Direct answer: We took the first 15 companies with an analyzable homepage from the 2024 Su-Xi-Chang first-set list and measured one product entry page each. 13 were analyzable and 2 had no valid product link under the fixed rule. Of the 13 pages, 6 have no H1, 6 have one and 1 has three; only 1 has JSON-LD; 2 have tables and 4 have inquiry forms; the median static text is 5,568 characters. Of the three homepages with JSON-LD, two did not carry it to the product page.
This is a pilot, not a benchmark and not a ranking. The product entry page is the first internal link in the static homepage HTML whose URL or text contains product, 产品, chanpin or prolist. It may be a product center, list or detail page and is not necessarily the most important product. With 15 companies and 13 pages, the sample is small; it describes these pages only.
A readable homepage does not mean a readable product page.
Structured data is not inherited from the homepage.
robots comes before structure: a blocked crawler cannot read a good product page.
Sample and definition
The parent frame is 30 Changzhou companies from the 2024 Su-Xi-Chang first-set major equipment list. We take the first 15 with an analyzable homepage, in list order, and measure one product entry page per company.
The rule is fixed: take the first internal link in the static homepage HTML whose href or text contains product, 产品, chanpin or prolist; exclude the homepage itself, project case pages and external links. If none is found, record no_product_link; we do not guess URLs or run JavaScript.
Method: one fetch, robots path checked first
For each product URL we checked robots.txt against the five AI crawlers and our own user agent using the RFC 9309 longest match, and fetched only if allowed. One GET per page, at least five seconds between sites, 20-second timeout, no JavaScript, no login, no form submission.
Metrics follow the homepage audit: status, HTML size, visible text characters, H1 count, JSON-LD types, tables, FAQ signal, inquiry form, mailto/tel, hreflang, meta description and viewport. We compare H1 and JSON-LD with the same site's homepage.
Result 1: 13 analyzable pages out of 15
| Result | Count | Rule |
|---|---|---|
| Analyzable product entry page | 13 | GET 200, parseable static HTML |
| No valid product link | 2 | First link returned to the homepage/project page, or the homepage was a JS shell; not proof that no product page exists |
| robots blocks five AI bots | 1 | Product path disallowed; our own agent allowed, so the fetch was for structural observation only |
"No valid product link" means not identified under the fixed rule, not that the company has no product page.
Result 2: structure against the homepage
| Metric (n=13) | Result | Note |
|---|---|---|
| H1 = 0 | 6 | No H1 on the product page |
| H1 = 1 | 6 | Unique H1 |
| H1 = 3 | 1 | Multiple H1s |
| JSON-LD present | 1 | E-commerce template output, not product-spec structured data |
| Tables present | 2 | One page with 6, one with 1 |
| Inquiry form present | 4 | Recognizable in static HTML |
| Median text | 5,568 chars | Range 559–15,338; list pages and detail pages differ |
| H1 differs from homepage | 3 | Count change only, not a quality judgement |
H1, JSON-LD, tables and forms are counted from static HTML; JavaScript injection is undercounted.
The more notable gap is JSON-LD: of these 13 sites, 3 had it on the homepage and only 1 on the product page; 2 did not carry it over. For tools that rely on structured data to understand what a product is and what its specs are, the product page is the empty one.
Result 3: robots before structure
One of the 13 product paths disallowed GPTBot, OAI-SearchBot, PerplexityBot, Bytespider and ClaudeBot; three sites had no robots.txt (absent); the rest were allowed. A blocked crawler will not read the page, however good the H1 and tables are.
That is why "fixing the homepage" is not the same as "making product pages citable": access policy, page structure and content depth are three layers, and any missing layer thins the product page from a machine's point of view.
Four checks to run now
- Is the product entry in static HTML? Menus, image maps and JS dropdowns can hide it; keep at least one plain link.
- Does the product page have one H1? Product name plus model/series is usually enough; do not replace it with an image.
- Does structured data carry over? Organization on the homepage does not cover product pages; each needs its own Product/Service or FAQ that matches visible content.
- Does robots allow the product path? Check more than the homepage; product, solution and download directories can be blocked separately.
For a self-check, use the 20-check AI readability audit. Homepage-level policy is in the 30-company equipment scan, and the SEO and GEO guide covers crawling, structure and measurement.
Limits
- Small sample: 15 companies, 13 product entry pages, below 30; pilot only.
- Entry not representative: the first product link is not the most important product and does not cover all product pages.
- Static layer: no JavaScript; the JS-shell homepage therefore had no product link, a method limitation.
- One fetch: no redesign, cache or regional variation; TTFB is only a proxy.
- Not linked to traffic or citations: indexing, AI mentions, clicks and inquiries were not measured.
Data and reproduction
The anonymized CSV (15 rows, sample IDs P01–P15, with the matching homepage sample ID, H1, JSON-LD, tables, forms and homepage comparison) is at equipment-product-page-pilot-2026.csv. Raw HTML is not published.
Recheck: P06 and P09 were re-parsed from stored HTML with the same parser; H1, JSON-LD, tables and text matched the first run. Other samples keep raw responses and timestamps. Empty cells mean "no analyzable page", not zero.
Sources and definitions
- 2024 Su-Xi-Chang first-set major equipment list notice (Changzhou, Suzhou and Wuxi industry bureaus, 2024-12-16): notice.
- RFC 9309, Robots Exclusion Protocol: rfc-editor.org.
- schema.org Organization / Product (reference for the structural signals): schema.org.