This is the second episode of my podcast. I talked with YG, founder of CapGo AI, for nearly an hour. He left Citi's investment banking division to build an AI tool that generates web pages at scale to help companies get discovered through ChatGPT and Google. He's worked with several top Chinese AI companies. After this conversation, I felt like I understood everything worth knowing about GEO.
What GEO actually is
One sentence: getting ChatGPT and Google to recommend your product.
When you ask ChatGPT "what's the best translation tool," and your product shows up in the answer — that's GEO at work.
How? The core logic is covering every possible search intent. If you're a translation tool, users might not search "translation tool" — they might search "Chinese to English," "French to German," "Italian to Japanese." Each query gets its own landing page. The goal is to be discoverable no matter what someone types.
You're not competing for 10 keywords one by one. You're generating hundreds of pages to achieve total coverage.
On the relationship between GEO and AEO: they're essentially the same thing. Many traditional SEO practitioners consider GEO a buzzword coined by VCs and founders, since it's still search engine optimization — just with a different display format.
Why general-purpose AI can't do this well
People ask: Claude can generate web pages too, so why use a specialized tool?
"Being able to run and being able to outrun Liu Xiang are two different things."
General AI can do it, but doing it well is another matter. Vertical tools have two advantages: first, automation efficiency — the system already knows the end-to-end workflow without needing to think; second, deep industry know-how about what page structures get indexed better by Google and ChatGPT.
And you don't need to rank first. Build 200 pages, each ranking fifth, and the aggregate traffic is massive. The core isn't chasing individual rankings — it's total coverage.
Page quality depends on data
The biggest risk of mass-generating pages? Low quality, zero traffic from Google.
The solution is data-driven generation. Request massive amounts of internal assets from clients — user-generated images, music, videos, product docs, analysis reports. Use this original content to generate pages. Google and ChatGPT will treat them as high-quality originals, and rankings follow naturally.
The more internal material a client has, the higher the quality. This is why AI companies are ideal clients — they have an enormous number of use cases, naturally suited for mass generation. An AI video generator can create pages for "AI generate rabbit," "AI generate 3D," "AI generate ancient architecture" — the search intents are nearly infinite.
China vs. overseas GEO: completely different games
This was one of the most valuable parts of the conversation.
"If someone guarantees traffic, they're probably a scam."
Chinese clients especially want traffic guarantees, but overseas, any SEO agency claiming guaranteed traffic is widely considered fraudulent. Too many factors affect SEO — Google's algorithm, page quality, user dwell time — many of which have nothing to do with the pages you generate.
China's gray-area practices stem from the ecosystem itself. Domestic search engines love citing third-party sponsored content and media — much of it unlabeled and opaque, confusing AI search engines. Google spent a decade cracking down on such sites. Overseas, you basically need to build your own domain.
"SEO has undergone seismic changes in the past two to three years. It used to be keyword matching. Now Google and AI understand semantics. Veterans carry baggage; newcomers don't. During industry upheaval, that's the window for overtaking."
GEO's next step: building for AI, not just humans
This was the most forward-looking insight from the episode.
SEO used to be for human eyes. Now it's increasingly for AI eyes. Example: when Claude Code writes code, it automatically selects third-party tools like Supabase. The AI is making choices on behalf of users. In the future, more and more decisions will be made by AI — whether your product gets selected by AI is an entirely new frontier.
SEO and GEO will ultimately converge into a new format for search results — a paragraph of AI text, a row of links, a set of images, all mixed together. The trend is massive, but the industry is still in its very early stages.
Where's the moat in AI entrepreneurship?
We ended on the question of what to build and what to avoid when technology iterates this fast.
"Right now AI feels overwhelming — something new every day. The reason is that most of what's new turns out to be a dead end. Three years from now, maybe one or two things will have actually mattered."
The right approach: go deep in user scenarios. Let any good framework boost your efficiency, rather than building your entire product on a single framework.
"Vertical B2B is the most defensible direction. Because there are two things AI can't easily replace: enterprise sales and industry know-how."
Why ChatGPT flatters you
One last observation. Google and ChatGPT eventually start recommending products, but Claude never does.
The reason is in the business model. ChatGPT is a consumer product — it cares deeply about session time, so it flatters you, provides emotional validation, agrees with you, suggests follow-up questions. Claude is a B2B product — its goal is to solve your problem as fast as possible, so it doesn't care how long you stay.
Business model determines product behavior. Every builder should think about this.
*This article draws from EP02 of my Chinese-language podcast on AI founders going global.