AI SEO: How to Get Found in Google AI Overviews and ChatGPT
Your customers are asking AI assistants for recommendations. Here is how those assistants decide which businesses to mention, and what your website needs so the answer is you.
For twenty years, getting found online meant one thing: ranking on the first page of Google. That still matters. But a growing share of your future customers will never see a list of ten blue links. They will ask a question and get an answer.
“Who’s a good landscaper near me?” typed into ChatGPT. “Best bridal hairstylist in Grand Rapids” asked of Gemini. A Google search that returns an AI Overview at the top, summarizing three businesses before the traditional results even begin.
In each case, an AI assistant decided which businesses to mention. This article is about how it decides, and what your website needs so that the answer includes you.
How AI answers differ from search rankings
A search engine ranks pages. An AI assistant composes an answer. That difference changes what your website has to do.
To rank a page, Google mostly needs to know the page is relevant and trustworthy. To recommend your business, an assistant needs to understand it: what you do, where, for whom, what you charge, how you are rated, and why it should believe any of that. It then needs to cite a source, because assistants are increasingly built to show where their answers came from.
So the question is no longer “does this page mention the right keywords?” It is “can a machine read this site and come away with a clear, confident description of the business?” Most small business sites fail that test, not because the business is unclear, but because the site was built for humans to look at and nothing else.
What AI assistants actually look for
Across the major assistants, the same signals keep showing up.
Clear, specific statements about the business. Your name, your services, and your location stated plainly in text, not buried in an image or implied by a tagline. If your homepage says “Solutions for your journey,” an assistant has no idea what you sell. If it says “Custom cabinetry for kitchens and baths in West Michigan,” it does.
Structured data, meaning schema markup. This is code behind the page that describes your business in a format machines are built to trust: business type, address, phone, hours, services, and aggregate rating. It is the single highest-leverage thing you can add, and it is invisible to visitors.
Consistency everywhere. Your name, address, and phone number should match exactly across your website, your Google Business Profile, and every directory that lists you. Assistants cross-reference. Inconsistency reads as unreliability.
Reviews with detail. Assistants lean heavily on reviews, especially ones that describe a specific situation and outcome. “Great service” tells an assistant nothing. “They rebuilt our booking flow and we picked up several new clients” is a sentence an assistant can use.
Dedicated pages for each service and location. One page that lists everything is harder to cite than one page per thing. If you serve two cities, have a page for each. If you offer five services, give each its own page with its own clear description.
Clean, semantic structure. Real headings, real lists, real sections. Assistants parse structure to understand hierarchy. A page built from generic boxes with styled text reads as noise.
The llms.txt file
There is one new piece worth explaining on its own. An llms.txt file is a plain-text document, placed at the root of your website, that describes your business and points to your most important pages, written specifically for AI assistants to read.
Think of it as the machine version of your elevator pitch. It tells an assistant, in a few paragraphs, who you are, what you offer, where you operate, and where to look for details. It is a young convention, but it is being adopted quickly, and Google now checks for it as one of the three criteria in its Agentic Browsing audit.
We add one to every site we build. It takes an afternoon to write well, and it may be the cheapest visibility improvement available right now.
From being read to being used
There is a second shift happening behind the first one. Assistants are becoming agents. Instead of telling someone your phone number, an agent will call it. Instead of pointing to your booking page, it will book the appointment. Instead of describing your contact form, it will fill it out.
For that to work, your site needs more than readable content. It needs to be operable by software. Google’s Agentic Browsing audit, now part of PageSpeed Insights, measures exactly this with three checks:
- A sound accessibility tree. The same structure that lets a screen reader navigate your site is what lets an agent navigate it. Buttons must be buttons, forms must be labeled, and navigation must be real navigation.
- Layout stability. If elements jump around while the page loads, an agent trying to click something will miss. Stable layouts are a speed metric and an agent metric at once.
- An llms.txt file. Covered above.
A site that passes all three shows a score of 3/3. Most sites built more than a year ago do not pass, and the fixes are usually structural. This is why we treat agent-readiness as a build requirement on every plan rather than an add-on: it cannot be retrofitted cheaply, and it will only matter more.
What to do this month
If you take one step, add schema markup and check that your business details match everywhere. If you take two, add an llms.txt file. If you take three, run your homepage through PageSpeed Insights, look at the Agentic Browsing score, and find out where you stand.
If the answer is “not great,” that is normal, and it is fixable. Every site we build launches with all of the above in place, because being readable by search engines was the last decade’s requirement. Being readable and usable by AI is this one’s.
