Doubao, DeepSeek, Tongyi and Kimi now answer questions like "Which Beijing cleaning company is reliable?" or "Who makes the best semiconductor-grade PFA tubing?" — and name specific suppliers. If your company isn't in that answer, you don't exist in that market, no matter how polished your website is.
What we actually do: llms.txt, structured data, citable comparison content, third-party platform listing, and live monitoring of whether AI really recommends you.
SEO fights for position on a search-results page; GEO fights to be written into the answer. Different logic, different checklist — but once you're in, the moat is deeper than classic SEO, because AI citations compound.
Schema.org JSON-LD in the <head>: LocalBusiness, Service, FAQPage, Product. This is how a machine confirms you are a real, verifiable business.
A plain-text file at your site root that tells AI which pages are worth citing: about, products, case studies, FAQ, contact. Very low cost, but a clear differentiator when peers haven't done it.
AI loves "lists + tables + sourced" content. Specs and parameters must render as real HTML <table> — not images, not canvas, not divs pretending to be a table.
Titles are the exact questions buyers ask; answers are specific — numbers, tolerances, certifications, lead times. No marketing fluff.
Name, address, phone must match exactly across your site, maps, review platforms and business registries — otherwise AI reads you as two different companies.
AI judges authority from outside signals: industry-association status, trade-media coverage, third-party收录, steady publishing cadence.
| Dimension | Traditional SEO | GEO / GSO |
|---|---|---|
| What you optimise for | Baidu / Google results pages | Doubao, DeepSeek, Tongyi, Kimi, ChatGPT |
| Ranking logic | Keyword density, backlinks, page authority | Authority, structured data, content quality, citation frequency |
| How it's shown | A list of links | A full natural-language answer |
| User behaviour | Clicks a link on purpose | Gets the answer directly, may never click |
| Conversion path | Search → click → landing page → enquiry | Ask → answer → contact directly (may skip the site) |
| What decides it | Page-level optimisation | Entity-level authority + machine readability |
A bilingual audit of the high-purity fluoropolymer tubing industry (semiconductor / photovoltaic grade) — precisely the category where international majors monopolise buyer perception.
We ran 12 bilingual query sets. The conclusion isn't "your site is ugly" — it's that your company appears in AI's sources as a name only, with no authority profile.
| Audit dimension | Finding |
|---|---|
| llms.txt | Missing across all categories |
| JSON-LD | Not detected on major supplier sites |
| English FAQ structure | Missing — no one answers buyer questions |
| Spec tables | Images / fragments, not HTML tables |
| Crawl strategy | Some majors block AI crawlers outright |
| Blue-ocean gap | High-purity UPVC / PV / ultrapure water |
| Semantic trap | Chinese query terms point to the wrong category |
In the category we audited, not one supplier ships llms.txt, several international majors block AI crawlers outright, and none of them publishes spec tables an engine can lift. Their advantage was built for search engines that no longer decide who gets recommended.
| AI citations by source | Jul 2025 | Mar 2026 |
|---|---|---|
| Came from a top-10 Google ranking | 76.1% | 37.9% |
| Came from a page outside Google's top 100 | — | 31.0% |
Ahrefs, March 2026: 863,000 keyword SERPs and 4 million AI Overview citations. Being ranked high used to be the ticket to being cited. It no longer is — which means a young site can now be quoted inside the answer without ever fighting the incumbents for page one.
Chaozhou Sanbo Ceramic has been making tableware since 2008 and runs 500+ staff. Semrush puts its Authority Score at 14 and its organic traffic at 123 visits a month. It is still cited by 11 of its own pages across AI search — 9 of them inside Google AI Mode.
It did not buy its way there. Every product page is written as a spec sheet — material, firing temperature, weight, exact dimensions, FDA / LFGB / SGS certification, shipping terms, a five-question FAQ — because that is the format an engine can quote. Authority Score 14, cited anyway.
The deliverable is the same either way. What changes is how many competitors are already sitting in the source pool when you arrive. Start now and you are one of a handful of citable sources. Start in a year and you are one of dozens — and you will be paying a premium to be noticed at all.
Every audit we run finds fewer empty slots than the last one. That gap closing is the whole clock.
| When you start | What it costs |
|---|---|
| Early (now) | A landing page, llms.txt, one real spec table |
| In 6–12 months | Same work — now as catch-up against occupied slots |
| After the category fills | Buying attention inside a source pool that is already full |
"Who makes this?" questions — AI grabs market-report pages and comparison tables. Build the same structured content pages — lists, tables, with sources — and you're in that source pool. No big team needed; you just need to know what AI actually reads.
Every week, ask Doubao, DeepSeek, Tongyi, Kimi the real buyer phrasing: "Which Beijing waterproofing company is good?", "semiconductor grade PFA tubing supplier China". Record whether you're mentioned, in what context, at what rank — then fill the un-cited gaps.
Domain, code and content are yours. I add a GEO layer to your existing site, or build one if you don't have one.
For companies with a website AI can't read.
llms.txt + JSON-LD structured dataFor export-facing factories chasing English AI search traffic.
Protect the visibility you've already earned.
They're complementary, not substitutes. The technical layer overlaps heavily (quality content, structured data, mobile experience), but GEO additionally requires machine readability and entity authority. A site can rank #1 on Baidu yet never be cited by AI — because assistants judge authority differently. If budget only allows one, do the shared technical layer first.
Not for the basics. Adding LocalBusiness structured data, writing llms.txt, building an FAQ page — all doable with a CMS plugin or by hand, no coding. Advanced work (custom schema types, large-scale multilingual export sites, crawler diagnostics) does need real technical skill.
GEO is a cumulative effort, not a one-click switch. After structured data and FAQ pages go live, you usually see the first movement in 4–8 weeks; citation growth becomes clearer after 2–3 months. Once you're in the model's source pool, the effect keeps compounding — exactly why the moat is deeper than classic SEO.
It depends on whether you're a trusted entity in AI's sources. In many niches — high-purity industrial tubing, specialised local services, regional providers — international or national majors haven't done this, so the small and focused can win. The audit specifically hunts these "major-absent gaps"; that's where budget produces real differentiation.
Yes. I'm in Beijing but serve clients nationwide, including export-facing factories using English AI tools. The upfront audit and content work are remote; Beijing clients can have me on-site at no extra charge.
One NAP fact sheet: full legal business name, detailed address, phone, domain, core services (3–5), a one-line positioning statement (for clients and AI), existing certifications. If you have a site, give me the address. I audit first, then we talk scope.
Tell me your company and the questions buyers ask. I'll run Doubao, DeepSeek, Tongyi and Kimi with real queries and send you the gap list. Free, no hard sell.
Not a bulk-content spam agency, and not a machine that submits your site to hundreds of directories. Three things only: turn the authority you already have into something machines can read, prove it with citations, and keep the momentum going.