How AI Assistants Decide Which Local Business to Recommend

Ask ChatGPT for “a reliable AC repair company in Boca Raton that can come today” and you’ll get a short list. Usually three to five names, each with a line or two about why.

Nobody on that list paid to be there. Most of them didn’t do anything on purpose to get there, either. They made the cut because the assistant found enough evidence about them, in enough places, and it all lined up.

That last bit is really the whole story. When an assistant answers a local question, it isn’t ranking pages the way the old ten blue links did. It’s putting together an answer and trying hard not to get it wrong. So anything that makes your business easy to describe and easy to double-check works in your favor. Anything confusing or contradictory works against you, and usually nobody notices it happening.

What happens between the question and the answer

Each platform does this a little differently, and they change things all the time. But the rough sequence holds up well enough to plan around.

1. It figures out what’s being asked. “Can come today” means it’s urgent. “Boca Raton” pins down the place. “Reliable” tells it to go looking for signs of trust, which mostly means reviews and reputation.
2. It pulls a set of candidates. Depending on the product, that comes from a web search index, a maps and places database, or both. Google’s AI features use Google’s own index plus Business Profile data. ChatGPT search uses its own crawler and partner data. Perplexity cites the pages it reads.
3. It reads around and compares notes. Your website, your Google Business Profile, review sites, directories, local news, trade associations. It wants to see the same business name, the same services, the same service area and the same phone number everywhere it looks.
4. It writes the answer, often with sources attached. The businesses that make it are the ones it can describe with confidence in a sentence or two.

None of these companies publishes a formula for this. If someone tells you they know the formula, they’re guessing. What the platforms do say in their own documentation all points the same way. Let their crawlers in. Say plainly what you do. Keep your facts straight, and have other sources back you up.

How the big assistants differ

They share the same basics, but each one leans on its own sources. It helps to know roughly where each one looks.

• Google AI Mode and AI Overviews run on top of Google Search. They pull from Google’s index and from Business Profile, Maps and review data. That’s why a complete, accurate profile matters so much for local questions.
• ChatGPT searches the web with its own search crawler and partner data when a question needs current information, and it shows you the sources. If your site blocks that crawler, you can’t show up in those results.
• Claude can search the web and read pages directly when that feature is turned on, and it usually cites the pages it used. Clear, well-organized pages are easier for it to quote correctly.
• Perplexity is built around search and citations. It likes pages that answer a specific question directly and are easy to credit.

Good news here. You don’t need a separate plan for each one. A clear website, a solid profile, reviews with some detail in them and a few outside mentions will cover all of them at once.

The four kinds of evidence that count most

1. A website a machine can actually read

Assistants work with text. If your home page is a big video, a slogan and a phone number, they’ve got almost nothing to go on.

The businesses that get described well usually have a separate page for each main service. Plain language. What the service is, who it’s for, where you offer it and what the job usually involves. A catch-all page that crams twenty services into one paragraph is hard to quote, so it tends to get skipped.

Two technical things matter more than most owners expect. First, your content needs to be in the page’s HTML itself. If it only shows up after some script runs, a crawler may never see it. Second, your site shouldn’t be blocking the crawlers these assistants use. OpenAI, for example, says in its documentation that sites blocking its search crawler can’t appear in ChatGPT search results. A lot of small business sites block AI crawlers by accident. Usually it’s a security plugin or a firewall setting nobody ever looked at.

2. A Google Business Profile that matches everything else

For local questions, your profile is often the most important record of your business out there. It feeds Maps and Google’s AI features both. Your categories, your listed services, your hours, your service area and your description all get read.

Say your profile says you serve all of Palm Beach County, and your website only ever mentions Boca Raton. That’s a small contradiction. One of those is no big deal. They add up, though.

3. Reviews that talk about the actual work

Star ratings matter. To an assistant, the words in the review matter more.

“Great service!” tells a model next to nothing. “They replaced the water heater the same afternoon and hauled the old one away” tells it you do water heaters, you do same-day work and you handle removal. Getting customers in the habit of mentioning the specific job, in their own words, is one of the most useful things a service business can do. It costs you nothing but the ask.

4. Other people talking about you

This is the one most local businesses underrate. Assistants look for backup from somewhere other than you. A listing on a trade association site. A mention in a local news story. A spot in a reputable directory, a manufacturer’s certified installer page, a chamber of commerce membership.

Each of those is an independent source confirming you exist and do the work you say you do, where you say you do it. A business with a beautiful website and zero outside mentions is basically asking the model to take its word for everything. Models aren’t great at that.

Why good businesses get left out

When I look at a business that should be showing up and isn’t, it’s rarely one missing trick. Usually it’s something on this list.

• The facts don’t match. An old address on some directory, a former business name on a review site, a phone number that changed two years ago. Every one of those makes the model a little less sure it’s looking at one current business.
• The services are vague. If your site never actually says “tankless water heater installation,” the assistant has no evidence you do it. Doesn’t matter how many you put in last month.
• The location signals are thin. You serve five cities and only name one of them anywhere.
• Nobody else mentions you. Reviews on one platform and that’s it.
• You’re blocking the door without knowing it. A security setting turns crawlers away, or the important stuff is stuck inside a script or a PDF.

The order I’d work in

If I were starting from scratch with a local service business, this is the order I’d go in. Each step makes the next one work better.

1. Get your facts straight. Pick the one correct version of your name, address, phone, hours and service area. Then make every profile and directory match it.
2. Open the doors. Check your robots.txt and firewall settings so search and AI crawlers can read the site. Make sure the important content is plain HTML.
3. Give every service its own page. Plain language, specific, and answer the questions customers actually ask right there on the page.
4. Be honest about your service area. List the cities you really serve, on the site and on the profile, using the same wording in both places.
5. Ask for reviews with detail. Make the ask specific. Something like “It really helps other homeowners if you mention what got fixed.” Put it in your own words.
6. Earn some outside mentions. Associations, supplier directories, local press, community sponsorships. It’s slow and it builds on itself, and a competitor can’t copy it overnight.
7. Track it with the same questions every time. Ask the same buying questions on the same platforms on a regular schedule. Write down who gets named and which sources get cited. Watch the trend over months and ignore any single day.

What to expect, honestly

Nobody can promise that a particular assistant will name your business for a particular question. The answers shift by platform, by wording, by location, even by day. Anyone guaranteeing otherwise is selling something.

What you can control is the evidence. How good it is, and whether it all agrees. In my experience that work pays off twice. The same fixes that make you easier for an AI assistant to recommend also help you show up in regular Google results and on Maps. And they make you easier to trust once a real person lands on your site.

Curious how your own business looks to these systems? The free AI Visibility Check scores the parts you control and explains every gap it finds.