How AI Assistants Pick Which Local Businesses to Recommend

Here's a buying journey that barely existed two years ago. Someone in Hollister opens ChatGPT and types, "My laptop is running really slow. Is there someone local who does in-home computer repair?" The AI answers with three or four business names and a sentence about each. Only then does the person open Google or Yelp, look those businesses up, check the reviews, and call one.

If your business isn't one of those three or four names, you never make the comparison. The customer doesn't see a list of ten blue links with you at number seven. They see a shortlist, and you aren't on it.

We've already written about the SEO fundamentals that help AI assistants find your website. This post goes a level deeper. It covers what happens between the question and the shortlist, which signals the evidence says actually matter, what we found when we tested our own business, and how to measure where you stand.

AI Is Now the First Stop, Not the Last

The shift happened fast. BrightLocal's 2026 Local Consumer Review Survey (1,002 U.S. adults) found that 45% of consumers used AI tools for local business recommendations in the past year, up from 6% the year before. That makes AI the third most-used tool for local recommendations, behind Google and Facebook and ahead of Yelp and TripAdvisor. ChatGPT was the most-used AI tool for it (31%), followed by Google's AI Mode (23%) and then Gemini.

The same research shows how people use those answers. Among consumers who used AI for local recommendations, 63% trust what it recommends, and 97% sometimes double-check the recommendation against real reviews. In BrightLocal's words, AI has become "the first step in local business research rather than the final authority." That's the whole pattern: the AI builds the shortlist, and the customer verifies it.

Google itself is moving the same direction. Google says AI Overviews have more than 2.5 billion monthly active users and AI Mode has passed 1 billion. When those summaries appear, people click less. In Pew Research Center's analysis of real browsing data from March 2025, users who saw an AI summary clicked a traditional search result in 8% of visits, compared with 15% when there was no summary, and clicked a link inside the summary itself in just 1% of visits. Questions were especially likely to trigger a summary: 60% of searches starting with words like "who" or "what" produced one. "Who does web design in Gilroy?" is exactly that kind of search.

What Actually Happens When Someone Asks an AI for a Recommendation

A person running a Google search skims results and picks one. An AI assistant with web search does much more before it answers. Here's the process, based on what the platforms themselves publish.

1. It rewrites the question into searches. OpenAI's help documentation says ChatGPT search "typically rewrites your query into one or more targeted queries" before sending them to search providers, uses your approximate location from your IP address (so "good restaurants near me" becomes something like "top restaurants San Francisco"), and may send more specific follow-up queries after it reads the first results.

2. It fans out across subtopics. Google describes the same idea as "query fan-out." Its documentation for site owners says AI Overviews and AI Mode may issue "multiple related searches across subtopics and data sources" to build one answer. For harder questions, Google says AI Mode's Deep Search can "issue hundreds of searches." A question like "who's a good roofer in Salinas who handles insurance claims?" might turn into separate searches for roofers in Salinas, reviews, insurance experience, and pricing.

3. It reads many sources and synthesizes them. In Pew's data, 88% of Google's AI summaries cited three or more sources. For local businesses, those sources are often not your website. When BrightLocal ran 20 local searches across ChatGPT, Gemini, Perplexity, and Google AI Mode, it found that Yelp was used as a source in a third of all searches, niche industry directories showed up regularly, and businesses' own websites were also important sources.

4. It looks for agreement. BrightLocal puts it simply: AI "looks for patterns across multiple sources" rather than reading reviews the way a person does. A business that shows up with the same name, the same services, and the same city on several independent sites is easy to state as fact. A business that appears in one place, or appears with conflicting details, is easier to leave out.

5. It writes a short answer and names a few businesses. The output is a handful of names, not a ranked page of 10. Every step before this decides whether you're one of them.

What the Evidence Says AI Systems Weigh

There's a lot of confident advice about "ranking in ChatGPT." Most of it isn't backed by data. This is what is.

• Mentions of your business around the web. Ahrefs studied 75,000 brands and found that branded web mentions had the strongest correlation with appearing in Google's AI Overviews (0.664), far ahead of backlinks (0.218). The top three signals were all off-site. About 26% of the brands they checked had zero AI Overview mentions at all. That's a correlation, not proof of cause, but it points the same way as everything else here: AI visibility is mostly earned off your website.
• Reviews on more than one platform. BrightLocal's CEO points out that ChatGPT and other AI tools "can't see inside Google's walled garden of reviews. If your reputation only exists on Google, you are effectively invisible to the millions of people using ChatGPT to find local services."
• A website that's crawlable and says things plainly. Google says a page only needs to be indexed and eligible for a snippet to appear in AI Overviews or AI Mode, with "no additional technical requirements." Its listed best practices include keeping important content in text form, making sure structured data matches the visible text on the page, and keeping your Business Profile up to date. OpenAI says sites must allow its OAI-SearchBot crawler to be eligible for ChatGPT search results.
• Specific, verifiable content. The research paper that coined "generative engine optimization" tested rewrites of web pages against AI answer engines. Adding cited sources, quotations, and statistics raised a page's visibility in AI answers by roughly 30–40% on the researchers' main measure, while old-school keyword stuffing didn't help. The gains were largest for pages that ranked lower in normal search results, which is good news for small businesses. It's a lab benchmark, not a guarantee, but the direction is clear: concrete facts beat filler.

And what doesn't matter: at a Google Search Central event, Google's Gary Illyes reportedly said Google doesn't support llms.txt files and isn't planning to, and Google's documentation says "specific optimization isn't required" for its AI features. If someone sells you an "AI schema package" or an llms.txt file as the fix, be skeptical.

What We Found When We Audited Ourselves

Before offering this to clients, we ran our own business through our GEO audit tool (more on how it works below). We asked 20 real customer-style questions across five AI engines, three times each, for 300 answers in total. The results were humbling, and useful:

• GS46 Tech was mentioned in 11% of answers and recommended in 5%. Most of those mentions came from questions that already used our name. Of the 20 questions, 17, including "best web designer in Hollister CA," produced zero mentions on every engine and every repeat.
• The engines disagreed with each other. For "is there someone local in Hollister who does in-home computer repair?", one engine recommended us in all three runs and the other four didn't mention us at all. Results also shifted between runs of the same question on the same engine, which is why checking once tells you very little.
• AI got basic facts about us wrong 13 times. Engines confused "GS 46 Tech" with Genie GS-46 scissor lifts, a compressor oil, Goldman Sachs, and even the Hollister Co. clothing brand. If your business name overlaps with anything else, AI can mix you up with it, especially when the web doesn't clearly and consistently say who you are.
• Third-party sites fed the answers. Across all 300 answers, the most-cited site in our category was Yelp. Most of the other top sources were other local businesses' websites and directories like Thumbtack and Clutch. That tells you exactly where the AI is getting its shortlist.

Our website is built on the same SEO practices behind our local SEO results for clients. AI visibility still turned out to be a separate problem that needs its own measurements. It's worth checking yours, because Google Search Console won't tell you what ChatGPT or Perplexity says about your business.

Why Being the Name Matters So Much

A Google results page gives you ten chances plus a map pack. An AI answer usually names a few businesses, and most people who use AI for local recommendations trust it. They check reviews to confirm the pick, not to start over. Being named and holding up when they look you up is what gets the call.

The flip side is quieter. If an AI tells customers you're closed on Saturdays, don't serve their town, or are a different company entirely, you'll never know. The phone just doesn't ring.

What to Do About It: A Practical Checklist

1. Ask the AIs yourself, the right way. Don't search your own business name. Ask the questions a customer would ask, like "best [your service] in [your city]" or "who can fix [problem] near [town]?" Ask in ChatGPT, Gemini, Google, and Perplexity, repeat each question a few times, and write down who gets named and which sites get cited.
2. Fix your facts at the source. Make your name, address, phone, hours, and service area identical on your Google Business Profile, Yelp, and every directory in your industry. If your name collides with something else, describe what you do and where, in the same words, everywhere.
3. Get onto the pages the AI is already reading. Find the directories and "best of" lists that get cited for your category and city, then get listed or corrected on them. In our own audit, this landed near the top of the fix list.
4. Earn real mentions. Local news, your chamber of commerce, supplier and partner pages, sponsorships, and community groups. Web mentions like these were the signal most strongly correlated with AI Overview visibility in the Ahrefs study.
5. Build reviews beyond Google. Keep asking happy customers for Google reviews, but also point some of them to Yelp or the review site that matters in your industry.
6. Make your website state facts plainly. Give each service and city its own page. Put real details on them: what you do, where, typical pricing, and how to reach you. Keep it all in text, not images, and make sure your hosting doesn't block search crawlers. Our local SEO service covers this.
7. Skip the snake oil. Nobody can guarantee a ChatGPT recommendation, and the evidence says llms.txt files and keyword stuffing won't get you one.
8. Measure again every month. AI answers change from run to run and month to month. One screenshot isn't a strategy.

How the GS46 Tech GEO Audit Works

Checking all of that by hand is slow and unreliable, so we built a custom tool that does it systematically. Here's what our GEO AI Search Audit actually does:

• Real customer questions. We write a set of 20 questions built around your business, phrased the way buyers really ask. They include "best [service] in [city]" questions, problem-first questions ("my website doesn't show up on Google, who do I hire?"), comparisons, price and specialty questions, and ready-to-buy questions. At most two mention you by name, because a business that only appears when you name it isn't actually visible.
• The biggest AI models, with live web search. The tool sends every question to up to five AI engines, each with live web search turned on: OpenAI's GPT models (the company behind ChatGPT), Google's Gemini grounded in Google Search, Anthropic's Claude, Perplexity, and xAI's Grok.
• Three runs per question, per engine. AI answers aren't stable, so every result is reported as a rate ("named in 2 of 3 runs") rather than a one-time snapshot.
• Every answer scored. Each answer gets a 0–4 score: not mentioned, mentioned in passing, your website cited as a source, named in the recommendation, or the first or only pick. We also record the tone, your position in the list, every competitor named, and any factual error the AI states about you.
• A report you can act on. You get your presence, recommendation, citation, and share-of-voice rates, plus an overall AI Visibility Score. That score is our own composite, and the report says so. You also get the factual errors AI is telling customers, who's being recommended instead of you and how often, the third-party sites the AIs actually cited, an engine-by-engine breakdown, verbatim excerpts of what each engine said, and a prioritized fix list ranked by how strong the evidence is. It even includes a "don't bother" section for tactics the evidence says don't work.
• Honest about limits. The tool queries the models directly, which isn't identical to your customer's personal ChatGPT app with its memory and location settings. We say so in the report and hand-check your most important questions in the real consumer apps before we send it. The audit also includes a technical and on-page SEO review and a Google Business Profile review.

After the audit, our ongoing GEO AI Visibility Automation plan re-runs the same locked question set every month, so you can see whether the numbers are actually moving. See the GEO section of our pricing page for details on both.

Find Out What AI Is Saying About Your Business

Your customers are already asking AI who to call. The only question is whether it names you, names a competitor, or gets your details wrong. GS46 Tech is based in Hollister and works with businesses across Gilroy, Morgan Hill, Salinas, Monterey, and beyond. The GEO audit is done remotely, so location doesn't matter.

Call or text (831) 288-3299 or request your GEO audit here. We respond within 24 hours.