AI Search, Schema, and the New Three-Audience Web

Ben cohen

Think about how you bought software three years ago compared to today. Three years ago, you searched Google, clicked five links, opened ten tabs, right? Today, a B2B buyer goes straight to ChatGPT or Perplexity or Claude, types in give me a shortlist of the top three tools for automated billing, and they make a buying decision before ever clicking a single traditional search link.

Ido

Yeah, or, or even looking at a website at all in some cases. It's totally shifted.

Ben cohen

It's completely zero click. And here is the crazy thing, Ido. Most companies have no idea if they are even in the running for those recommendations.

Ido

Because, well, look at the web today. WordPress alone powers over 43 percent of all websites on the internet. Forty three percent! But most of those sites, they were built for a totally different era. They are packed with heavy page builders, like Elementor or Divi, generating mountains of code bloat, div inside div inside div. And, and half the time, the site owner doesn't even realize their security plugin is actively blocking GPTBot or ClaudeBot in their robots dot txt file!

Ben cohen

Right! You're literally slamming the front door shut on the very machines that are deciding whether to recommend you.

Ido

Exactly. If an AI engine cannot scan structured, extractable signals from your site, it just skips you. Not because your product is bad, but because you made it computationally expensive or literally impossible for the model to read.

Ben cohen

I love the analogy you used the other day for this. The, the restaurant analogy.

Ido

Oh, yeah! Explain that, because it makes it so intuitive.

Ben cohen

Okay, picture a physical restaurant. Traditional web design focuses entirely on the dining room. The lighting, the plush leather seats, the fancy artwork on the walls, the ambient music. That's your visual frontend, your JavaScript, your CSS. A human walk in and says, wow, beautiful place. But an LLM? An LLM doesn't sit in the dining room. An LLM is a rushed delivery driver who walks around back to the kitchen window and says, give me the plain text menu right now so I know what you serve.

Ido

Yes! Give me the raw ingredients and prices, no fluff.

Ben cohen

Exactly! That plain text menu is what assets like llms dot txt provide. It is a clean, low noise Markdown file that strips away all the visual layout, all the code bloat, and gives the language model pure structured context. It maximizes token efficiency.

Ido

Which means you're really designing for three entirely different audiences now when you build a website. You have audience one, the human visitor, who still wants the nice dining room and visual experience. You have audience two, the machine interpreter, tools like Perplexity or Gemini that process clean markdown and schema markup to answer questions. And then you have audience three coming fast, autonomous agents.

Ben cohen

Ah, the agents that don't just read the menu, they need to place the order.

Ido

Right, they need structured, predictable pathways to execute an action without getting confused by a pop up or a broken form.

Ben cohen

Which brings us to how we actually build for this new web foundation.

Ido

Yeah, how do we bridge that gap from crawler discovery to actual agent execution?

Ben cohen

So if you want machine interpreters to cite you, where do you actually start on the technical side?

Ido

It starts with schema markup. Schema is basically the explicit ID badge your content wears for AI models. Without it, the AI has to guess what your page is about. With it, it knows instantly.

Ben cohen

Like Article schema?

Ido

Article schema, yes, especially ensuring you include the dateModified field, not just datePublished. Freshness signals tell the LLM this information is actively maintained. Person and Author schema establish your E E A T credentials, showing who wrote it. But the real silver platter for GEO, for Generative Engine Optimization, is FAQ schema.

Ben cohen

Why FAQ schema specifically?

Ido

Because when you structure a clear question as an H2 heading and follow it immediately with a direct one to two sentence answer, you are handing the language model a pre packaged quote. It can literally lift that two sentence answer and drop it directly into a ChatGPT response with a citation link back to you.

Ben cohen

That makes total sense. And this is why traditional SEO tracking is starting to feel completely broken. Old school rank trackers tell you, hey, you're position number three for this keyword on Google. But that doesn't tell you if ChatGPT is recommending your software or your competitor's software when someone asks for a comparison.

Ido

Right, rank isn't really a static thing anymore with LLMs because generative answers are fluid and probabilistic.

Ben cohen

Exactly! It's stochastic. So instead of tracking traditional keyword ranks, we have to look at bot analytics and prompt tracking. That's why we're seeing tools like LLMagnet pop up. LLMagnet actually tracks real AI bot traffic visiting your site, measures your overall AI visibility score, and monitors specific prompt mentions across ChatGPT, Gemini, and Claude so you can see where your brand appears and what to optimize next.

Ido

That is huge. Being able to see which specific prompts include your brand versus where competitors are stealing the citation.

Ben cohen

It completely changes the game. But, okay, tracking citation visibility is step one. What about step two? You mentioned audience three earlier, the autonomous agents.

Ido

Yeah, this is where we move from reading to acting. Agent readiness means preparing your website so AI agents can perform tasks on behalf of users. If an agent comes to your site to book a demo or buy a product, it shouldn't get stuck because your submit button isn't labeled semantically or because your booking form is trapped inside a complex iframe without clear API endpoints.

Ben cohen

Clean HTML labels, unblocked form fields, simple lightweight integration hooks.

Ido

Simple, predictable structures. Make it frictionless for the machine to complete the workflow.

Ben cohen

So for anyone listening right now who wants to know where they stand today, what is the immediate action step?

Ido

Take fifteen minutes this week. Open ChatGPT, Gemini, and Perplexity. Type in five critical questions or buying prompts that your potential customers would ask when looking for a product like yours.

Ben cohen

See if you are cited, see who else is cited, and check if the AI is hallucinating or getting your messaging completely wrong.

Ido

Spot those citation gaps before your competitors do. Alright, that's the playbook for today.

Ben cohen

Go check your robots dot txt file and talk soon!