TLDR
For twenty-five years, a website had one audience: a person, with eyes, reading a screen. That’s changing in two ways at once. First, AI answer engines now read your site to answer someone else’s question, like “who’s the best roofer near me,” and the person asking may never click through. Second, AI agents with their own browsers are starting to visit sites on someone’s behalf to compare, book, or buy. Neither one watches your hero animation. Neither one is moved by your urgency banner. This is what they actually see, what stops them, what the early numbers say, and the checklist I’m now using to build sites for both audiences.
Two New Kinds of Visitors
Picture three visitors hitting the same service business website.
The person. They scroll, they look at photos, they get a feel for you, and maybe they call. Everything about web design for the last two decades was built for them.
The answer engine. Someone asks ChatGPT, Gemini, or Perplexity for a recommendation. The AI reads what it can find about businesses like yours, forms an answer, and may name you, or may name someone else. The person asking might never visit your site at all. The AI visited for them.
The agent. Someone tells an AI assistant, “find me three electricians who can come this week and book the one with the best reviews.” The assistant opens a browser, visits your site, looks for availability and a way to book, and either gets it done or moves on to the next site.
The first visitor is who you designed for. The second is already here. The third is coming.
The First One Is Already Here
Here’s a number that should get every local business owner’s attention. SOCi’s 2026 Local Visibility Index, covering about 350,000 business locations, found that Google’s local 3-pack surfaced businesses 35.9% of the time. Gemini, 11%. Perplexity, 7.4%. ChatGPT, 1.2%.
In other words, being recommended by an AI assistant is roughly 30 times harder than showing up in Google’s map results. That gap is also the opportunity: most of your competitors aren’t there either.
I’ve seen it move in real life. When I rebuilt the Oregon Ghost Conference platform, the new site went from appearing in 0 of 20 AI-answer checks to 20 of 20 between the July and August audits. It wasn’t a trick. The big change in that window was the real domain, with fourteen years of history and press behind it, moving onto a fast, clean, well-structured site.
And here’s what didn’t matter: special files. Google’s own documentation on its AI search features says no special files or markup are needed to appear in them. I keep an llms.txt on this site, a plain-text summary written for AI crawlers, because it costs nothing. But I treat it as a courtesy, not a ranking trick, and I’d be skeptical of anyone selling you one as a growth strategy.
The Second One Is Coming
The agent is the stranger visitor, and I’ve been building them, so I can tell you how they behave from the inside.
One of my own agents has a browser. It runs in its own isolated Chrome profile and does tasks I hand it. And it has house rules written into every job: no purchases, no payment details, no sending anything unless told, and stop at login walls and captchas. I wrote about what six months of running agents like that taught me.
Those rules exist for good reasons. A responsible agent should stop at a login wall. It should stop at a captcha. It should treat anything written on your page as information, not as instructions to follow.
Now flip that around and look at your own site from the agent’s side. Every login wall in front of a price, every captcha in front of a booking form, every “call us for pricing” is a place where a well-behaved agent stops and goes to the next business.
I don’t know how fast agent traffic grows. Nobody does. But the direction is clear enough that I’m building for it now.
What an Agent Actually Sees
An agent doesn’t see your site the way a person does. It reads the page’s structure and text: headings, labels, buttons, links, forms.
Here’s what helps it:
- Real text. Your services, your service area, your prices or price ranges, your hours, written as words on the page.
- Labeled forms. A field labeled “Phone number” is obvious. An unlabeled box with a gray placeholder that disappears when you click it is a guess.
- Clear buttons. “Book a free estimate” tells an agent exactly what happens. A button that says “Let’s go!” doesn’t.
- A booking path without walls. If a person can book without creating an account, an agent can too.
- Structured data that matches the page. Schema markup that says the same thing the visible page says, not a different, keyword-stuffed version.
Here’s what stops it:
- Prices only inside images, PDFs, or “contact us for a quote”
- Critical information only inside a video or an animation
- Pop-ups that cover the page and have to be dismissed in a particular way
- Booking that requires an account before you can even see availability
- Forms that only work with one specific sequence of clicks
Notice something about both lists. Almost everything that helps an agent also helps a person using a screen reader, a person on a slow phone connection, and a person who’s in a hurry. Accessible, clear, honest pages were always good design. They’re about to become necessary design.
Designing for Two Audiences
Here’s the tension. A lot of “conversion design” was built to work on human psychology: urgency timers, gated pricing, exit pop-ups, forms that make you commit before you see the price. Some of that works on people. Almost none of it works on an agent, and a lot of it actively blocks one.
My approach now is to split the job:
- For the person: emotion, trust, proof, story, a reason to choose you. Photos, reviews, a real face, a clear promise.
- For the agent: facts, structure, and a clear path. What you do, where, for how much, when you’re available, and how to book.
They conflict less than you’d think. The fastest way to satisfy both is honesty: tell the truth plainly, put it where it can be read, and make the next step obvious. A site that’s clear enough for an agent is almost always clearer for a person too.
What I’m Changing in My Own Builds
This is the checklist I’m using on every site now:
- Prices or price ranges written as text. Even a range beats “call for pricing.”
- Every service described in plain words. What it is, who it’s for, where you offer it.
- Hours, service area, and contact details as text, not only in an image or a map.
- Every form field labeled. Real labels, not just placeholder text.
- Buttons that say exactly what they do.
- Booking without an account. Let people, and agents, see availability and book.
- No critical info locked in images, video, or PDFs.
- Structured data that matches the visible page. Business details, services, reviews, FAQs.
- Fast pages. Agents have time limits too.
- Answers first. FAQ sections that give the answer in the first sentence, not after three paragraphs of story.
And one test you can run today: ask an AI assistant with browsing to find your price for your most common service and tell you how to book it. Watch where it gets stuck. That’s where a real customer’s agent will get stuck too.
The Bet, and What I Don’t Know
I want to be clear about what’s a fact here and what’s a bet.
The facts: AI assistants are already recommending businesses, and they recommend far fewer businesses than Google’s map results show. Moving a site’s real history onto a clean, fast structure moved one client from 0 of 20 AI-answer checks to 20 of 20 within a month. And agents with browsers exist and behave the way I described, because I’ve built them.
The bet: that within the next few years, a real share of first visits to service businesses will be an agent doing research for a person. I don’t know how big that share gets, or how fast. Nobody does, and I’d be suspicious of anyone who claims to.
But the cost of preparing is low, and most of it is just good design: clear, honest, fast, accessible pages. If the bet is wrong, you end up with a better website for people. If it’s right, you’re already there when the agents arrive.
I appreciate you reading this one. If you want to know how your site looks to an AI, the AI Visibility Check is a good start, and if you want a site built for both audiences, book a call from the home page.
Keep reading
- I Taught My AI to Forget (Why Memory Needs a Garbage Collector)Everyone's racing to give their AI more memory. Mine has about 700 memories, and the most useful thing I built this year was a way for it to forget. Every memory now has a birth date, an expiration date, a source, and a confidence level, and on the day I checked, only about one in nine made it into the room. Here's why forgetting is the next frontier for AI that works with you for years.
- Your Screen Is the New API (Every App You Own Just Became Automatable)For thirty years the rule was simple: no API, no automation. This year I watched an AI mix a record by clicking through Pro Tools, an agent run errands in its own browser, and a video editor build timelines from a script. The screen became the integration. Here's what that changes, why it's slower and stranger than the demos, and the rules that make it safe.
- Six Months of AI Agents That Never Sleep (And Why I Turned Most of Their Updates Off)In March I built an AI agent that runs 24/7, dreams at night, and sends me briefs every morning. By September one agent had sent me the same two suggestions five times in ten days, and I'd accepted zero of them. So I shut off every scheduled update, kept every real alert, and gave one agent hands instead. Here's what always-on AI actually looks like six months in.