What gets a local business recommended by AI

In short, if you run a local business

I asked ChatGPT, Gemini, Perplexity and Google’s AI Overviews to recommend a professional in 20 different trades, from dentists to plumbers, using Catania, in Sicily, as the sample city. Then I compared the businesses they named with the top 20 Google Maps listings for the same searches.

The main idea that comes out of the data is simple: AI hasn’t retired local SEO, it runs on it. To decide whom to recommend, AI assistants read the Google Business Profile, the reviews, the industry directories and the business website. Five things to know:

1. Every AI makes a short list: 3.9 names per answer on average. Google Maps shows you twenty listings; a chat answer shows you four.
2. Having your own website roughly doubles the odds of being named, even with the same reviews and rating. A Facebook or Instagram page instead of a website helps much less.
3. Reviews and rating work as a filter. Below 10 reviews, 11% of businesses get named; above 200, 44%. With a rating under 4.8, the odds halve.
4. Industry directories carry a lot of weight. In Italy that means MioDottore (the local Doctolib) for health, Houzz for architects, Matrimonio.com for wedding photographers, Fresha and Treatwell for hair and beauty.
5. Answers change often. Three in four businesses that get named appear on only one engine, and the same question asked a few hours later can return different names.

If you want to skip to the practical part, the checklist is in section 9. Below is the full study, with method, numbers and a comparison with the research published so far.

1. Why I ran this study

In the United States, 45% of consumers used ChatGPT or another AI tool to find a local business in the past year, up from 6% the year before (BrightLocal, Local Consumer Review Survey 2026). For anyone working in a city, that means part of their customers no longer scroll through Google results: they ask an assistant and choose from a list of a few names.

The studies published so far on how AI picks local businesses come almost entirely from the US and often cover chains and multi-location brands. I found none about Italy, and none about the independent professionals who make up most local businesses here. This study tries to fill part of that gap, with one clear limit: the data comes from a single city.

I work from Catania, building websites and doing local SEO mostly for local businesses and professionals. Over the past few months a new question has become one of the ones I hear most: “how do I show up when someone asks ChatGPT?”. I had impressions, not numbers. I chose to measure, and to publish data and method so that anyone can check the maths.

2. How I collected the data

I picked 20 trades: lawyer, accountant, notary, dentist, psychologist, physiotherapist, nutritionist, vet, architect, building contractor, electrician, plumber, estate agent, restaurant, photographer, hairdresser, beauty salon, gym, driving school, mechanic.

For each one I wrote two questions, phrased the way a person would ask them (in Italian):

• a generic one, e.g. “Can you recommend a good dentist in Catania?”;
• one with a specific need, e.g. “I’m looking for a dentist in Catania for an implant: who would you recommend?”.

The 40 questions went to four engines: ChatGPT and Gemini in their consumer versions, AI Overviews on Google’s desktop results page, and Perplexity with the sonar model. User location Italy, Italian language, one answer per question and engine. 160 requests in total, collected on October 15 and 16, 2026 through DataForSEO. 141 came back with text: ChatGPT, Gemini and Perplexity always answered, AI Overviews only in 21 cases out of 40 (I explain why further down).

To compare the answers with Google Maps, I collected the top 20 listings for two searches per trade (“dentist catania”, “dentist implants catania” and so on): 621 distinct listings. For each listing I looked at reviews, rating, photos, categories, whether it is claimed, whether it shows a public address and which website it links to. I then opened each of the 379 linked websites’ homepages once (340 responded) to check whether they link to social profiles, have a blog and use structured data for local businesses.

I counted the names by reading the answers one by one, and matched them to the Maps listings by hand. The question design follows a 2025 BrightLocal study (20 searches, 10 industries, 4 engines), so the results can be compared (BrightLocal, July 2025).

3. How many names AI gives, and where they come from

On average each answer names 3.9 businesses: 3.7 for ChatGPT and Gemini, 3.8 for AI Overviews, 4.4 for Perplexity. Only 2 answers out of 141 name nobody. These figures are close to BrightLocal’s measurements across 200,000 US searches: 4.1 for ChatGPT, 3.5 for AI Mode, 2.5 for AI Overviews (BrightLocal, September 2026).

Across all answers, 212 different websites are cited as sources. Directories and listing platforms are the largest group: 89 domains and 305 citations. Business websites account for 91 domains and 118 citations. Google Maps appears in 46 answers, professional bodies and public institutions in 14, social networks in 13, newspapers, guides and blogs in 18.

Each engine has its own habits:

Percentages of each engine’s answers. Gemini relies mostly on Google Maps data. Perplexity always leans on directories. AI Overviews mix directories and website pages. ChatGPT sits in between and often shows the rating and review count next to each name. To be found by all of them you need several sources at once: the Google Business Profile, the right directory and a website AI can read.

Each industry also has its reference platform (Italian platforms; the closest international equivalent in brackets where useful):

4. A website matters more than social media

This is the result that surprised me most, because of how clear it is. Out of 621 Maps listings, here is how many get named by at least one AI depending on what they link to as a website:

The other 40 listings link to a directory profile or a chain’s website.

Businesses with a website also tend to have more reviews, so the raw numbers could mislead. To separate the two effects I used a statistical model (a logistic regression) that holds reviews, rating and trade constant. The result holds: with the same reviews, rating and trade, having your own website multiplies the odds of being named by about 2.3 (odds ratio 2.26, 95% confidence interval 1.36 to 3.76). For a social page only, the estimated effect is smaller and not distinguishable from chance (1.76, 0.64 to 4.87), partly because that group is small.

BrightLocal’s September 2026 study reaches the same conclusion at a much larger scale: 93% of the domains AI uses as sources for local searches are business websites, accounting for 42% of all citations. That is notable because outside local search the opposite holds. In Semrush’s study of finance, technology and fashion, AI mostly cites Wikipedia, Reddit and review sites, and company websites lag behind (Semrush, September 2025). For a local business the website carries more weight than for a company selling online.

Within the website, what makes a difference? I looked at four things on the 313 business websites I could read:

Links to social profiles add nothing. The blog points the right way, but I need more data to say so with confidence: I’ll re-check it with the second collection. Structured data is the strongest technical signal, but only 36 sites use it: I read it as a sign of a site looked after by someone who does SEO rather than as a lever that works on its own.

5. Reviews and rating act as a filter

Listings named by AI have a median of 71 reviews; ignored ones have 21. The chance of being named grows with the number of reviews (see the first chart above): 11% up to 9 reviews, 24% between 10 and 49, 40% between 50 and 199, 44% from 200 up.

Rating works as a threshold. With 4.8 stars or more, 34% of listings get named; between 4.5 and 4.7, 18%; below 4.5, 20%. In the model, with the same reviews and trade, a rating of 4.8 or higher multiplies the odds by 2.4.

ChatGPT makes this explicit: next to each name it often shows the rating and the number of reviews. In the answers collected, the median is 155 reviews and the lowest rating shown is 4.1 (in trades where clinics have thousands of reviews, such as vets). When it searches the web before answering, 34 of its 43 searches include the word “recensioni” (reviews).

Other data points the same way. According to SOCi, businesses recommended by ChatGPT average 4.3 stars (SOCi, Local Visibility Index 2026). On the consumer side, 47% of Americans won’t choose a business with fewer than 20 reviews (BrightLocal 2026). AI applies roughly the same criteria as people.

6. Google Maps, search results and AI: how much they overlap

Ranking high on Maps helps, but less than I expected:

More than six in ten top-three listings are not named by any AI. AI also looks beyond Maps: only 42% of the businesses it names are among the top 20 listings of my searches. Among those named by at least two engines the share rises to 73%, so businesses that convince several AIs are usually strong on Maps too.

The same happens with Google’s classic results. AI Overviews sources rank in the top 10 results for the same search in 66% of cases. Ahrefs, on a far larger and not only local sample, measured 76% in July 2025 and 38% in March 2026 (Ahrefs, Search Engine Journal). In local search the first page seems to matter even more than elsewhere, but with 21 answers mine is a hypothesis rather than a measurement.

One detail changes the way you work. For generic questions (“can you recommend a good dentist”) Google shows AI Overviews in only 6 cases out of 20: almost always the Maps local pack appears instead. For questions with a specific need (“a dentist for an implant”) AI Overviews appear in 15 cases out of 20. Whitespark found the same pattern in the US: AI Overviews on 15% of pure local searches, 92% of informational ones (Whitespark, May 2025). For a business that means two front doors: the Maps listing for people searching for the trade, and service-specific website pages for people searching for a solution.

Inside AI Mode, too, Google is shifting citations towards its own listings: according to Profound, google.com’s citation share grew 8.4 times between April and June 2026, almost entirely through business profiles, especially in restaurants, home services, real estate and healthcare (Profound).

7. Answers change often

Of the 365 businesses named at least once, 265 (73%) appear on only one engine. 77 appear on two, 18 on three and just 5 on all four. Kevin Indig found that 91% of citations appear on only one of ChatGPT, Perplexity and AI Overviews (Growth Memo, July 2026).

Answers also change over time. In a first test I repeated the same questions about two and a half hours apart for two trades: in one case AI Overviews changed all four names, in another ChatGPT kept three out of four. BrightLocal measures an average overlap of 20-33% between repeated searches. The second collection, two weeks after the first and covering every question, will measure this volatility properly.

The practical consequence for a business: a single question to ChatGPT doesn’t tell you whether you’re “in” or “out”. You need to look at several engines and repeat the checks over time.

8. Digital PR, “best of” lists and shortcuts to avoid

Almost half the answers (65 out of 141) cite at least one “best” or “ranking” page as a source. Perplexity does it almost always (39 answers out of 40), AI Overviews in 9 out of 21, ChatGPT in 12 out of 40, Gemini in 5 out of 40. Most are rankings generated by platforms from their own reviews (Fresha, Edilnet, Tripadvisor, Addlance, TopDoctors, Matrimonio.com), and only a few are blog posts or local websites. Newspapers, guides and blogs in general appear in just 15 answers.

This fits Whitespark’s 2026 survey, in which 47 local SEO experts rank presence on “best of” lists as the top factor for AI visibility, followed by a dedicated page for each service (Whitespark, Local Search Ranking Factors 2026). Coverage in major national outlets, on the other hand, seems to matter little: Seer Interactive found a near-zero correlation between presence on top news sites and mentions in ChatGPT (Seer Interactive, 2025).

For a small business, then, useful digital PR means getting onto the lists AI reads: the rankings on your industry platform (which depend on the reviews collected there), “best … in [city]” articles from local blogs and publications, and industry guides. The weight varies a lot by trade: high for restaurants, weddings and wellness, low for emergency trades like plumbers and vets, where Maps and reviews decide.

Two shortcuts work, for now, but are worth avoiding.

The first is “best [trade] in [city]” pages that a business writes about itself. I saw a couple cited in the data. A study by Lily Ray across 100 searches, reported by Search Engine Journal, shows that when AI Overviews cite such a list they recommend a competitor 69% of the time (Search Engine Journal, July 2026). In an Ahrefs experiment, AI still recommended a competitor 43% of the time (Ahrefs, July 2026), and according to Lily Ray Google has started demoting sites full of these pages (Lily Ray). For lawyers and healthcare professionals, professional conduct rules on self-promotion add further limits.

The second is a Google Business Profile name stuffed with keywords, along the lines of “Name - Best [trade] [city] 24h”. In my data these listings get named slightly more often (33% vs 23%; 1.54 in the model, borderline significant), because Gemini repeats the listing name as it is. Google’s guidelines forbid it, though, and listings that do it risk suspension: a big risk for a small gain.

9. What to do

1. Have your own website, with a page for each service and your city in the copy. Pages built around specific needs (“dental implants”, “uncontested separation”) are the ones AI Overviews cite when the question is specific.
2. Complete and claim your Google Business Profile: right categories, opening hours, services, linked website. It is Gemini’s main source and increasingly AI Mode’s.
3. Ask for genuine reviews consistently, without incentives (Google’s rules forbid them). Getting past 10 reviews and keeping a high rating are the two steps with the biggest effect in the data.
4. Look after your profile on your industry platform (table in section 3), reviews included.
5. Try to get onto local lists written by others, instead of writing one about yourself.
6. Check every month, across several engines, whether and how you’re named. One test isn’t enough.

At this point, the groundwork of local SEO and the work of getting cited by AI largely overlap.

10. Limits of the study

The data comes from one city, Catania, and so far from one collection. The relationships I found are correlations: they show which features go together with more mentions, not that adding them gets you named. Some groups are small (27 social-only listings, 36 sites with structured data, 22 listings without a public address) and results for those groups should be read with caution. I matched the names in the answers to the Maps listings by hand, so some errors are possible. The questions were asked by a generic user in Italy, not from a phone in central Catania. The second collection at the end of October will reduce some of these limits.

11. Transparency and data

Aloi Digital builds websites and does local SEO, so I have an obvious interest in the website result. That is exactly why the method and data are public: you can download the list of sources cited by AI (CSV, with the number of answers per engine; individual business websites are aggregated) and reuse it by crediting the source with a link to this page. I excluded my own category (SEO consultants and web agencies) from the study, so as not to measure myself.

The names of the most cited businesses, trade by trade, will appear in a separate article about Catania (in Italian), due in November after the second collection.

Studies cited

• BrightLocal, Local Consumer Review Survey 2026, February 2026
• BrightLocal, How AI search uses listings sources, July 2025
• BrightLocal, Local AI Visibility Study, September 2026
• SOCi, 2026 Local Visibility Index, February 2026
• Semrush, AI Search Visibility Study, September 2025
• Ahrefs, Search rankings and AI citations, July 2025; update reported by Search Engine Journal, March 2026
• Ahrefs, Self-promotional content AI SEO experiment, July 2026
• Whitespark, AI Overviews in local search, May 2025
• Whitespark, Local Search Ranking Factors 2026, November 2025
• Profound, google.com is now AI Mode’s #2 cited domain, 2026
• Kevin Indig, AI Halftime Report H1 2026, July 2026
• Lily Ray’s research reported by Search Engine Journal, July 2026
• Seer Interactive, Does being mentioned on top news sites impact AI answer mentions?, March 2025