The way people find a plumber, a clinic, a consultant, or a contractor is quietly splitting in two. Some still type into Google and click a blue link. More and more are asking an assistant — ChatGPT, Google's AI results, Perplexity, the assistant on their phone — to do the finding for them: "who does emergency HVAC near me, and can you line up a quote?"

That means your website now has a second reader. Not just the human skimming it on their phone, but the AI reading it on the human's behalf — deciding whether to recommend you and, increasingly, whether it can act for them. Here's what that reader actually needs, and, just as important, what it doesn't.

Not magic. Legibility.

First, the un-hype

There is no secret "AI SEO," and anyone selling you an "AI ranking boost" is selling theater. Google's AI answers draw from ordinary search eligibility. The assistants pull from the open web and from structured data. There's no magic token you sprinkle on a page.

What actually helps is boring and durable: be clear, be structured, be reliable. An AI reading your site for a customer is trying to answer a short list of questions, and it can only answer them if your site says them plainly. So the work isn't chasing an algorithm. It's making the truth about your business legible — to a person in a hurry and to a machine reading on their behalf.

The machine is not looking for poetry. It is looking for something it can trust enough to repeat.

The six questions an assistant is trying to answer

When someone asks an assistant to find a local service, the assistant is quietly checking whether your site answers:

  1. What do you do? In plain words, not a slogan.
  2. Where do you do it? The actual service area, named.
  3. Who is it for? The kind of customer and job you take.
  4. What does it cost? Even a range, or a starting point.
  5. How does someone start? The next step, spelled out.
  6. Can I act here safely? Is there a real way to submit a request or book — and will it actually go through?

Most small-business sites answer one or two of these and bury the rest in a hero video, a PDF, or an About page written like a wedding toast. A human will dig for it. An assistant won't — it'll just recommend the competitor whose site said it out loud.

What makes a site legible to both

None of this is exotic. It's the same handful of things that make a site good for humans, done on purpose:

  • Say the literal facts on the page. Services, service area, a price signal, how to start — as text, near the top, not trapped inside an image. If a person can copy-paste the answer, an assistant can quote it.
  • Add structured data (JSON-LD). LocalBusiness, Service, and FAQ markup are the machine-readable version of those facts. It doesn't change what humans see; it hands the same truth to machines in a format they don't have to guess at. One rule: only mark up what's actually visible and true. Schema for things that aren't on the page is how you get penalized, not promoted.
  • Be crawlable and fast. Clean HTML, a real sitemap, and an llms.txt that points to your key pages. Static, fast-loading pages aren't just nicer for humans — they're what crawlers can actually read.
  • Choose your crawler policy on purpose. You get to decide which AI crawlers may read your site for search and answers. Allowing the search-oriented ones (like OpenAI's OAI-SearchBot) is a legitimate, configurable choice — not a growth hack, just a decision worth making rather than accepting by default.

The part everyone forgets: reliability

Here's the piece missing from every "optimize for AI" article: an assistant that recommends you can also send you a lead — and if your form silently fails, you lose an AI-referred customer exactly the way you lose a human one.

Picture it. Someone asks their assistant to line up three quotes. The assistant reads your site, likes what it sees, and submits the quote request through your contact form. If that form drops it — a dead inbox, a broken plugin, an email that lands in spam — you didn't just lose a lead. You lost one that an AI actively chose to hand you, and you'll never know it happened.

Getting found by the machine is worthless if you can't reliably catch what it sends. Which is why the unglamorous stuff — saving every submission to a database you own, before anything else can lose it — matters more in an AI buying journey, not less. (I wrote about exactly how forms fail here.)

Readable is nice. Catchable is better.

The forward edge — without the hype

The step past "readable" is "actionable": websites an assistant can do something with — submit a structured quote request, check availability, start a booking — through a defined, safe endpoint instead of guessing at a form. The plumbing for this is real and emerging; the Model Context Protocol, the standard behind a lot of assistant tooling, is a genuine, supported thing. Most websites can't do any of it yet.

I'm not going to tell you that you need an agent-action layer today. You probably don't. But here's the honest edge: the sites that will be able to add one without a painful rebuild are the ones already built on real foundations — a database you own, reliable capture, machine-readable service data. If you're building or rebuilding now, that foundation is cheap insurance against a trend clearly heading this way.

What to actually do

You don't need to chase every AI headline. You need a site that:

  • says plainly what you do, where, for whom, roughly what it costs, and how to start;
  • backs those facts with honest structured data;
  • loads fast and lets the crawlers you choose read it;
  • and, above all, reliably captures the lead — whether a human or an assistant sends it.

Do those, and you're legible to the AI buying journey and the human one at once, because it turns out they want the same things: clarity, and a contact form that actually works.

If you want to know how your current site scores on any of that, send it to me and I'll take a look — free, no pitch.