Original video in Chinese.
Key Takeaway
- Changes in traffic rules: in the AI era, SEO shifts to GEO, from ranking competition to improving citation rate; AI direct answers replace opening webpages, so seize the change and make money quietly.
- GEO starting methods: optimize the whole site for AI friendliness (llms.txt manual, robots.txt authorization); page-level instructions (script tag/GEOHead component); structured data enhancement for fast extraction.
- Call to action: the author is applying this to products such as Prompt House and monitoring optimization in the backend; there are few global early movers, so the timing is good, and I’m determined to invest.
Whenever the rules for traffic allocation change, you have to jump in decisively.
Because traffic can almost be directly equated with money, and a lot of money at that. If you know even a little about the SEO industry, you know how satisfying it is for people in this line of work to make a fortune quietly, and to do it for many, many years once it starts.
Now, we are standing at a turning point where the rules of traffic allocation are changing completely. Because AI has arrived.
The old logic was: users entered keywords to search; the search engine returned a list, and users chose which webpage to open themselves; the focus of webmasters’ competition was the “ranking position.”
Now, with AI, users can directly obtain an integrated answer through natural language conversation — no longer needing the intermediate step of opening a webpage, but getting a direct answer.
This change is basically a table-flipping-level move. After AI barged in like this, the traffic rules changed completely. So the focus of webmasters’ competition has shifted from “ranking position” to “citation rate.”
So, how do you improve the “citation rate,” or more bluntly, how do you make the AI middleman see your website and cite your content when answering? That is what GEO (Generative Engine Optimization) is for.
So, how should you get started?
At present, there isn’t yet a complete, systematic summary of GEO tactics. I’ve summarized some of my own, and I previously posted them in the Knowledge Planet newtype community. Let me lay them out here in full.
I think the foundation of GEO must be making it very easy and clear for AI to access your website and understand your website and every page. So there are three things you can do.
First, optimize the entire site. The goal is to turn the website into an “AI-friendly infrastructure.”
You can create llms.txt. For AI, this is a “manual” for your website. It tells AI what your website does, what its core value is, and how to correctly use your content.
For example, this is what I created for Prompt House. It clearly tells AI what Prompt House does; whether the site’s content may be used for model training; whether the site allows crawler access; which pages and resources on the site are important; and when users ask what kinds of questions, they can come to this website for answers. And so on.
In addition, you can clearly authorize mainstream AI crawlers to access the site in robots.txt, and suggest a reasonable crawling delay.
You see, with this manual, AI gets to know the Prompt House website very well. That is what I meant earlier by “AI-friendly infrastructure.”
Second, optimize key pages. The goal is to guide AI to understand and recommend more precisely through page-level instructions.
On this front, Vercel previously proposed a very good method. They suggested that you can embed instructions in a script tag to tell AI what needs special attention on this webpage.
For example, tell AI what this page is, what its main function is, what content categories it contains, what user queries should lead to this page being recommended, and what format should be used for source attribution when citing it.
Because it is an unknow type to the browser, it won’t affect rendering, but it can still catch AI’s attention.
But this method has one drawback: you need to manually embed it on every page. If your website has multiple pages, you need to copy and paste it into each HTML file, which makes maintenance a bit costly.
So I made a GEOHead component. You only need to import the component in the layout or page, and dynamic injection is achieved, which is especially convenient.
For example, for the Blog page, I emphasize to AI that this is where AI tool usage tips and best practices are shared, and so on; for the Docs page, I emphasize to AI that this is where the MCP integration guide and technical documentation are, and so on.
So through the first step, AI gets a general understanding of the entire website; through the second step, AI gets a further, more precise understanding of specific pages. However, all of this is still pure text interpretation and remains relatively limited.
To go a step further, we can also enhance the website with structured data. Simply put, this means adding machine-readable markup to web pages, helping AI quickly extract key facts, entities, and relationships as if it were reading a database, instead of relying only on pure text interpretation.
This part is too technical, so I won’t show it.
Everything I just mentioned has already been applied to my products, namely Prompt House and LearnFromYouTube. In addition, I’ve also built a backend monitoring page to make it easier to optimize based on data feedback. Once there’s progress after a while, I’ll make another video to share it with everyone.
As for GEO, both in China and globally, people have started researching and practicing it. Some people may have run a bit faster, but the lead isn’t that big. So with such a huge opportunity and such a good timing, there’s no reason not to invest.
OK, that’s all for this episode. If you want to understand AI, want to become a super individual, and want to find like-minded people, come join our newtype community. See you next time!