Background
Before the rebuild we diagnosed our own site. The problems were specific:
- Two domains served the same content with no redirect between them, leaving engines to pick one of two identical copies
- Our English name collides with a same-named company registered in the UK (company number 13279332), which makes the two entities easy to merge
- The home page carried under 3,400 characters of body text and not a single table — there was almost nothing an engine could quote whole
What we did
- Rebuilt the information structure: entity facts, services, cases and insights each get their own URL and structured data
- Wrote the disambiguation facts into crawlable body text: legal name, ICP filing numbers, registered address, former domain
- Published llms.txt / llms-full.txt as a machine-readable summary of the site
- Gave Chinese and English separate URLs, with paired hreflang and canonical declarations
- Scheduled three baselines (B0 / B1 / Round 1) using the same prompts and the same engines
Technical approach
| Item | Approach |
|---|---|
| Site generation | Eleventy 3 + Nunjucks, static output |
| Languages | Chinese on root, English on /en/ URLs |
| Structured data | Organization / WebSite / Service / FAQPage / CollectionPage |
| Machine-readable entry | llms.txt, llms-full.txt, sitemap.xml |
| Build-time checks | A script verifies hreflang, canonical, JSON-LD, image dimensions and dead links |
Results
| Metric | Value |
|---|---|
| Duplicate-domain redirect | (to be filled after Round 1) |
| Entity identified correctly | (to be filled after Round 1) |
| Buyer questions where we are cited | (to be filled after Round 1) |
| Engine coverage | (to be filled after Round 1) |
Authorization
Our own site: the diagnosis and the rebuild are fully public. The baseline numbers are filled in once Round 1 is measured.