/entity_signal_probe.py and scripts/entity_signal_audit.py; unit tests: scripts/test_content_entity_signals.py (11) and scripts/test_content_entity_audit.py (9, both offline). Run:

python -m unittest scripts.test_content_entity_signals scripts.test_content_entity_audit

Related: the homepage enquiry-path pilot, the homepage HTTPS certificate and TLS pilot, and the homepage canonical and scheme-redirect pilot, and the homepage English-entry pilot.

Sources and definitions

  1. Sample: the 25 homepages from study 4 (same sample_id).
  2. Fetch: 2026-11-08 07:42–07:43 Asia/Shanghai, serial, 0.6 s apart, 15-second timeout.
  3. Processing: script/style/comments/tags removed; visible text matched only; no JS execution and no OCR.
  4. Publication: booleans and counts only; filing numbers, phones and addresses are not published.
  5. Audit: 2026-11-08 08:40–08:41; homepage source plus up to three same-origin contact/about pages, falling back to /contact.html and /about.html.
  6. Inner-page comparison: 2026-11-08 09:40–09:42; up to two same-origin contact/about pages, visible text for address/phone and raw HTML for ICP.

Frequently asked questions

Do numbers inside images or JavaScript count?

No. This study reads visible HTML text only, without running JavaScript or OCR; a filing number rendered by an image or script is recorded as absent and needs manual review.

Does a missing filing number mean non-compliance?

Not necessarily. The study only records whether a filing number appears in visible homepage text; it does not query the filing registry or judge registration or compliance.

Must a public-security filing appear with ICP?

No. All three sites with a public-security filing also show ICP here, but that is an observation across these 25 sites, not a rule.

Why count address and phone separately?

They help B2B enquiries and entity checks. Ten homepages show an address and six show a phone, but only four carry ICP, address and phone together.

Do 25 sites represent the industry?

No. This is the same 25 equipment-company homepages from study 4; it describes only these pages.

Can the data be re-checked?

Yes. The homepage CSV has 25 rows by 10 columns, the audit 12 by 7 and the inner-page comparison 25 by 13; the two parser modules have 20 offline unit tests.

Why do the visible-text and source numbers differ?

The filing number on E23 exists in the homepage source but not in visible text, likely inside a script or a hidden element; after the audit the source-based count is 14 of 25 while the visible-text count stays 13 of 25.

Why does ICP improve little while address and phone improve more?

The filing number usually sits in a site-wide footer shared by the homepage and inner pages, while address and phone more often appear only on the contact page. ICP improved by one site, address by four and phone by six.

About the author

Lingyi Information Official (delivery team) works on requirements, proposals and acceptance for corporate websites, brand and smart-manufacturing projects. This article records filing and contact signals in visible homepage text from one fetch; it does not query the filing registry, judge compliance or reach conclusions about any site; the 25-site sample is a pilot.

Missing canonical is a common homepage gap

25 sites: 9 canonical, 16 missing; 8 og:url.

Request a recheck plan