Doubao, DeepSeek, Kimi and ChatGPT now answer "which supplier should I choose?" with a written recommendation. If your company isn't in that answer, you don't exist in that market — no matter how good your website looks.
Practical, measurable work — llms.txt, structured data, comparison content, third-party listings, and monitoring that proves whether AI actually recommends you.
Traditional SEO chases positions in a results list. GEO chases inclusion in a generated answer. Different mechanics, different checklist, and a deeper moat once you are in the model's source pool.
Schema.org JSON-LD in the <head>: LocalBusiness, Service, FAQPage, Product. This is how a machine confirms your company is a real, addressable entity.
A plain-text file at your root that tells AI which pages are worth citing: company profile, product pages, cases, FAQ, contact. Cheap to build, and a visible differentiator versus competitors.
AI harvests lists, tables and cited sources. Specs and parameters must be rendered as real HTML <table> elements — not images, not canvas, not <div> soup.
Headings written as the exact questions buyers ask, answered with specifics — numbers, tolerances, certifications, lead times. No marketing filler.
Name, address and phone must be byte-identical on your site, maps, review platforms and business registries — or AI reads two different companies.
AI judges authority from outside signals: industry memberships, trade press, association listings, consistent recent activity.
| Dimension | Traditional SEO | GEO / GSO |
|---|---|---|
| Optimised for | Baidu, Google results pages | Doubao, DeepSeek, Tongyi, Kimi, ChatGPT |
| Ranking logic | Keyword density, backlinks, page authority | Authority, structured data, content quality, citation frequency |
| Result format | A list of links | A generated, complete answer |
| User behaviour | User clicks a link | User gets the answer — may never click |
| Conversion path | Search → click → landing page → enquiry | Question → answer → direct contact (may skip your site) |
| What wins | Page-level optimisation | Entity-level authority + machine readability |
Live audits of the high-purity fluoropolymer piping industry (semiconductor and photovoltaic grade) — the category where international incumbents dominate buyer awareness.
Twelve bilingual query sets run across high-purity piping categories. The finding was not "your site is ugly" — it was that your company had a name but no authoritative profile in the sources AI crawls.
| Audit dimension | Finding |
|---|---|
| llms.txt | Absent across the whole category |
| JSON-LD | Not detected on leading supplier sites |
| English FAQ schema | Missing — buying questions unanswered |
| Spec tables | Images / fragments, not HTML tables |
| Crawler policy | Some majors block AI crawlers outright |
| Blue ocean | High-purity UPVC / solar / ultrapure water |
| Semantics trap | Chinese query terms route to wrong product |
Market-report pages are the highest-cited sources for "who makes this" questions. Building a comparison-grade page with the same structure — list, table, cited sources — puts you into that source pool. None of this requires a large team; it requires knowing what AI actually reads.
Weekly checks of real buyer prompts against Doubao, DeepSeek, Tongyi and Kimi: "Beijing waterproofing company, which is good?", "semiconductor grade PFA tubing supplier China”. We log mentions, context and position — then fix the content that was not quoted.
You own the domain, the code and the content. I do the GEO layer on top of your existing site, or build the site if you do not have one.
For firms with an existing website that AI cannot read.
llms.txt + JSON-LD structured dataFor exporters and manufacturers who need to win English-language AI searches.
Keep the visibility you earned.
They are complementary, not alternatives. The technical overlap is large (quality content, structured data, mobile performance), but GEO adds machine readability and entity authority. A site can rank well on Baidu and still never be quoted by an AI assistant, because the assistant evaluates authority and structure differently. If you can only fund one, do both at the technical layer first.
No, for the foundation. Adding LocalBusiness JSON-LD, writing an llms.txt and creating an FAQ page can be done through your CMS or by hand — no code required. Advanced work (custom schema types, large-scale multi-language export sites, crawler diagnostics) does need technical help.
GEO is a compounding project, not a toggle. Expect the first movement in 4–8 weeks as your structured data and FAQ pages get picked up, and clearer citation growth over 2–3 months. Once you are in the model's source pool, the effect accumulates — that is where the moat is versus classic SEO.
It depends on whether you are a credible entity in the sources AI reads. In many niches — especially high-purity industrial supply, specialist trades, regional services — the international or national leaders have not done this work, and the smaller specialist wins. That is the specific opportunity I look for in the audit phase: where the incumbents are absent.
Yes. I am based in Beijing and work with clients across China, including export-facing manufacturers whose buyer journey runs through English-language AI tools. The first audit and the content work are done remotely; on-site visits are arranged for Beijing clients at no extra charge.
A single NAP fact sheet: registered company name, full address, phone, domain, core services (3–5), one-line positioning for customers and AI, and any certifications. If you already have a website, point me at it. I run the audit first, then we talk scope.
Tell me your company and the questions your buyers ask. I run a real audit of how your brand appears across Doubao, DeepSeek, Tongyi and Kimi, and send you the gap list. Free, no obligation.
Not an agency with a content farm, and not a tool that submits your site to a hundred directories. The work is: make your existing authority machine-readable, prove it with citations, and keep it alive.