Zero-Click Buying and the Three-Audience Web

Ben cohen

So I, I, I was talking to a software buyer last week, and they told me something that, that honestly kind of blew my mind. They picked their last three enterprise vendors, like, shorted them, evaluated them, completely decided on them, inside ChatGPT and Perplexity. They never, not once, clicked through to a company homepage until it was time to request a demo.

Ido

Zero click buying. I mean, it's, it's totally shifting how software is bought, right? People aren't using search engines to find links anymore. They're using AI as an answer engine to make decisions.

Ben cohen

Yeah, exactly. And if your site isn't readable to those AI engines, you aren't even on the menu. You're just, uh, completely invisible.

Ido

Which is, okay, think about it like a physical restaurant. You build this gorgeous dining room with fancy lighting and nice music for human guests. That's your visual website. But when GPTBot or ClaudeBot shows up at the back door, they don't care about the lighting. They just want a clean, structured menu. If you hand them a tangled mess of JavaScript token bloat and custom styling, they just walk away.

Ben cohen

JavaScript token bloat, right! Because LLMs pay for tokens. Every extra layer of div tags and layout code is literally just noise to them.

Ido

Exactly. We have to start building for three completely different web audiences now. You've got human visitors who need visual trust, machine interpreters parsing explicit schema, and now autonomous agents that need to actually navigate your site architecture to perform tasks.

Ben cohen

Wait, wait, go back to that second one. Machine interpreters. How big of a difference does it actually make if you strip away the design and give them plain structured markdown?

Ido

I, I actually tested this on a site I was working on. It was built with a heavy page builder, tons of nested code. Claude kept getting the product specs wrong when summarizing it. So I created a clean markdown endpoint, like a stripped down raw text version. The moment Claude read that instead, the accuracy jumped to nearly one hundred percent. The summary completely transformed overnight.

Ben cohen

That is wild. One hundred percent accuracy just by changing the format it reads.

Ben cohen

So if someone is listening right now and thinking, okay, how do I actually fix this for my site today? Where do they start? I usually tell people the first step is just auditing how AI sees you right now. See what prompts mention your brand, track where you're getting cited, and see if AI bots are even crawling your pages.

Ido

Right, and you need actual visibility into that bot traffic, not just guessing based on standard Google Analytics, because standard analytics misses half of these crawlers.

Ben cohen

Yeah, that's why tools like LLMagnet have become so useful lately. LLMagnet helps you understand how large language models read, rank, and connect with your content. It tracks real AI bot visits, shows you which pages they hit, and helps generate files like llms.txt automatically.

Ido

Though, okay, I have to jump in with a warning on llms.txt. A lot of people are treating llms.txt like it's some magic silver bullet, like the old meta keywords tag from twenty years ago. They throw up a clean text index, but then their on page JSON LD schema is full of duplicate entities or outdated product prices.

Ben cohen

Ah, so if the underlying data on the page is messy, a nice directory file isn't going to save you.

Ido

It won't! If the model follows your llms.txt link to a page with conflicting schema, it gets confused and drops your confidence score. Clean index, dirty content, still fails.

Ben cohen

And that gets even worse as we move into autonomous agents, right? It is one thing for an AI to cite your blog post in a search summary. But what happens when an AI agent is trying to book a demo or configure a product on your site for a user?

Ido

Oh, it breaks completely. If your form fields aren't clearly labeled or if your action buttons rely on complex click handlers without clear text labels, the agent hits a brick wall.

Ben cohen

Right! You can be perfectly readable to a search model, but completely unusable to an action agent. That's the next big wall people are going to hit.

Ido

So here is the assignment for anyone listening today. Take your main brand value proposition or your top product query, and run it across three different engines right now. Run it in ChatGPT, run it in Perplexity, and run it in Gemini.

Ben cohen

See who gets cited, see how your product is described, and see where the gaps are. Alright, that's it for today. Talk soon!