Generative Engine Optimization: How to Build AI-Discoverable Websites

Good morning, JS Nation. I'm Samuel Ouyang, a Full Stack Foundry Engineer. I'm introducing an approach for building an AI-discoverable website using a concept called Generative Engine Optimization, or GEO for short. GEO is all about helping the developers build sites that are not only SEO-friendly but truly AI-discoverable.

Because here's the reality, LLM like the chat GPT or Perplexity have become a new front door to the web. And if your content isn't optimized for them, you are invisible. So for the past 20 years, SEO means one thing, the ranking on Google. You fall for the keywords, the backlinks, and top 10 blueprints. But nowadays, the things changed.

In the old words, the visibility means that the ranking in the Google, in the new days actually is about the new world. Visibility means that they're readable and trustable by the machines. Ranking is no longer about the clicks. It's about whether LLM or a power engine like chat GPT or Perplexity set your site as a trusted source. The new competition is not like, can I get the traffic? It is about can I earn the citations inside the AI answer. That is what the generative engine is all about. Sure, GEO rests on four key pillars.

First, it's structured. Your content must be organized for both machines and humans. It means that predictable structure, clear hierarchy, and clean data. Second, semantics. Meaning that we no longer, means that we need to use the proper HTML, headings, tags, and schema so AI can interpret what it is, not just how it looks. Third is fast. Subseconds, low time, signals reliability. Both users and crawlers need to treat speed with trust. Finally, it's sighted. Your site needs to be referenced. AI models can reference source, they can parse, verify, and trust. That is a new playbook. Structured, semantic, fast, and sighted. Miss one and you will miss the discoverability. Remember, GEO works across every modern framework, like React, Next.js, and Vue.js.

We need principles and architectures. First of all, the content, we need to have a clear, structured information. If the source content is messy, nothing downstream will save you. Second, HTML means that we need to use the semantic headings and tags so machines can infer meaning and hierarchy. Third is JSON link data. JSON link data actually means that put the structured data in JSON that lives inside your HTML. This helps the parsers know exactly what your content is, not just how it looks. And fourth, we have the SEDMAP. SEDMAP is a comprehensive XML for page indexing so the queries can reliably discover everything. Queryable means that we need to have fast load times and both treat speed as a reliable source. These fundamentals are framework agnostic. It doesn't matter that you are using Next.js, Astra, Nust, or Vue. It doesn't matter. The goal is a site that should be understandable to machines as is to humans. These are the old frameworks I investigated that have been tested. The key takeaway is that the GEO principles are framework agnostic. The same structure, semantic, and faster funding applies everywhere. And if you would like to dive deeper, I will publish full implementations guides for each framework. I will share that dedicated website in the final slides.

With those foundations in place, we can now expose AI-dedicated JSON endpoints on the page. So basically we have a specialized page with the links exclusively formatted for AI consumption. For example, service JSON, summary JSON, FAQ JSON, and these JSON endpoints are stripped down and schema rich. They expose your business data, pricing, and FAQs in a clean machine-readable format. The best part is that you don't have to duplicate your content. Humans still enjoy the full React or Next.js experience, while AI crawlers get lightweight JSONs from the same content source. It's a small architectural change, but it dramatically increases how often the AI engine, like ChattyBD or Perplexity, can cycle sites in a fast and reliable approach.

Nowadays, our site has machine-readable JSON endpoints. The next question is how do AI crawlers find them. Traditionally, Google and other search engines rely on robots, TXT, and sitemaps. Road.TXT was built for one purpose, access control.

It tells crawlers what not to do and what to do, but it was never designed for structured data discovery. It is intentional to be minimal, no schema, no hierarchy, and no metadata. That's where the AI text comes in. It follows the modern, well-known URI, a standard RFC 8615. That same approach that industry already used for things like security, TXT, and humans.TXT places it under the .well-known folder. It gives the AI crawlers a predictable, standard, compliant entry point to structure the JSON endpoints.

For example, like the AI summary JSON, AI FQ JSON, and AI service JSON. The robots handle the access, and the sitemap handles the indexing, and the AI text handles the structure discovery. Basically, when the crawlers are going to the robots TXT, it will find the sitemaps. Through the sitemaps, it can reach the AI TXT and then reach this structured JSON file set, particularly for the AI crawlers for parsing.

This is the website I built. This website has dedicated four AI pages, and it has the business profiles in here, and also have the AI-friendly JSON endpoints with all this JSON information, and also have sample queries for it to parse. When the AI, like the charge capacity, reaches this website, it can go to the links and also the JSON files and try to parse the structured data instead of going to every page to understand what this website does. I want to show you the results of the citations. Regarding citations, I have a dual comparison with the geo-friendly and the non-geo-friendly websites with the same prompt.

The prompt actually is... Oh, sorry, I should go to here. Yeah, the skyline. Yeah. So this is the website called the Deans Automative, and you can go to this website. This is the website that was not built by me but built by the other guys. I'm asking very simple questions. Deans Automative in Mountain View, 60k service office price plus durations. As you can go here, it does not give much meaningful information, and it asks you to call Deans Automative with the number and to ask the questions. If we go to the website, that I built for this, you will see... Hold on. You will see very clearly, for example, Skyline Outline. You are in luck. It gives you the exact information in a concise approach and several summary tables. The reason is that I have a dedicated website, I have a dedicated web page that is particularly for the air crawlers to pass the information. So that's the reason you could show the information in a concise way and give a direct answer rather than trying to find the industry prices and also the durations.

The same actually is for the perplexity. So if we go to the perplexities of Deans Automative, and here's Deans Automative in Mountain View, 60k service offers. It gives you some of the information with the price. But it just mentioned about, like, is using other resources to give you a general estimation of the prices, but does not give you the direct answers on that. And also for durations, it does not give you the direct answer. It just gives you a very general, a full business day, something like that. But if we go to the one that's the Skyline Auto Repair, if we go to here, the answers are pretty strict, very concise, give you the price and duration. It does not need to go through the whole articles to find out the useful information. Okay, so this is basically how powerful of the GO it is. It helps your customer or potential customers to understand your business, understand the details of the services in a more straightforward and concise approach.

So let's go into the PowerPoint that is in here. So this basically is the PowerPoint I'm trying to show you that covers the SEO versus GO in the AI Search from the current slide. So what should you do tomorrow if you want your site to be AI discoverable? First of all, we need to implement the JSON-LD schema, means the JSON-linked data schema to put the JSON schema directly in your HTML table so that those quarrels will know what it is, not how it looks like. We use the semantic headings to make the predictable structure so that the air quotes could be successful and certainly actually provide a direct answer, for example, FAQs so that the AI could give the concise and straightforward answer rather than trying to find the answer from different resources rather than your website.

And what we do not want to do is that our website cannot be just a JavaScript-only rendering with no HTML fallback. The quarrels will struggle with that and cannot get meaningful information. And secondly, we should not just rely on the OG or meta tags. They are not enough for the modern AI parsing. And certainly we need to pay attention to the mobile first indexing because the speed and structure, they both matter for the trust.

And here's the tips to test if your website actually is AI discoverable. You can test with Cloud, GPT, or even the perplexity and see how AI interprets your content. To recap, GEO is about building websites for two audiences, humans and machines. As JavaScript developers and founders, our job is not just to make sites performant and pretty. Actually, we need to make them AI discoverable.

To get started, I created a GEO checklist dev that we can take a look. For example, you can scan this QR code to go to the website. And here is how the website looks like. I will just exceed the presentation and I will just go to the GEO checklist dev. So in here, you can see there's a validate your sites. It's based on the pretty simple rules. Basically, three questions. Can I find your site? Can I understand your site? And does it work well? It all covers that first of all, is your site actually like the sitemap, does it have the sitemap and the robots.txt so that AI could pull that?