Meta published the blueprint for the agentic economy. No Mexican business fits in it.

We measured three verticals — hotels, dentists, and doctors — to see what separates the businesses AI names from the ones it ignores. In hotels, the ones AI names have 26 times more reviews than the ones it doesn't. In dentists, 5.5 times. In doctors, no difference at all: reviews predict nothing. The same number — how many reviews you have — explains almost everything in one vertical and nothing in another. There's no single formula for “showing up in AI”: you have to measure, vertical by vertical. And that's exactly what the blueprint Meta just published doesn't do.

That finding isn't about paying agents or shopping with AI. It's simpler and more urgent: before a business can transact with an agent, the agent has to know the business exists. And today, for most Mexican businesses, it doesn't. That's the blind spot I want to take apart in this article, because an important document just appeared that illuminates it precisely by what it leaves out.

Meta already drew the blueprint, but it's talking to a COO, not a Monterrey clinic

In 2026 Meta published "Beyond chatbots: The agentic economy is here," authored by Alex Schultz, Meta's Chief Marketing Officer (CMO). It is, without exaggeration, the blueprint for the economy that's coming: how businesses will operate once AI agents buy, book, and hire on people's behalf. It's not just another blog post; it's the written-down vision of one of the platforms with the most business contact in the world.

To size that economy, the document leans on third-party projections about how fast adoption arrives. Consulting firm Forrester cites 75% and 15%; analyst firm Gartner projects 60% by 2028; McKinsey puts the figure at 55%. These are the analysis houses big companies pay to read, and they all point the same way: agent adoption isn't a lab experiment, it's a curve that already started. The report's underlying message is that this is infrastructure, not a fad.

The document organizes that infrastructure into five layers. Identity: the agent knows on whose behalf it acts. Directory: it finds other agents and businesses. Messaging: they communicate with each other. Commerce: they can pay. And protocols: the common rules that make all of the above possible. About the second layer, the directory, the report says something that sounds very familiar to us: "agents need a directory to find other agents and businesses."

The cases the report showcases are Trendyol, Movida, and Sem Parar: a Turkish e-commerce giant, a Brazilian car-rental company, a toll operator. All corporate, all with tech teams and budget. And each section closes the same way: "talk to your Meta client partner." There's the clue to who the document is speaking to. No average Mexican business fits in those pages. They speak to whoever has an assigned Meta account executive, not to the owner of a neighborhood clinic that doesn't appear in those answers.

The five layers describe tomorrow. The Mexican business's question is about today

Here's the gap, and it has to be stated precisely to avoid the cheap comparison. Meta's five layers describe the infrastructure to transact with agents: the economy that's coming. They aren't wrong or vapor. They're tomorrow, and a likely tomorrow. When those rails are laid, they'll matter a lot.

But there's an earlier layer the document takes as solved: being named. Meta's layer 2 brushes against it when it says directories are needed, and treats it as a plumbing problem, a registry someone will build. For a Mexican small business that same layer isn't plumbing, it's a market problem you can measure today: when someone asks an AI for a service in your city, does it name you? That question comes before the five layers. An agent can't pay you, book you, or hire you if it never put you on the list of options in the first place. And it's exactly the question nobody was measuring on the side of the small business, in Mexico. That's what we did, and what follows are the numbers.

Start with dentists: the signal that separates is review volume

We measured the dentists guide against ChatGPT: 185 answers to real dentist-search questions. The clinics the AI does name have a median of 74 reviews (n=103) versus 13.5 for those it doesn't (n=496): 5.5 times more. Meta's agentic economy assumes you're already on that short list. In dental, the list is short and already taken by businesses with review histories well above average.

The same pattern, 26 times sharper: hotels on Google AI

We switched vertical and surface: hospitality against Google AI answers. Here the gap turns brutal. The hotels Google AI names have a median of 2,211 reviews (n=32); the ones it doesn't name, 85 (n=458). A 26-fold difference. In hospitality, the signal separating the named from the invisible is enormous and consistent.

The practical reading is harsh. If your hotel in the AI answer competes against properties with thousands of reviews accumulated over years and you have dozens, the model decided before you optimized a single line of your listing. It's not that the algorithm has it in for you; it's that the review-volume signal in this vertical is so strong it flattens almost everything else. That's the kind of data Meta's blueprint doesn't touch, because its document describes how transactions will happen tomorrow, not who's winning visibility today.

Where the pattern breaks: doctors, where reviews don't separate

And here comes the data point a marketing brochure would hide. We measured doctors against ChatGPT expecting the same pattern as the other two verticals. It didn't show up. Named doctors have a median of 30 reviews (n=49); unnamed ones, 24.5 (n=540). Practically the same. In medicine, reviews don't separate who the AI names from who it doesn't.

That null is why this is a measurement and not a campaign. If we were selling "get reviews and the AI will name you," we'd have to delete the medical vertical from the deck. We didn't delete it: we published it, because the one that doesn't fit is worth as much as the two that do. The real lesson isn't "reviews." It's that the correlate changes by vertical, and that's why you measure instead of assume. What decides visibility in hospitality isn't what decides it in medicine, and probably isn't what decides it for notaries or auto shops either. Anyone selling you a universal formula simply didn't measure the three verticals.

What to do before building on the five layers

Back to the clinics that don't appear. Meta's blueprint is real, and we run on its rails: Plexa uses the WhatsApp Cloud API and click-to-message ads (CTWA). We don't fight Meta, we flag an omission. Its document assumes you already exist for the AI. For most Mexican businesses that assumption doesn't hold: in the first edition of our study, the AI named no business from the directory in 62.3% of the answers — we count at the answer level, not by session or click (DOI 10.5281/zenodo.21433409). Our catalog holds 12,507 businesses, but the number that matters isn't how many we list: it's that in nearly two out of three answers the AI names none of them.

Let's call it a correlation, because that's what it is: no formula guarantees the AI names you. But you can't optimize what you don't measure. Before Meta's layer 1, identity, there's a layer zero for your business: existing in the answer. That's the one you can attack today, while the other five are still being laid.

Start there. The study tells you in what fraction of AI answers your vertical and city appear, and the living guides keep each vertical's figures current, which is where this article's numbers come from. If you want to understand why we wrote this and how we work, there's the manifesto and agent made. Meta's tomorrow is coming. Today's question can already be answered.