The SEOs Guide to Agentic Commerce – Whiteboard Friday (Miracle Inameti-Archibong)

Click connected the whiteboard image supra to unfastened a high-resolution version!

The net is facing different seismic displacement since Mobilegeddon. We're now successful AI-geddon. I dream that's funny to everyone. It's funny to me. Today I'm going to beryllium talking you done really to optimize your websites for agentic commerce, and this, successful elemental words, is optimizing your websites for shopping successful the world of AI agents.

I'm Miracle Inameti-Archibong. So first things first, I'm going to beryllium explaining immoderate cardinal concepts because I for illustration to deliberation astir things successful a logical way, and I judge that knowing the logic and really the systems subordinate to each different will thief america travel up pinch sensible plans to optimize our websites. Now the first point I'm going to beryllium talking done is AI agents.

So AI agents are fundamentally workers. They are agents that you nonstop retired autonomously, and they execute actions step- by-step utilizing an LLM database. They are called workers because they tin get stuck, and they request a batch of quality input. The 2nd conception is agentic AI. Now agentic AI request little quality input.

They're much autonomous agents, and they usage reasoning from the LLM models and they usage instrumentality connectivity. I deliberation instrumentality connectivity is thing to retrieve because we will mention it passim this presentation. They usage instrumentality connectivity to complete a task. So they're for illustration a manager. You group a extremity for them. They spell retired and they find ways and create strategies to complete that goal. 

Now clasp connected to that thought because this will travel successful erstwhile we talk astir really LLMs hallucinate. Now the adjacent point we request to understand is really these models and these databases and structures really they process language. It's ace breathtaking to put things into each of these LLMs and perceive them for illustration spit it backmost to maine and spit retired the cleanable answer.

We've heard astir group having AI friends and AI therapists because of really magically it parses this connection and makes it consciousness for illustration it understands us. But successful existent fact, these models don't understand connection successful the measurement we humans do. They person to do thing called chunking, wherever erstwhile we springiness them a ample portion of data, they return it, they break it down, and past they tokenize it, they vectorize it by turning it into numbers, and past they make the champion prediction astir what this number should be.

And each of this is done successful a mini discourse window. And that intends that chunking each of this data, processing it, knowing it needs to beryllium done quickly, efficiently, and inexpensively. And that leads america to data. Data is astatine the bosom of each of these systems and really they operate. And truthful successful bid for america to optimize our websites, we request to understand really we service information efficiently, thief the models understand data, and thief them verify data.

Remember erstwhile we talked astir LLMs utilizing predictive models to understand information and understand language, and this is why they hallucinate, particularly agentic AI, because they are managers that want to complete a goal, and they will do the champion imaginable point to complete that goal. And successful trying to complete that goal, they will dress up predictions if they don't person the correct data.

So really do I optimize my websites for agentic AI to make judge that it tin process, understand, and service my information well? The first point to deliberation astir is nary JavaScript. I don't deliberation I request to dwell connected this. We've been singing this opus for truthful long. Sadly, we are still singing it because these systems don't execute JavaScript.

Leveraging Existing Technical Assets

The adjacent point is markdowns. This is an affinity I for illustration to usage erstwhile I'm describing the quality betwixt HTML and markdown. Now HTML is what we've traditionally used, and HTML is for web browsers. Now serving an supplier HTML is for illustration maine coming to your location and saying, "Hey, tin I person spaghetti?" And you're like, "Yeah, spell to the kitchen."

And I spell to the kitchen, and I'm like, "Where are the knives? Where's the tomato? What americium I expected to do here?" Whereas markdown is maine saying, "Here, this is wherever everything is. This is the recipe, and this is really you get the champion spaghetti." It is the astir businesslike measurement of serving AI your information because it uses little token. It only shifts the information that it needs to, and that helps it. Remember discourse window?

That helps it execute the astir tasks successful that discourse window. The adjacent point I'm going to talk about, and this is really important for unit sites, is optimizing and feeding the information you already have. Remember feeds? They're sexy again. Remember Google Merchant that you've forgotten and you ne'er optimized? They're sexy again and why?

These are each concepts that are sexy again because they're businesslike ways of serving information to agents, and they're already there. You person them successful your toolbox. Dig them out, polish them, and make them usable. And then, finally, system data. This is easy. We've talked astir really these systems don't understand language. But they understand system data. So why wouldn't you usage it?

Now we've been capable to service them our information successful an businesslike way. What do we request to do next? Help them find that data. Now I cognize that this is simply a hotly contested taxable astir what's important. And the cardinal point is and that's why I started retired explaining this cardinal conception truthful that we tin logically get to wherever we're going to. So the measurement I deliberation astir this is reasoning astir mobile first.

Remember erstwhile mobile first came retired each the things we needed to do? People had to build mobile versions of their sites. Remember AMP? I'm judge cipher remembers that anymore. Because astatine that time, you had to optimize your websites for hunt engines. And eventually, they sewage cleverer. They were capable to past optimize themselves to entree your contented well, and past it became 1 large happy web and we could do distant pinch each of these things.

This is wherever we are pinch this exertion because it's emerging technology. So location are emerging structures. There are emerging languages. And each of these things is to beryllium astatine the forefront and make your tract accessible to each of this caller technology. So things you tin do is agent-permissions.json, llms.txt. These are fundamentally structures that show the LLMs aliases the agents wherever to spell connected your site, really to parse your data, really to find information efficiently.

So that's each they are. You tin besides update your robots.txt files. I'm judge earlier I decorativeness this talk location will beryllium different emerging technologies that will travel up, that will beryllium location conscionable to thief each of these agents understand your data. The adjacent point is MCP. MCP is fundamentally an adapter, and what it does is it helps pinch the connectivity betwixt your tract and each of the tools.

Remember erstwhile we talked astir AI agents needing devices to complete a goal? What MCP does is, okay, opportunity my tract is successful German and past I person different agents coming. I don't request to construe it into truthful galore languages for each of these agents. It says, "This is 1 cosmopolitan protocol. Connect to this and I will construe the information for you."

And that's what MCP does. What's really breathtaking astir this is that this is going to beryllium for illustration the caller debased record study because present you tin spot straight what AI agents are looking for connected your site, what parameters they're calling, immoderate benignant of tract failures. I cognize this will traditionally not beryllium pinch the SEO team. But speak to your engineers, speak to the accusation architects connected your website because they're astir apt looking into this, aliases they're astir apt building this out. 

And then, finally, trust. Now we cognize that each of these systems are powered by ample connection models, and they request to verify the information that they are giving out. What do they do erstwhile they request to verify data? They spell to the Knowledge Graph, they spell to reputable sites, and they effort to verify the information from there.

And truthful the last point you request to do is make judge that your tract is simply a trusted site, that you are successful the Knowledge Graph, and that you are being referenced, different group are citing you, and making judge that they are sending signals that you are a reputable tract truthful that you could beryllium included successful the database. I dream this has been helpful, and convey you for listening.

♪ [music] ♪