A customer who used to type “dog groomer near me” into Google now sometimes asks their phone a whole question: “Who’s a good dog groomer near me who can handle an anxious rescue, and roughly what does it cost?” They get back a paragraph, with two or three names in it.
You want to be one of those names, and you want what the paragraph says about you to be true. Nobody outside these companies knows exactly how those names are chosen. Some things are known, though, and the things that help are mostly things that were already worth doing.
One warning before anything else. This is the fastest-moving subject I write about. These tools change how they find and read information every few months. Treat what follows as the state of things in October 2026, and be suspicious of anyone, including me, who sounds certain about the details.
What these tools actually look at
When an assistant answers a question about local businesses, it isn’t working from one source. Broadly, it draws on some mix of these:
What it learned in training. Each model was trained on a large snapshot of the web, taken months or years before you’re talking to it. If your business was around and written about back then, the model may “know” of you. That knowledge is old and it can be wrong: an old address, a closed location, a service you stopped offering.
A live web search. For questions about the current world, most assistants now run a search and read the pages they find before answering. Some use their own search index, some use a partner’s, and Google’s AI answers sit on top of Google Search itself. Which company uses which index isn’t always public, and it changes.
Map and listing data. For “near me” questions, assistants often lean on business listings: Google Business Profile for Google’s products, and other listing and map data elsewhere. The exact sources vary by assistant and aren’t fully disclosed.
Reviews and what others say about you. Review sites, local news, directories, forum threads. An assistant summarizing “is this place any good?” reads the same kind of things a careful person would.
Your own website, when it’s fetched during that search, or when the person pastes your address in and asks about you.
So there is no single “AI listing” to fill in. There’s the web, read by a machine that is impatient in some ways and very literal in others. That’s why most of what helps is the same work that helps you on Google and with real customers.
On Google’s side, there’s an official answer worth knowing. Google’s guidance for its AI Overviews and AI Mode says there’s no special markup or file needed to appear in them, and that the same things that make a page good for Search apply. That’s Google only. The other companies have said less.
Plain text beats clever design
A person can look at your site and understand it from a photo, a layout, an icon. A machine reading your site mostly reads its text. If the important facts aren’t written down as text, they may as well not exist.
Common ways small business sites hide their own facts:
- Prices and menus as images or PDFs. The bakery’s price list is a photo of a chalkboard; the restaurant menu is a scanned PDF. Some tools can read these. Many readers won’t, and none of them will read them as reliably as plain text on a page.
- Hours only in a graphic, or only on your Facebook page.
- Text that appears only after the page’s scripts run. Some sites, particularly ones built with heavier tools, send an almost empty page and fill it in with JavaScript. Google’s crawler runs that code. Many AI crawlers, as far as is publicly known, read the page as it first arrives and don’t. They see the empty version.
- Key details inside tabs, sliders and pop-ups. Often fine, sometimes not, depending on how they’re built.
- Vague copy. “Your trusted partner for all your needs” tells a machine nothing. Nor does it tell a person anything.
Do this now, five minutes: open your home page on a laptop, right-click and choose “View page source” (or press Ctrl+U, Cmd+Option+U on a Mac). Press Ctrl+F (Cmd+F) and search for your phone number, then your town, then the name of your main service. If they’re not in that source, a crawler that doesn’t run scripts can’t see them either.
And write the specific thing, in a sentence, where a person would look for it:
Before: Premium grooming experiences tailored to your pet.
After: Dog grooming in Tempe, Arizona, by appointment. Full groom from $65 for small dogs, $95 for large. We take anxious and reactive dogs; first visits are quieter and longer, at no extra charge.
The second version answers the question the customer actually asked the assistant. That’s the whole idea.
Answer your customers’ questions, in your own words, on your own site
Think about the questions you get on the phone every week. Do you take walk-ins? Is there parking? Do you work on older houses? Do you do gluten-free? How long does a first visit take? What does it usually cost?
Those are, more or less, the questions people now type into an assistant, in full sentences. If the answer is written plainly on your site, a tool that reads your site can find it and use it. If it isn’t, the tool fills the gap from somewhere else, maybe a two-year-old review, maybe a competitor’s page, maybe a guess.
A questions page works. So do answers placed where they belong: parking on the contact page, prices on the service page, “we take anxious dogs” on the grooming page. Use your customers’ wording, not industry wording. People ask whether you’ll “fix a running toilet”, not whether you offer “residential fixture repair”.
Two cautions. Write answers that are true and that you’ll keep up to date; a wrong price on your own site is worse than none. And don’t pad. Twenty real questions answered in a sentence or two each are worth more than sixty invented ones written to fill a page.
Google stopped showing the expandable FAQ boxes in search results for most sites in 2023. That doesn’t make the questions less useful to the people, and the machines, reading your page.
Structured data: say the facts in a form machines don’t have to guess
Structured data is a small block of code on your page that states your business facts in a standard vocabulary called schema.org. Instead of a program having to work out that “Mon–Fri 8–5” is your opening hours, you say so in a format built for exactly that.
For a local business, the relevant type is LocalBusiness, or a more specific one if it exists for your trade (schema.org has types such as Bakery, Plumber and Dentist). A minimal version looks like this:
{
"@context": "https://schema.org",
"@type": "Bakery",
"name": "Rise & Shine Bakery",
"url": "https://example.com",
"telephone": "+1-480-555-0137",
"address": {
"@type": "PostalAddress",
"streetAddress": "12 Main St",
"addressLocality": "Tempe",
"addressRegion": "AZ",
"postalCode": "85281",
"addressCountry": "US"
},
"openingHoursSpecification": [
{
"@type": "OpeningHoursSpecification",
"dayOfWeek": ["Tuesday", "Wednesday", "Thursday", "Friday", "Saturday"],
"opens": "07:00",
"closes": "14:00"
}
],
"priceRange": "$"
}
It goes in the page inside a <script type="application/ld+json"> tag. Many website platforms and SEO plugins add a version of it for you; check that what they add is filled in and correct, not half-empty defaults. Google’s free Rich Results Test will show you what it finds on a page.
What it does, honestly: Google uses structured data to understand pages and says so in its documentation. Whether ChatGPT, Claude or Perplexity give it any special weight when they read a page isn’t something those companies have spelled out, as far as I know. It’s a reasonable bet that clear, standard, machine-readable facts help any machine that reads them. It’s not a proven lever, and nobody can honestly sell it to you as one.
Two rules regardless. The structured data must match what’s visible on the page; Google’s guidelines say not to mark up things visitors can’t see. And when your hours change, it has to change too. Stale structured data is just a wrong answer in a more official format.
Be the same everywhere
Since an assistant can draw on your website, your Google profile, review sites and directories in one answer, disagreement between them shows up directly. Old hours on Yelp, a former address in a directory, a phone number you stopped using. A person might notice the conflict and check. A summary might simply pick one.
So the dull local search work matters here too:
- Keep your Google Business Profile complete and current. It feeds Google’s own AI answers, and it’s the listing people check first.
- Claim your free listings on Apple Business Connect and Bing Places, and fix your name, address, phone and hours on Yelp, Facebook and the one or two directories that matter in your trade.
- Use one exact version of your name, address and phone number everywhere.
Reviews matter in a particular way. When someone asks an assistant whether a place is good, the answer is often a summary of what reviewers say. Reviews that mention specific things (“they were patient with my anxious terrier”, “fixed it the same afternoon”) give a summary something real to say. You can’t write your reviews, and you mustn’t pay for them. You can ask happy customers to leave one, and you can reply to the bad ones calmly, because the replies get read too.
Don’t lock the door by accident
The big AI companies publish the names of the crawlers they use, and most now separate the one that collects material for training from the one that fetches pages to answer a live question. OpenAI, for example, lists GPTBot for training and OAI-SearchBot for search. You can allow or block each one in a file on your site called robots.txt.
Whether to let your content be used for training is your call, and there are fair reasons either way. But know what you’ve chosen. Some hosting and security services now block AI crawlers by default or offer a one-click switch to, and a business that blocks everything may also be blocking the crawlers that would have read its hours and prices to answer a customer. If you don’t know what your robots.txt or your host’s bot settings say, ask whoever runs your site. It’s a quick answer.
You may also hear about llms.txt, a proposed file that gives AI tools a plain-text summary of your site. It’s cheap to add. As far as I know, none of the major assistants has said it relies on one, so don’t pay anyone extra for it.
Some visitors now aren’t people
There’s a newer change, and it’s moving fast. Several of the big assistants can now open a browser and use websites on someone’s behalf: compare three plumbers’ prices, check whether a class has space on Thursday, fill in a booking form. Their owner asks for the outcome and the agent goes and gets it.
How much this happens in your trade today, I don’t know, and I don’t think anyone outside these companies does. It’s early. But you can see what these agents need, because it’s what a careful, literal, slightly impatient visitor needs:
- Prices and availability written on the page, not “call for a quote” on everything. An agent can’t phone you. It may report that it couldn’t find a price and suggest someone who listed one.
- Buttons and form fields with real labels. “Book a cleaning” rather than an icon; a field labeled “Phone number” rather than an empty box. Some of these agents lean on the same page structure a screen reader uses, so a site built properly for blind visitors is already most of the way there.
- A way to book or ask that doesn’t need an app download or a login.
This is also why some sites now offer a stripped-down version of each page for software to read. davis9001.com, for example, serves its pages as plain Markdown to agents that ask for it. That’s a nice extra, not a requirement. The requirement is that your normal pages say the important things in plain text.
See what the assistants say about you now
Do this now, ten minutes: open two or three assistants (whichever of ChatGPT, Claude, Gemini and Google’s AI answers you or your customers use) and ask the questions your customers would. Not your business name first; the question:
What’s a good [your trade] in [your town] for [the thing customers most often want]?
Then ask about you directly:
What are [your business name]’s hours, prices and services? Where did you find that?
Write down what’s wrong, and where it says it got it. A wrong answer is usually traceable to an old listing, a stale page, or a gap you haven’t filled. Fix the source, not the symptom. Then ask again in a month. The answers change, which is annoying, and also means a fix can show up.
Don’t panic if you’re not mentioned at all. Answers vary from one try to the next and from one person to another. Look for patterns over several questions, not a single result.
Honest closing
There’s a growing industry selling “AI search optimization” with confidence it hasn’t earned. Some of what it sells is reasonable work with a new label. Some of it is guesswork about systems the sellers can’t see inside. If someone promises to get you recommended by ChatGPT, ask them how they know, and listen for a real answer.
What I’m confident of is narrower. These tools read text. They cross-check sources. They repeat what reviewers say. So write down, plainly and on your own site, what you do, where, for whom, what it costs and when you’re open. Keep it the same everywhere. Answer the questions your customers ask. Keep your listings accurate and your reviews honest.
That would have been good advice in 2015 for a different reason. It will probably still be good advice after the next five changes to how these assistants work.