Direct answer: among 15 Changzhou equipment companies with an analysable homepage, the first static product link led to 13 product entry pages; two could not be identified (P04's link returned to the homepage/project page, P14's homepage is a JS shell). Of the 13, 6 have no H1, 6 have exactly one and one has three; only one keeps JSON-LD (P09), while two of the three homepages with JSON-LD lose it on the product page. Two pages have tables, four have an inquiry form; body text runs 559–15,338 characters (median 5,568), HTML 7.9–232.4 KB (median 93.8).
More importantly, P09 has JSON-LD but its product path is disallowed to all five AI bots (GPTBot, OAI-SearchBot, PerplexityBot, Bytespider, ClaudeBot). Good structure is useless if the crawler is refused.
Method: take the first product link from the homepage
The parent frame is the 30 Changzhou companies on the 2024 Su–Xi–Chang first-set equipment list; the sub-frame is the first 15 in list order with an analysable homepage. For each, we took the first internal link whose href or text contains product, the Chinese word for product, or chanpin/prolist; if none existed we recorded no_product_link and did not guess a URL. Product-path robots rules were checked with RFC 9309 longest-match first; each page received one GET, at least 5 s between sites, 20 s timeout, with no JS, login or form submission.
| Item | Rule | Boundary |
|---|---|---|
| Product entry | First internal link with product / Chinese product / chanpin / prolist | Not necessarily the most important product |
| H1 | Number of h1 tags | Copy quality not judged |
| JSON-LD | Any valid JSON-LD block | Only presence and type count |
| Inquiry form | A form with message/quote signals | Not submitted or backend-tested |
| robots | Five AI bots' path rules | State at capture time |
Result: missing H1, a JSON-LD gap, thin pages
| Metric | Result (13 pages) |
|---|---|
| H1 = 0 / = 1 / = 3 | 6 / 6 / 1 |
| With JSON-LD | 1 (P09) |
| With tables | 2 pages (P06 6 tables, P15 1) |
| With an inquiry form | 4 pages (P02, P08, P13, P15) |
| With a FAQ signal | 1 page |
| mailto / tel links | 5 / 8 pages |
| meta description / viewport | 12 / 12 pages |
| hreflang | 0 pages |
| Body min / median / max | 559 / 5,568 / 15,338 characters |
| HTML min / median / max | 7.9 / 93.8 / 232.4 KB |
| Median TTFB | 303 ms |
Three pages have under 2,000 characters (P01 559, P09 1,088, P12 1,163) and four under 3,000; at the other end P06 has 15,338 characters and six tables. A product entry page can be a thin list page or a thick detail page; they are not one type.
Page-by-page (anonymous P01–P15)
| Page | Result | H1 | JSON-LD | Tables | Inquiry | Text | HTML KB | AI bots allow/deny |
|---|---|---|---|---|---|---|---|---|
| P01 | 200 | 0 | 0 | 0 | 0 | 559 | 7.9 | 5/0 |
| P02 | 200 | 0 | 0 | 0 | 1 | 10,040 | 232.4 | 5/0 |
| P03 | 200 | 0 | 0 | 0 | 0 | 5,692 | 159.3 | 5/0 |
| P04 | no product link | — | — | — | — | — | — | — |
| P05 | 200 | 1 | 0 | 0 | 0 | 5,683 | 158.0 | 5/0 |
| P06 | 200 | 1 | 0 | 6 | 0 | 15,338 | 100.7 | 0/0 (robots 404) |
| P07 | 200 | 0 | 0 | 0 | 0 | 4,218 | 93.8 | 5/0 |
| P08 | 200 | 3 | 0 | 0 | 1 | 5,568 | 29.5 | 5/0 |
| P09 | 200 | 0 | 1 | 0 | 0 | 1,088 | 41.1 | 0/5 |
| P10 | 200 | 1 | 0 | 0 | 0 | 2,030 | 14.8 | 0/0 (robots 404) |
| P11 | 200 | 1 | 0 | 0 | 0 | 5,485 | 94.9 | 5/0 |
| P12 | 200 | 0 | 0 | 0 | 0 | 1,163 | 13.4 | 0/0 (robots 404) |
| P13 | 200 | 1 | 0 | 0 | 1 | 12,060 | 170.4 | 5/0 |
| P14 | no product link | — | — | — | — | — | — | — |
| P15 | 200 | 1 | 0 | 1 | 1 | 5,746 | 40.2 | 5/0 |
The homepage conclusion does not carry over
H1 counts differ from the homepage on 3 of 13 pages (P02 1→0, P08 10→3, P13 2→1). Three homepages had JSON-LD (P07, P09, P11); only P09 keeps it on the product page, so P07 and P11 lose structured data. A unique homepage H1 does not imply a product-page H1, and homepage JSON-LD does not imply product-path JSON-LD.
| Comparison | Result |
|---|---|
| H1 count differs from homepage | 3/13 (P02, P08, P13) |
| Homepages with JSON-LD | 3 pages (P07, P09, P11) |
| Product pages still with JSON-LD | 1 page (P09) |
| Product pages losing JSON-LD | 2 pages (P07, P11) |
| Product path disallowed to all five AI bots | 1 page (P09) |
P09: structure present, crawlers refused
P09 is the only product page that keeps JSON-LD, but its robots.txt returns disallowed for GPTBot, OAI-SearchBot, PerplexityBot, Bytespider and ClaudeBot. This is policy over structure: even with CollectionPage/BreadcrumbList/WebSite markup, all five AI crawlers are explicitly refused. Acceptance cannot stop at "the page has JSON-LD"; it must also check the product path's robots rules.
Two sites with no identified product entry
P04's first product link returned to the homepage/project page, and P14's homepage is a JS shell with no product link in the static HTML. Both are recorded as no_product_link, not "no product page". That is itself an AI-readability issue: the static layer offers no product entry for a crawler to follow from the homepage.
What this means for acceptance
- Audit product pages separately. Homepage H1, JSON-LD and body length do not represent the product entry page.
- Add a unique H1 first. Six of 13 have none and one has three; a product page needs one H1 that names the product or category.
- Carry JSON-LD to product pages. At least Product/Offer and BreadcrumbList; losing it between homepage and product page is common.
- Put an inquiry path on the product page. Only 4/13 have a form; the rest can add a form, phone, email or WeChat.
- Check product-path robots. P09 has JSON-LD but all five AI bots are refused; policy beats structure.
- Deepen thin pages. Three pages have under 2,000 characters; a product entry page needs specs, use cases and delivery/service information.
Limits
- 13-page pilot: two of 15 sites have no product link; not the full 30-company sample and not an industry estimate.
- First product entry: not necessarily the most important product; PDF catalogues, JS menus and image hotspots are excluded.
- One page per site: not all product pages on the site.
- Single time window: captured 2026-11-01; no recheck; robots rules may change.
- No JS: product links behind JS shells or dynamic menus may be missed;
no_product_linkis not proof that no product page exists. - No backend test: inquiry forms are recorded as present, not submitted or delivery-tested.
Data and reproduction
Anonymised data: equipment-product-entry-pilot-2026.csv (15 rows × 24 columns: result, robots status, HTTP, HTML KB, TTFB, text, H1, JSON-LD/type count, tables, FAQ, forms, inquiry, mailto/tel, hreflang, meta, viewport, AI-bot allow/deny, homepage H1/JSON-LD). In-repo copy: docs/research/2026-11-equipment-product-entry-pilot.csv.
Capture: scripts/site_audit_probe.py; product-path robots use the rules in js/robots-matcher.js (with scripts/test_robots_matcher.js). Site and product URLs are not published.
Related: the 30-site AI-crawler robots scan, the homepage semantic-structure pilot, and the homepage inquiry-path pilot.
Sources and definitions
- Parent frame: 2024 Su–Xi–Chang first-set equipment list, 30 Changzhou companies; sub-frame: the first 15 with an analysable homepage.
- Product entry: first internal link with product / Chinese product / chanpin / prolist; otherwise
no_product_link. - Capture: one GET per page, ≥5 s between sites, 20 s timeout, no JS/login/form submission; robots by RFC 9309 longest match.
- Measured 2026-11-01; sites anonymised, product URLs not published.
Related: the product entry image-delivery pilot.