Din e-handel syns på Google. AI-agenten hittar dig inte.

## Introduction

[A] "Welcome to the deep dive. If you've spent, you know, the last decade or two operating a business online, you probably already know the holy grail of digital retail."

[B] "Oh, absolutely. It's always been that one thing."

[A] "Right. It's beautifully straightforward, honestly. You fight tooth and nail for a top spot on a page of 10 blue links."

[B] "Yeah, you optimize your meta tags, you build out those backlink empires."

[A] "Exactly. You're basically just catering to search engine crawlers all day. But today we have this really fascinating and honestly highly urgent piece of source material on the table."

[B] "It really is."

[A] "Yeah. We're looking at a brilliant analytical article by Stefan Sånnell. It's from the tech publication Spinout and it's titled: Your e-commerce ranks on Google. The AI agent can't find you."

[B] "Which is such a striking title. I mean, the premise of that alone should send a shiver down the spine of anyone who relies on traditional search traffic for revenue."

[A] "Oh, totally. Because it points to this massive systemic breakdown in how products are actually discovered and bought today. Which perfectly brings us to our mission for this deep dive. We are going to explore this really aggressive shift away from those traditional 10 blue links."

[B] "Right. And we're moving toward this new frontier of AI-driven agentic shopping. And, you know, more importantly for you listening, we're going to decode the literal infrastructure here. The new plumbing, essentially."

[A] "Exactly. The backend plumbing that's currently being built by the biggest tech companies on earth. Because this infrastructure will absolutely determine which businesses survive this transition and which ones just turn completely invisible."

[B] "Yeah. Invisible is the right word for it."

## The E-Commerce Reality Check

[A] "Right. So to set the stage for why this is happening right now, I want you to imagine for a second that you have access to the most capable assistant in human history."

[B] "Okay."

[A] "Like, this assistant has synthesized the entire internet. It can debug complex software in seconds. It knows everything."

[B] "It knows everything. Now imagine asking that same genius assistant to do something incredibly mundane, like buy you a new set of bike tires."

[A] "Right."

[B] "And then imagine watching it completely fall apart."

[A] "And that's not just a hypothetical, right? That striking image is actually playing out in reality right now. It exposes this massive blind spot in our current artificial intelligence models. The smartest entity in the room is totally paralyzed by a basic retail transaction."

[B] "It's crazy."

[A] "It is. But to understand why the AI fails, we kind of need to look at the current state of e-commerce architecture. Because we often operate under this assumption that retail has been, you know, entirely digitized at this point."

[B] "Oh, sure. I mean, we buy everything online."

[A] "Right. But the reality is that e-commerce still only accounts for roughly 20% of total retail sales."

[B] "Wait, really? Just 20%?"

[A] "Just 20%."

[B] "Wow. I mean, for anyone whose entire career revolves around digital storefronts or, you know, platform migrations and customer acquisition costs, hearing that 80% of transactions still happen offline — in the physical world — it feels almost contradictory to our daily experience."

[A] "It does feel super counterintuitive. But looking at that 20% figure as like a limitation of digital adoption kind of misses the broader historical impact. E-commerce didn't just carve out a one-fifth slice of the retail pie. It acted as this massive forcing function that transformed the remaining 80%."

[B] "Oh, I see."

[A] "Yeah. Physical retail had to radically evolve just to survive the Amazon effect."

[B] "Right. Right."

[A] "So you get omni-channel strategies, buy online, pick up in store, highly sophisticated inventory routing."

[B] "Seamless return logistics, all of that."

[A] "Exactly. All that physical infrastructure was built in response to digital pressure. E-commerce basically forced the physical store to become this hyper-efficient node in a network."

[B] "Okay. So if the first wave of e-commerce forced physical retail to upgrade its logistics, this current wave of AI is going to force digital retail to upgrade its underlying data structures."

[A] "Yes. Spot on."

## The Bike Tire Story

[B] "Which actually brings me to the core anecdote from that Spinout article. It perfectly illustrates this friction that's happening right now. So the piece follows this analyst at the ad tech firm Criteo."

[A] "Right. The guy who lives in New York City."

[B] "Exactly. And he cycles everywhere. And if you've been to New York, the asphalt there fluctuates between merely bad and just completely catastrophic."

[A] "Oh, it's terrible. It's an environment that demands highly specific hardware."

[B] "Right. He can't just buy a generic tire. He needs something engineered for glass, potholes, and unpredictable urban debris."

[A] "Yeah."

[B] "So instead of doing a standard keyword search and, you know, sifting through SEO-optimized blog posts about top 10 bike tires for 2026, he does something different. He spends an hour consulting an AI agent."

[A] "And he gives it really strict parameters."

[B] "Super strict. Puncture resistant, specific millimeter dimensions, a certain weight class, and high durability. And the thing is, during the discovery phase, the agent performs brilliantly. It does what it's built to do."

[A] "Exactly. It asks clarifying questions about his riding style. It synthesizes reviews across all these random cycling forums, and it presents a highly curated list of options that perfectly matches his criteria."

[B] "Yeah. Because the large language model executed its primary function flawlessly right there. It parsed unstructured data, it reasoned through the user's constraints, and it generated an intelligent recommendation. It mapped the territory perfectly."

[A] "But the map isn't the territory."

[B] "Yep."

[A] "Because after this incredible hour of personalized curation, the analyst ultimately closes his laptop, walks down the street to a local physical bike shop, and just buys the tires in person."

[B] "Yeah. The AI totally failed at the finish line."

[A] "Exactly. But wait — if the AI is so smart at finding the exact right product, why did he end up going to a local physical shop?"

[B] "Well, when he tried to actually execute the purchase through the agent, the links it provided were dead."

[A] "*Nah.*"

[B] "Yeah. The highly recommended tires were actually discontinued models. The agent confidently quoted a price of like $60, but clicking through to the retailer revealed a price of $90."

[A] "That's incredibly frustrating."

[B] "Or, worst of all, the product was listed, but there was zero real-time stock information available. You just couldn't tell if you could actually buy it."

[A] "So it's basically a systemic failure at the point of transaction."

[B] "Exactly."

## The Sommelier Analogy

[A] "The author uses this phenomenal analogy to explain the bottleneck here. An LLM is, fundamentally, a well-trained conversationalist."

[B] "Okay."

[A] "Think of it as a world-class sommelier at a really high-end restaurant."

[B] "Oh, I like this."

[A] "Right. That sommelier can stand at your table and speak poetically about the terroir of a specific vineyard, or the weather patterns that affected a vintage, the subtle tasting notes of a Bordeaux."

[B] "But the problem arises when it's time to actually pull the cork."

[A] "Yes."

[B] "Like, the sommelier can have all the abstract knowledge in the world, but if they don't have an accurate, real-time look at the restaurant's physical wine cellar, their expertise is totally useless."

[A] "Exactly."

[B] "If they recommend a bottle that sold out three hours ago, or a vintage where the supplier suddenly doubled the wholesale cost, the whole customer experience is ruined."

[A] "Right."

[B] "The AI knows everything about the physics and manufacturing of bike tires, but it has zero visibility into which specific SKU is sitting on a warehouse shelf in Brooklyn at 2 p.m. on a Tuesday."

[A] "And that visibility requires an entirely different category of data. Agents need deterministic, real-time transaction feeds. They need the actual wine list."

[B] "Exactly."

[A] "They need normalized product information, inventory counts, dynamic pricing APIs. Without that accurate wine list, the intelligence of the agent is basically just trapped inside a vacuum."

[B] "Okay, so if the fundamental problem is that we have these brilliant conversational agents trapped in a box without access to the world's real-time inventory, I guess the solution is building the pipes to connect them."

[A] "That's exactly what's happening. And I want to signal to you listening that we are about to get a little bit into the architectural weeds here."

[B] "Just a little bit."

[A] "Just a bit. But I promise we'll keep it practical, because this is the crucial plumbing that will dictate your revenue over the next few years. We really need to dissect three critical acronyms."

## MCP — The Universal Adapter

[A] "Let's start with the first piece of infrastructure: MCP, or the Model Context Protocol."

[B] "Right, so MCP is the architectural layer designed to break the AI out of that vacuum we just talked about. It was developed initially by Anthropic, and it has rapidly become the open-source standard for how AI models communicate with external tools and data sets. It is basically the interpreter sitting between the reasoning engine of the AI and the hard data of external systems."

[A] "Okay, let's ground that mechanism for the listener, because calling it an interpreter almost undersells the engineering, I think."

[B] "Fair enough."

[A] "Like, if I look at the architecture of the web today, an AI model like Claude inherently outputs text, right?"

[B] "Yes."

[A] "But my company's inventory ERP system doesn't understand conversational text. It requires specific structured queries. So if Claude speaks English, your inventory ERP speaks a second language, and Gmail speaks a third. MCP is essentially the universal translator, so the AI isn't just sitting in an isolated box."

[B] "You nailed it. It functions a lot like a universal power adapter for data."

[A] "Oh, that's a good way to put it."

[B] "Yeah, it doesn't matter if you're traveling to a country with a 220-volt wall socket or a 110-volt socket. The adapter standardizes the current so your laptop doesn't fry. MCP standardizes the API endpoints using JSON."

[A] "Got it. So it allows the AI agent to request inventory data from a local database without the developer having to write custom bespoke code for every single integration."

[B] "*Wow.*"

[A] "And that standardized connection is absolutely vital, isn't it?"

[B] "Well, without it, developers are stuck building fragile one-to-one bridges between every single AI model and every single database."

[A] "It's a nightmare."

[B] "Right. So MCP provides the secure universal protocol for the AI to dynamically read a company's internal knowledge base, check a CRM, or query a live inventory system — and it does all that while maintaining strict permissions."

[A] "Exactly. It gives the agent functional hands and eyes."

## UCP — The Commerce Standard

[A] "But MCP is a really broad protocol for all data. When we narrow our focus specifically down to retail and e-commerce, we hit the second piece of the puzzle: UCP."

[B] "The Universal Commerce Protocol. And honestly, the gravity of this specific protocol cannot be overstated based on the players involved."

[A] "No, it's massive."

[B] "Google officially launched UCP at the National Retail Federation Conference. Sundar Pichai himself presented it."

[A] "Right."

[B] "But Google launching a proprietary e-commerce standard would immediately trigger massive antitrust scrutiny, right? And tons of industry pushback."

[A] "Oh, absolutely. It would be a legal nightmare."

[B] "So they bypassed that completely by co-authoring UCP as an open standard. They basically walked on stage holding hands with Shopify, Etsy, Target, Walmart, and Wayfair."

[A] "That coalition completely changes the reality of this rollout. I mean, the documentation is already live at UCP.dev. The sheer market capitalization of that group guarantees adoption."

[B] "Oh, 100%."

[A] "And UCP solves that wine list problem we discussed earlier by establishing a single universal schema for commerce."

[B] "The Spinout article visualizes the necessity of this beautifully through a restaurant delivery analogy."

[A] "Right."

[B] "Imagine owning a local restaurant today. If you want to capture digital orders, you are forced to manually upload your menu, your pricing, and your operating hours to Uber Eats. Then you have to log into DoorDash and do it again in a completely different format. And then again for Grubhub. It is a massive operational headache. Every platform demands its own specific data structure, and keeping them all synchronized in real time is just a nightmare."

[A] "It's impossible."

[B] "But UCP acts as the singular, definitive menu. A business registers its inventory, pricing, shipping mechanics, and return policies exactly once using the UCP format."

[A] "Just once."

[B] "Just once. And because it is a universal standard, every single AI application — whether it's Google's Gemini, OpenAI's assistants, or a custom agent built on Anthropic — can natively ingest that data without any translation errors."

[A] "That's incredible."

[B] "Furthermore, UCP doesn't just allow the agent to read the data. It actually facilitates the action. The protocol is designed to handle the complex variables of a transaction — like calculating localized tax, applying dynamic shipping rates, that kind of thing."

[A] "Exactly. And processing payment tokens."

[B] "The AI agent can securely fill out the forms, process the transaction, and arrange the logistics automatically. The human buyer never has to type their credit card or shipping address into a fragmented checkout page ever again."

[A] "It effectively eradicates the friction that caused our New York cyclist to abandon his search in the first place."

[B] "Precisely."

## AEO — The New SEO

[A] "Okay. So we have MCP acting as the universal adapter connecting the AI to the outside world."

[B] "*Wow.*"

[A] "And UCP acting as the standardized commerce language flowing through those connections."

[B] "Yes."

[A] "So if you are a business owner or a digital strategist listening to this, the immediate question is obvious. How do you actually use these pipes to make sure the AI chooses your product instead of your competitors?"

[B] "And that operational shift brings us to the third and maybe most important acronym. It serves as the direct successor to SEO."

[A] "Right."

[B] "We are entering the era of AEO."

[A] "Answer Engine Optimization."

[B] "AEO. This feels like the pivot point of the entire deep dive, honestly. Because for 20 years, search engine optimization was a probabilistic game. You were fighting a war of attrition to rank third on a page of results. Knowing that users would scroll, click, bounce back, click again — you were optimizing for a web crawler that essentially just matched text strings and counted backlinks."

[A] "But AEO operates on a fundamentally different premise. You are no longer fighting to be an option on a list."

[B] "You are fighting to be the singular, definitive answer the AI agent provides when a user asks for a recommendation."

[A] "Like which bike tire should I buy?"

[B] "Exactly. The margin for error drops to zero."

[A] "The source material uses a really brilliant GPS analogy to illustrate this shift in optimization."

[B] "Oh, the sign above the shop."

[A] "Right. When you put a physical sign above a brick and mortar shop, you design it with bright colors and large fonts to catch the eye of a human driving by. But an autonomous vehicle doesn't care about the aesthetic of the sign. It's not looking at the neon lights."

[B] "No."

[A] "To ensure the autonomous vehicle stops at your business, the sign must be written in a language the GPS natively understands. Longitudes, latitudes, and digital waypoints."

[B] "Okay. So applying that to a digital storefront, what does that mean practically for the listener's business? Because right now, most e-commerce sites are designed entirely as human-readable signs."

[A] "Right. If I go to an online store and see a price tag of $50 with a red line slashed through it and a new price of $40 next to it in bold green text, my human brain instantly understands the context."

[B] "You know this item is on sale."

[A] "Exactly. Or if I see a greyed-out Add to Cart button with a red X next to it, I intuitively know the item is out of stock."

[B] "A machine intelligence does not intuitively interpret visual cues or clever marketing copywriting. It does not parse the vibes of your landing page."

[A] "Right. It doesn't care how pretty the website is."

[B] "Not at all. For an AI agent to confidently recommend that product, the underlying data structure must explicitly declare those variables. It looks for standardized schema markup."

[A] "So it's looking for a machine-readable context."

[B] "Yes. It needs a machine-readable JSON payload that explicitly states the price is valid until a certain date, or a dynamic inventory API that returns a Boolean true or false for stock availability. You are no longer optimizing for a crawler indexing text, but for an agent needing up-to-date, reliable product data."

[A] "The agent demands hard deterministic facts."

[B] "Exactly. And the reason it demands those facts is self-preservation. Because an AI agent's entire value proposition is utility and trust. If an agent recommends your specific bike tire to a user, and the user clicks a generated link only to discover the item has been out of stock for weeks, the user doesn't blame the retailer."

[A] "The user blames the AI for hallucinating."

[B] "And the AI hates that. To protect its own user experience, the AI will actively penalize and filter out businesses with poor data hygiene."

[A] "So it creates this ruthless feedback loop."

[B] "Very ruthless. If your data isn't structured, pristine, and accessible via these new protocols, the AI doesn't skip you because its algorithm determined your product is inferior. It skips you because, to the machine's parsing logic, your business literally does not exist."

[A] "*Wow.*"

## Why This Time Is Different

[B] "Okay. But I can hear the skepticism brewing for some of you listening, and it is a completely valid reaction. Because it's easy to dismiss this as just another tech trend that won't materialize for years."

[A] "Sure. People say that a lot."

[B] "We've lived through these grand industry-altering tech proclamations before. I mean, the Spinout article even acknowledges the graveyard of failed social commerce experiments."

[A] "Oh, yeah. There's a big graveyard."

[B] "Didn't Meta try to pull off a massive native checkout feature across Instagram and Facebook back in 2022?"

[A] "And it largely just fizzled out."

[B] "It did. And didn't Google aggressively push Buy on Google for years before finally deprecating it in 2023?"

[A] "Yep."

[B] "So why should the listener care now? If the most well-resourced tech monopolies on the planet have previously attempted to build universal checkout layers and failed, why jump on this now? Isn't this just another hype cycle?"

[A] "The skepticism is historically justified, but I think it misdiagnoses why those previous attempts failed. Meta and Google were trying to build the roof before the foundation was poured. They attempted to force seamless native checkout experiences, while the underlying data was still totally fragmented, proprietary, and reliant on scraping human-readable HTML. The infrastructure — these open protocols like MCP and UCP — simply did not exist yet."

[B] "So they were trying to force the sommelier to guess what was in the cellar without giving him the wine list."

[A] "Exactly. But the author counters this skepticism by invoking that famous observation from Bill Gates."

[B] "We overestimate change in the short term and underestimate it in the long term."

[A] "Right. It is highly probable that fully autonomous agents — the kind that independently manage your budget and negotiate purchases without any human oversight — are still a little ways off from mainstream adoption. But the protocols that will govern that future, the standards that will rigidly separate the visible from the invisible, are being written and deployed into production right now."

[B] "And the early adopters here — the scale of deployment — it isn't just a sandbox experiment anymore."

[A] "No, not at all."

[B] "Google AI Mode and Gemini already process queries for hundreds of millions of active users. OpenAI just launched their instant checkout feature, partnering directly with the UCP coalition: Shopify, Walmart, Etsy, and DoorDash."

[A] "We are well past the theoretical phase here."

[B] "We really are. And if we look back at the history of digital commerce, the platform shifts are always sudden and they are non-negotiable."

[A] "Yeah, I'm thinking back to mobile again."

[B] "Perfect example. For years, developers and brands argued that mobile optimization wasn't a priority. The prevailing logic was, you know, customers browse on their phones, but ultimately they open their laptops to complete a purchase, so investing in responsive design is a waste of capital."

[A] "Then Google casually announced that mobile friendliness would officially become a primary ranking signal in their search index."

[B] "And the debate ended overnight. Entire industries scrambled to rebuild their websites because the platform dictated the new reality."

[A] "And we saw the exact same compliance scramble with the transition from HTTP to HTTPS security protocols."

[B] "Oh, yeah."

[A] "And most recently, with the introduction of Core Web Vitals, where page load speeds directly impacted visibility."

[B] "Platform owners dictate the physics of the digital world, and the market is forced to comply instantly or just watch their revenue dry up."

[A] "And the exact same historical pattern is unfolding right in front of us with agentic commerce."

[B] "It really is. The window to prepare your infrastructure is open right now — before the penalty of invisibility is fully enforced."

## The Lights Out Future

[A] "Which brings us to the ultimate takeaway here. What will separate the winners from the losers in this next era of commerce is not going to be who has the biggest marketing budget or the cleverest brand campaigns."

[B] "No."

[A] "The new, insurmountable competitive advantage is simply data hygiene."

[B] "Clean data."

[A] "Clean, structured data is the absolute prerequisite for participation. Having real-time, machine-readable, schema-based product catalogs is no longer just an IT best practice. It is the core marketing strategy."

[B] "The good news is that if you're already utilizing sophisticated product feeds for Google Shopping or dynamic ad retargeting, you aren't starting from absolute scratch."

[A] "Right. UCP builds upon a lot of those existing principles."

[B] "But the fidelity of that data has to be absolute. Because if your catalog is riddled with ghost inventory, conflicting pricing variables, or broken variation links, the AI doesn't skip you out of malice."

[A] "It won't hold a grudge."

[B] "It simply won't see you."

[A] "Yeah. Like the local bike shops in the Criteo analyst's story — you might have the exact product the customer desperately needs sitting right there on the physical shelf, but bad data prevents the connection entirely."

[B] "And if we follow the trajectory of this infrastructure to its logical conclusion, we arrive at a profoundly disruptive concept for the future of retail. In industrial manufacturing, they use this term called lights out to describe a factory that is so heavily automated by robotics that it can operate 24/7 in the dark without a single human worker on the floor."

[A] "A completely autonomous supply chain."

[B] "Yes. And we are rapidly building the architecture for lights out e-commerce. If MCP and UCP function perfectly, we are approaching a reality where your personal AI agent holds the keys to your digital wallet. You give it a budget and a set of preferences, and it autonomously monitors your household inventory, negotiates with retailer APIs in milliseconds, and executes transactions purely based on logic, specifications, and data fidelity."

[A] "That is wild."

[B] "It is. Which leaves us with a rather unsettling, maybe provocative thought to consider here."

[A] "Yeah."

[B] "As a business, as we move quickly toward this lights out reality, how do you market to a machine?"

[A] "Right. If an AI is doing the shopping, the entire psychological playbook of human marketing — the impulse buys triggered by a flashy banner ad, the emotional resonance of a brand story, the artificial urgency of a countdown timer — all of that becomes completely obsolete."

[B] "Because a logic gate doesn't feel FOMO."

[A] "Exactly. An AI doesn't feel emotion. So how do you market to it? That is a staggering paradigm shift to leave you with. We are really moving from the psychology of persuasion to the mathematics of precision."

[B] "Well, thank you for joining us on this deep dive into the underlying architecture of tomorrow's commerce. As you go back to running your business or analyzing your market, take a hard look at your digital footprint. Stop looking at it through the eyes of a human customer susceptible to clever design. Look at it through the cold parsing logic of an AI agent. Make sure your sign is written in a language the machine can actually read. We'll catch you on the next deep dive."

[A] "Thank you."