The Zagat Machine: What Gets a Local Business Recommended by ChatGPT, Gemini and AI Overviews

Ybor City storefronts at dusk with paper stars in lit restaurant windows and an orange door, local AI recommendations
Law Smith, Founder and President of Tocobaga, a Tampa marketing agency

Law Smith

Founder & President, Tocobaga

Law Smith is the Founder and President of Tocobaga, a Tampa-based ROI-focused marketing agency and SMB advisory. He has strategically advised 1,000+ small and medium businesses, executed 600+ integrated marketing campaigns, and written 30+ business plans across industries ranging from professional services to e-commerce to B2B technology. Tocobaga is a Google Partner and Squarespace Gold Partner headquartered in Ybor City.

The complaint that built a restaurant empire was made over wine in Manhattan, and nobody remembers exactly who made it. Tim and Nina Zagat hosted a monthly dinner for friends who cared far too much about food, a club with the gloriously self-serious name of the Downtown Wine Tasting Association.* Both hosts were lawyers. They married in 1965, finished Yale Law School in 1966, and spent their first years of practice in Paris, where they kept a one-page cheat sheet comparing what the guidebooks and their friends said about Paris restaurants.*,† Back in New York, Tim did corporate work at Hughes Hubbard & Reed and Nina practiced at Shearman & Sterling.*,†

At one of those dinners, a guest griped about the restaurant critic at The New York Times. One critic, one palate, one set of bad nights, printed as gospel. Tim answered like a lawyer who would rather collect testimony than argue taste: why not survey our friends?* Nina dates the start to 1979.† Depending on who tells the story, somewhere between 100 and 200 friends and colleagues rated the places they actually ate, scoring food, decor, service and cost, and adding a line of comment.*,‡ Two famous rooms, Luchow's and Sign of the Dove, got savaged.*

The first three editions were free. By 1982, about 600 amateur critics were sending in ratings, far more than two lawyers could tally at home. Publishers passed on a bound edition, so the Zagats printed it themselves and hauled boxes into small Manhattan bookstores. It sold 7,500 copies in its first year and 18,000 in its second.* Tim kept his day job until 1987.*

The format is the part worth remembering. Food was scored on a 30-point scale, with separate numbers for decor, service and cost.† Then the editors did something no single critic could do. In Nina Zagat's words, they would "pull together quotes from lots of different people based on the numerical ratings received by a place."† Each restaurant became a mosaic of quoted fragments from strangers. Eventually about 200,000 people were filling out surveys for guides covering roughly 70 cities.* In 2011, Google paid $151 million in cash for the company and its "vast store of reviews and local data."§

Forty-seven years after that dinner, the little red book is back. It just lost its cover. When a homeowner in Brandon, east of Tampa, types "emergency AC repair near me open now" into ChatGPT, or a retiree in St. Petersburg asks Gemini for "the best estate planning attorney in St. Pete," the machine does what the Zagats did. It polls a crowd. It tallies. It quotes fragments. Only the scale changed: the surveyors are now every review site, list, forum and news story the engine can reach, and the survey re-runs every time somebody asks.

At Tocobaga, the Ybor City marketing agency and SMB advisory behind SolvingHow.com, that question arrives on intro calls in some version of the same sentence. Law Smith, Tocobaga's founder, hears it from HVAC owners, two-partner law firms and PE-backed platform companies alike: How are other owners getting their local businesses recommended by ChatGPT, Gemini or Google AI Overviews, and which changes have actually moved the needle? Tocobaga's earlier field guide, Phone a Friend, covered how AI engines choose what to cite in general. This is the local edition. It is about the survey itself: who the surveyors are, what they write down, and how a business earns better comments without stuffing the ballot box.

How do I get my business recommended by ChatGPT?

A local business gets recommended by ChatGPT by already ranking in ordinary search, by appearing on credible "best of" lists and industry directories, and by being described specifically, by service and by town, across many independent websites. The AI does not pick a favorite out of thin air; it summarizes the consensus it can find, the same way the Zagat Survey summarized its diners.

The strongest measured evidence comes from OppAlerts founder Ben Wills, whose study ran 403,000 prompts through large language models across 100 industries and 1,100 buyer personas. The number one predictor of AI visibility, across every industry, was simply being in the search results. Homepage keyword relevance ranked second and domain page rank third.‖ Whitespark's write-up, drawing on Darren Shaw's breakdown, added the obligatory warning that correlation is not causation and the more useful reminder that SEO is not dead.‖

The expert view lines up with the measured one. Whitespark's 2026 Local Search Ranking Factors survey, the first edition to score AI search visibility, asked 47 local search specialists to weigh 187 factors.¶ Expert-curated "best of" lists came out on top for AI visibility.** Shaw's summary of the whole exercise fits on a coffee mug: "In AI SEO, mentions (citations) are the new link."¶

So the fundamentals are not new. What changed is the weighting. A 4.9-star rating is a score, and Zagat readers never bought the guide for the score alone; they bought it for the quoted fragments that explained the score. AI engines work the same way. Picture a two-partner estate planning firm in Clearwater with 61 reviews that all say some version of "great experience." Now picture its competitor down Gulf to Bay Boulevard with 140 reviews that mention revocable trusts, probate, blended families and Clearwater by name. When someone asks for "an estate planning attorney in Clearwater who handles trusts for second marriages," only one of those firms has handed the machine a sentence to write. The firm that supplies the sentence gets the recommendation.

Where each AI surface gets its local data, and the first fix that feeds it.
AI surface Main local data it leans on First fix for an owner
ChatGPT Its own index, scraped Google and Google Maps results, licensed Yelp and TripAdvisor data (Peec, 2026) Clean up Yelp, keep Google rankings strong, earn best-of list placements
Google AI Overviews and AI Mode Google's index plus Business Profile and Merchant Center feeds (Google via Whitespark, 2026) Complete the Business Profile and build one page per service
Ask Maps in Google Maps 300 million places and reviews from 500 million contributors (Google, 2026) Invite reviews that name the service and the town
Apple Maps and Siri Apple Business place cards plus partner data such as Yelp (Axios, 2026) Claim the place card; add photos, hours, Showcases and action links
Microsoft Copilot Bing's index and Bing Places listings Claim Bing Places, import from Google, match name, address and phone
Best-of lists and industry sites Editorial lists, directories and associations cited as sources (Whitespark, 2026) Pursue curated lists with a visible method, not paid plaques

What do local AI search prompts actually look like?

Local AI prompts are longer, more specific and more conversational than the two-word searches most owners spent a decade optimizing for. Nobody asks ChatGPT for "HVAC Tampa"; people ask for "emergency AC repair near me open now," "a family-owned plumber in Riverview who installs tankless water heaters," or "a 3PL in Tampa that works with Shopify brands shipping under 5,000 orders a month."

Google built Ask Maps, launched in the U.S. and India in March 2026, around exactly this kind of question. Its announcement described "a new conversational experience that answers complex, real-world questions a map could never answer before," and offered sample prompts like "Any spots with a cozy aesthetic and a table for 4 at 7 tonight?"†† That is four filters in one sentence: mood, party size, time and availability.

The people typing these prompts are not bored teenagers. BrightLocal's 2026 consumer research found that 64 percent of adults aged 30 to 44 had asked an AI tool for a business recommendation in the past year, against 24 percent of people over 60. ChatGPT led the tools at 31 percent of consumers, with Google's AI Mode at 23 percent.‡‡ That 30 to 44 bracket is the homeowner with a dead compressor in August, the founder choosing a fractional CFO, and the adult child hiring an elder law attorney for a parent.

Location is the wrinkle. In late March 2026, OpenAI began letting ChatGPT users share their device location for "near me" answers, but the setting is off by default. Search Engine Roundtable reported on March 31 that one early tester, Glenn Gabe, asked for the "best steakhouses near me" and got several restaurants about 45 minutes away.§§ When the engine is guessing where "near me" is, the business whose website, listings and reviews actually name Brandon, Riverview and Valrico has a far better shot at being matched to someone in Brandon, Riverview and Valrico.

Every clause in a long prompt is a filter, and the AI can only satisfy a filter with text it found somewhere. "Open now" needs accurate hours on every listing, not just Google. "Financing" needs the word financing on a page or in a review. "Spanish-speaking" needs someone, somewhere, to have written it down. The Zagat editors could only print the comments their surveyors actually wrote. The machine has the same limitation at a much larger scale.

Where does ChatGPT get its local business information?

ChatGPT builds local answers from a patchwork: its own web index, scraped Google search and Google Maps results, licensed data from Yelp and TripAdvisor, and several other search providers. No single listing controls the answer, which is why consistency across all of them matters more than perfection on any one.

The most detailed public teardown comes from Tomek Rudzki at Peec AI, whose September 2026 research found that ChatGPT draws on its own crawler and index plus at least eight external providers. Google web results and Google Maps listings arrive through third-party scrapers. Yelp and TripAdvisor are licensed partners for local data. Rudzki's blunt verdict on the old conventional wisdom: "Bing is not a proxy."‖‖

The Yelp piece is new, and it is large. Axios reported on July 23, 2026, that Yelp is licensing its reviews, ratings, photos and business details to OpenAI, with Yelp branding and links appearing in ChatGPT answers and a "Request a Quote" feature on the way.¶¶ Whitespark's quarterly roundup put the size of that library at 330 million local business reviews.‖ For a Tampa plumbing company that has not logged into Yelp since 2019, that abandoned page is now an input to the AI tool consumers use most for local recommendations.‡‡

Earlier snapshots show how the method evolves without changing the principle. In a test documented by Damian Rollison for Search Engine Land, ChatGPT described scanning the first 20 to 30 Bing results for a local query, examining five to eight promising sources more closely, then choosing three to five that showed star ratings and were "verifiable and linkable."*** BrightLocal's study of 800 local searches across 20 industries and 20 cities, run in November 2024, found business websites made up 58 percent of ChatGPT's sources, business mentions 27 percent and directories 15 percent.†††

The plumbing will keep changing. The principle has held since 1979: a guide is only as good as its surveyors, and a business is only as visible as what the surveyors report. A wrong phone number on Yelp becomes a wrong phone number in an AI recommendation. Stale hours on TripAdvisor become a "closed now" in a chatbot. Every listing is a surveyor, and every surveyor gets a vote.

Which local SEO factors matter most for AI Overviews and ChatGPT in 2026?

Citations dominate. In Whitespark's 2026 ranking factors, experts placed expert-curated "best of" lists first for AI visibility, a dedicated page for each service second, prominence on industry-relevant domains third, the quality of unstructured citations such as news articles and blog posts fourth, and authority on third-party review sites fifth.**,‡‡‡

Whitespark's own accounting is that structured and unstructured citation sources dominate the entire top 20.** Further down, the quantity of unstructured citations ranked 7th, high Google ratings 8th, the diversity of third-party review sites 11th and the quantity of unlinked brand mentions 12th.** Translation: the AI wants to hear about a business from many independent voices, and it gives extra weight to voices with editorial standards.

The contrast with the Google map pack is where it gets interesting, because Whitespark scored the same factors for both. "Business is open at time of search" ranked 5th for the local pack and 104th for AI search. A visible address on the Business Profile, which punishes service-area businesses that hide their address, ranked 7th for the pack and 54th for AI. Scannable content structure across the website ranked 128th for the pack and 10th for AI.¶

Read those numbers as two different games. The map pack is mostly a proximity and profile contest. AI search is mostly a reputation and content contest. For a mobile pet groomer, a roofing contractor or a law firm that meets clients by video, that is the best news local search has delivered in years. The map pack will always favor whoever sits physically closest to the searcher. The AI favors whoever the internet describes most specifically.

One caution. Whitespark's AI column is expert opinion, not a controlled experiment, and Shaw himself noted the question was ambiguous, since Google's AI Mode and Gemini can see data that other models cannot.¶ Still, the OppAlerts correlations and the expert survey point the same direction. Zagat never cared how close a restaurant was to the reader's apartment. It cared what the diners said.

1981-style snapshot of survey sheets, a typewriter and an orange document box on a table, the Zagat survey method

How does a business get on "best of" lists in its city?

A business gets on credible "best of" lists by being nominated, researched or voted for under each list's published method, which starts with an inventory of every list the AI engines already cite in its category and city. The lists that matter are expert-curated ones that explain how they choose, not plaques sold to whoever answers the phone.

Start with the evidence the AI already shows. Run local prompts and write down every list ChatGPT and Google cite as sources for competitors. In BrightLocal's 2024 study of ChatGPT's local sources, two curated directory sites, Three Best Rated and Expertise, accounted for 24 percent and 18 percent of the directory citations, respectively.††† Every category has its own version: city magazine readers' polls, business journal awards, chamber of commerce small business awards, trade association honors and manufacturer dealer programs.

Then sort the list into three piles. Editorial lists with a visible method get a submission package: license numbers, years in business, review counts and two or three customer stories shared with permission. Reader polls get a campaign during the voting window, asking real customers to vote, which is exactly the crowd behavior the Zagats built a business on. Pay-to-play awards with no stated criteria get skepticism. There is no public evidence that AI engines discount them, but there is no evidence they reward them either, and an award anyone can buy says nothing specific about anyone.

Here is the math for a hypothetical two-partner estate planning firm in Clearwater. Its prompts surface 14 lists cited across its practice areas and cities. Five are editorial with open nominations, four are reader polls and five are pay-to-play. The firm targets the nine legitimate lists. If it lands on three in the first year, it moves from zero of the 14 sources the AI is reading to three, a 21 percent share of the room. If its two strongest competitors sit on six lists each, the firm still trails, but it has gone from invisible to on the ballot.

Zagat's authority came from the independence of its surveyors. The diners were not on the restaurants' payroll. The lists that carry weight with AI engines share that trait, and Whitespark's experts ranked them first for a reason: an editor with nothing to sell is the closest thing the internet has to a disinterested diner.

Do Yelp, BBB and industry directories still matter for AI search?

Yes, and more than they did three years ago, because AI engines treat them as surveyors. Yelp now licenses its reviews and business data directly to OpenAI, and Whitespark's 2026 experts ranked prominence on industry-relevant domains third among all AI visibility factors.¶¶,**

Consumers have spread out, too. BrightLocal's 2026 Local Consumer Review Survey found that the average consumer uses six review sites, and that Facebook, Tripadvisor, BBB, Trustpilot, Healthgrades, Yellow Pages and Angi all saw usage rise.§§§ Whitespark's advice to its own readers was blunt: review requests can no longer go only to Google.¶

Which directories matter depends on the category. In a 2025 Whitespark study of 153 queries across 17 local categories in nine U.S. cities, Facebook appeared on nearly one and a half times as many businesses' Bing listings as Yelp did, Yelp appeared more than twice as often as TripAdvisor, and Porch led for garage door repair.‖‖‖ For law firms, the industry domains are the state bar directory, Avvo, Justia and Martindale-Hubbell. For home services, it is Angi, Porch, BBB and the manufacturer's dealer locator. For a B2B consultancy or a manufacturer, it is the trade association's member directory and whichever platforms buyers actually use to vet vendors.

The cost of neglect compounds in a way owners rarely see. Consider a PE-backed HVAC platform in its second add-on acquisition: three brands, three legacy Yelp pages, two of them still routing to phone numbers the sellers disconnected at closing. Each stale page is a surveyor filing a report with the wrong facts, and those reports now flow into ChatGPT through one licensing deal and into Apple Maps through another.¶¶ A carefully modeled roll-up thesis can be quietly undercut by a free listing nobody owns.

The fix is boring and fast. Claim every listing, merge duplicates, match the name, address, phone and hours exactly, add real photos, and reply to reviews on every platform, not just Google. The Zagat guide reported the consensus; it never fact-checked a restaurant's hours. Neither will the AI.

Does ChatGPT use Google reviews, and does review text matter?

ChatGPT reaches Google review data indirectly, through scraped Google Maps results and through pages that quote ratings, while Google's own AI features read Business Profiles and reviews directly. Either way, review text has become raw material: the AI quotes and summarizes what customers wrote, so reviews that name the service, the town and the outcome give it something specific to say.‖‖,***

Google has been explicit about its side. Whitespark's quarterly roundup relayed guidance from Google's Lisa Landsman that generative AI features pull directly from Business Profiles and Merchant Center product feeds to recommend businesses and in-stock inventory in real time.‖ Ask Maps answers questions using information on more than 300 million places and reviews from a community of more than 500 million contributors.†† Meanwhile, the owner-influenced question-and-answer section on Google Business Profiles is gone. Google retired the Q&A API on November 3, 2025, and replaced public Q&A with Ask Maps.¶¶¶ The reviews answer the questions now.

Consumers lean on the summaries. BrightLocal found that 82 percent of consumers read AI-generated review summaries and 23 percent would rely on the summary alone. The same survey found that 56 percent look for consistent sentiment across reviews, 74 percent care only about reviews from the last three months, 47 percent will not use a business with fewer than 20 reviews, and 31 percent will only use a business rated 4.5 stars or higher, up from 17 percent a year earlier.§§§

So the review ask has to change. "Please leave us a review" produces "great service, highly recommend." Inviting the customer to mention what was fixed and which part of town the job was in produces "Replaced our 15-year-old system in Valrico the same day it died." The second review is a Zagat quote. The first is a star with no sentence attached. Prompting customers for specifics is honest. Scripting their words, paying for them or steering unhappy customers away from the ask is not, and Tocobaga's earlier piece on fake and incentivized reviews, The Velvet Rope Writes Reviews, explains why that line matters more every year.

The arithmetic is simple. A business with 300 reviews that all say "great service" hands the AI a rating and nothing to quote. A business with 60 reviews that mention "tankless water heater" and "Seminole Heights" hands it a ready-made recommendation for anyone asking about tankless water heaters in Seminole Heights. The editors at Zagat could not print a comment nobody wrote, and neither can Gemini.

Do Bing Places and Apple Business matter for AI recommendations?

Both are worth the hour it takes to claim them, for different reasons. Apple Maps nearly doubled as a review source in BrightLocal's 2026 survey, from 14 percent to 27 percent of consumers, while Bing Places is cheap insurance whose direct influence on ChatGPT is murkier than the 2024 advice suggested.§§§

Start with Apple. Over the same year that Apple Maps nearly doubled, Google's share of review readers slipped from 83 percent to 71 percent.§§§ Apple Business, the free platform that replaced Apple Business Connect in April 2026, lets a business claim its place card in Apple Maps, add photos and hours, publish Showcases for offers and seasonal promotions, and add action links such as order or reserve, with the information reaching customers across Maps, Mail, Wallet and Siri.**** Apple Maps also draws on Yelp, which Axios noted already supplies data to Apple as well as to OpenAI.¶¶ Fixing Yelp is a two-for-one.

Picture what that 27 percent means for a busy service business. If 400 people look up a Tampa roofing company in a month, roughly 108 of them are the kind of consumer who reads Apple Maps reviews. If the Apple place card shows a 2018 photo, no hours and a phone number that rings a former partner's cell, more than a quarter of the audience meets a ghost.

Bing is a more honest conversation. The 2024 advice was "ChatGPT runs on Bing, so claim Bing Places." Rollison's Search Engine Land test found ChatGPT pulling web content from Bing but not seeing Bing Places profile data directly.*** Peec's 2026 research found ChatGPT running its own index plus multiple providers, with Microsoft's Web IQ grounding platform confirmed as one of them.‖‖ Whitespark's Miriam Ellis wrote in 2025 that the link was unconfirmed and the industry had no firm consensus.‖‖‖

The practical answer is the same either way. Bing Places is free, it can import an existing Google profile, it feeds Bing's own local results for Microsoft's Copilot, and Microsoft's data still reaches ChatGPT through at least one pipe. Claim it, verify it, match it to Google and move on. Thirty minutes to cover a surveyor is a bargain. A quarter spent fine-tuning it would be a misallocation.

Five identical gold trophies with blank plates and one orange ribbon on an HVAC office shelf, best-of lists for AI

Do Reddit threads and local press mentions help a business get recommended?

They help, because AI engines lean heavily on unstructured mentions, but only when the mentions are real. Pew Research Center found that Wikipedia, YouTube and Reddit together made up 15 percent of the sources cited in Google's AI summaries, and Reuters reported in 2024 that Google's deal to license Reddit's content was worth about $60 million a year, according to a source.††††,‡‡‡‡

Whitespark's experts ranked the quality of unstructured citations, meaning newspaper articles, blog posts, government sites and industry associations, fourth among AI visibility factors, and their sheer quantity seventh.** Unlinked brand mentions, the plain-text "we used them and they were great" with no hyperlink attached, ranked 12th.** The internet's offhand comments have become a ranking signal.

Reddit deserves a warning label. City subreddits and neighborhood forums are where people ask "who's a good electrician in South Tampa," and the answers carry the authority of a stranger with nothing to gain. That authority evaporates the moment a business fakes it. Moderators ban sock-puppet accounts, regulars recognize the pattern, and a single exposed fake becomes its own thread, which the AI will also read. The honest playbook is slower and better: answer questions under a clearly identified owner account and let customers mention the business on their own. BrightLocal's 2024 study found no forum results at all among ChatGPT's local sources at the time, so Reddit is best treated as a long-term reputation asset, not a ranking hack.†††

Local press works the same way, with an editor in the middle. BrightLocal found that consumers' use of local news sites as a review source fell from 48 percent to 29 percent in a year.§§§ Readers drifted; the machines did not. A business journal profile, a TV segment about a free AC tune-up program for veterans, a chamber of commerce award writeup or a guest spot on a local business podcast all create the independent, editorial mention that AI engines weight heavily. Zyppy founder Cyrus Shepard offered a less obvious tactic in Whitespark's roundup: build a genuine resource of peer businesses worth recommending, which he called a strong route to AI citations.‖

The Zagat Survey never let a restaurant fill out its own ballot. The internet cannot stop a business from trying, but it is extraordinarily good at noticing.

What should a service page say so AI tools recommend the business?

A service page that gets picked by AI states, in plain sentences near the top, what the service is, who it is for, which towns it covers, what it typically costs, how fast it happens and which credentials stand behind it. Whitespark's experts ranked a dedicated page for each service second among AI visibility factors, and scannable content structure ranked 10th for AI against 128th for the map pack.‡‡‡,¶

Phone a Friend covered answer-first page structure in general. The local layer is more specific, and it lives in four kinds of sentences most service pages never write.

The first is service-area language. "Serving the greater Tampa Bay area" is fog. "Emergency AC repair in Brandon, Riverview, Valrico, Seffner and Plant City, 24 hours a day" is a match for five different prompts. It matters most for service-area businesses: when Claudia Tomina tested Ask Maps in 2026, it could not locate a service-area business at all.‖ A business without a public address has to put its geography in its words.

The second is price-range language. Prompts increasingly carry money: "affordable," "financing," "under $10,000," "flat fee." An AI cannot match a budget filter to a page that refuses to discuss money. A plain sentence stating that most repairs land between $250 and $900 and most full replacements between $8,500 and $14,000, using the company's real numbers, gives the engine something to quote and gives the customer a reason to call instead of bounce.

The third is credentials. Board certification in wills, trusts and estates law, a state contractor license number, NATE-certified technicians, a manufacturer's dealer designation: each is a verifiable claim the AI can repeat with confidence. Vague phrases like "experienced professionals" give it nothing to verify and nothing to say.

The fourth is the specific customer. "Trusts for blended families." "Shopify brands shipping under 5,000 orders a month." "Commercial kitchens with walk-in coolers." Each phrase answers a prompt somebody is typing right now. A Tampa 3PL that publishes its minimums, its pick-and-pack fee range and the platforms it integrates with will match prompts its competitors cannot. A management consultancy that names the industries and revenue bands it serves gives the AI a reason to put it in front of a $20 million distributor hunting for an operations advisor.

A Zagat score was never vague. A 26 for food meant something precise to the reader. A service page should be equally precise, because the machine reading it is looking for precisely that.

How can a business test whether ChatGPT recommends it?

A business tests its AI visibility with a prompt grid: every core service crossed with every target town, asked on each AI platform its customers use, logged monthly in a spreadsheet. There is no reliable rank tracker for local AI answers yet, so a disciplined manual grid beats guessing.

The rows are service-and-town combinations, phrased the way people talk: one "best" prompt, one "near me open now" prompt and one constraint prompt, such as "with financing" or "for blended families." The columns are the platforms: ChatGPT, Gemini or Google AI Mode, Ask Maps on a phone and Microsoft Copilot. Each cell records four things: whether the business was named, its position in the answer, which competitors were named, and which sources the AI cited.

Method matters. Name the town in the prompt rather than relying on "near me," since ChatGPT's location sharing is off by default.§§ Use a logged-out or fresh session so personalization does not flatter the business. Ask Maps personalizes answers based on places a user has searched for or saved, so test from a phone that has never looked up the business.††

The column that pays for the whole exercise is "sources cited." When ChatGPT names a competitor three times and cites the same city magazine list, the same Yelp page and the same news story each time, that is the to-do list, delivered by the judge.

Pair the grid with the data Google now provides. Google Search Console began showing some site owners impressions and URLs for AI Overviews and AI Mode in 2026, and SEO consultant Aleyda Solis summarized the pattern to watch in one line: "AI impressions rising with flat clicks = exposure without traffic."‖ Then add "an AI assistant like ChatGPT" as an answer to the lead-source question on web forms and in the call script.

Expect that field to undercount. A Yext survey of nearly 4,000 consumers found only 5 percent use AI as a direct path to purchase, with the rest moving through a "verification loop" of Google results, websites and reviews before buying.‖ BrightLocal found 97 percent of AI users at least sometimes double-check AI recommendations against real reviews.‡‡ Plenty of AI-sourced customers will tell the receptionist they "found the company on Google," and they will be telling the truth about the last step. The Zagats re-ran their survey every year. A monthly grid is the same discipline at a faster cadence.

What does the math look like for a Tampa Bay HVAC company?

For a hypothetical 12-truck residential HVAC company in Brandon, a realistic program means testing 144 prompts a month, nearly tripling the share of reviews that name a service and a town within six months, and landing on four more credible lists within two quarters. Every number below is an assumption chosen to show the arithmetic, not a promise of results.

The grid covers six core services (AC repair, emergency AC repair, AC replacement, heat pump installation, duct cleaning and maintenance plans) across eight towns (Brandon, Riverview, Valrico, Seffner, Plant City, Lithia, Apollo Beach and Tampa) on three platforms (ChatGPT, Google AI Mode and Ask Maps). That is 6 x 8 x 3, or 144 prompts. At about 90 seconds per prompt, including logging, the grid takes 216 minutes, or 3.6 hours a month.

The baseline is sobering. The company is named in 11 of 144 answers, 7.6 percent. Its strongest competitor is named in 41, or 28.5 percent. The sources the AI cites for that competitor: three "best of" lists, its Yelp page, its BBB profile, a manufacturer's dealer locator and a TV news story about its free tune-up program for veterans. The company itself has 412 Google reviews averaging 4.8 stars, but only 38 of them, or 9.2 percent, mention a specific service by name, and just four mention any town other than Brandon.

Now the plan, in numbers. Twelve trucks running three jobs a day for 22 workdays is 792 jobs a month. A text from the technician after every job, converting at 8 percent, yields about 63 new reviews a month. The new ask, which invites customers to mention what was fixed and in which town, draws a specific answer from roughly 45 percent of reviewers, or about 28 specific reviews a month. Over six months that is 378 new reviews, 168 of them specific. Total reviews climb to 790, and specific reviews rise from 38 to 206, from 9.2 percent of the total to 26.1 percent.

The other workstreams run in parallel. Of the nine lists cited across the grid, the company appears on one; it pursues the five editorial and reader-poll lists with open entries and lands on four within two quarters. Three stale listings (Yelp, BBB and Apple Maps) get fixed in the first week, and Bing Places gets claimed. Six dedicated service pages go live, each naming all eight towns, real price ranges and the company's license number.

Then the payoff scenario. Suppose that six months later the company is named in 36 of 144 answers, 25 percent, and the lead-source field shows AI-attributed leads rising from 4 a month to 14. At a 50 percent booking rate, that is 7 jobs. If five are repairs at a $450 average ticket and two are system replacements at $11,000, that is $2,250 plus $22,000, or $24,250 in monthly revenue. At a 40 percent gross margin, it is roughly $9,700 a month in gross profit, from work that costs a few hours of testing and a few hours of outreach each month. One replacement covers the effort many times over.

The grid shrinks gracefully for smaller firms. A two-attorney estate planning practice in St. Petersburg with four practice areas (wills, revocable trusts, probate and guardianship) across five cities (St. Petersburg, Clearwater, Largo, Seminole and Dunedin) on three platforms tests 60 prompts, about 90 minutes a month. At an assumed $3,500 flat fee for a trust package, one additional client a quarter pays for the testing time many times over. The Zagats proved the point in 1979: opinions become useful the moment someone bothers to count them.

Hand placing an orange magnet on a whiteboard grid in an HVAC dispatch office, a monthly AI prompt test by city

What is the strongest case for ignoring AI search and sticking with Google Maps?

The strongest case is that most local buying still runs through Google, and AI recommendations largely echo Google anyway. Google remains the review source for 71 percent of consumers, a Yext survey found that only 5 percent of consumers use AI as a direct path to purchase, and the OppAlerts data says the best predictor of AI visibility is ranking in traditional search.§§§,‖

The skeptic has more ammunition. The AI that most local customers actually encounter is Google's own. In Pew Research Center's 2025 study of 68,879 searches, AI summaries appeared on 18 percent of them, and when a summary appeared, users clicked a traditional result in 8 percent of visits, against 15 percent when none appeared.†††† That is a Google problem with a Google solution: a strong Business Profile, a fast site and the map pack. Google is also testing Local Services Ads inside AI Mode, and Whitespark's Shaw put the consequence plainly: "More ads means less space for organic results."‖ An owner with ten spare hours a month could reasonably conclude that excellent traditional local SEO is the whole job, and that AI visibility will follow on its own.

That owner is about 80 percent right, which is exactly the point. Most of the work overlaps. The other 20 percent is where the two games diverge, and it is where the cheap wins live. A visible address ranks 7th for the map pack and 54th for AI. Best-of lists rank first for AI. Yelp feeds ChatGPT and Apple Maps, and no amount of Google ranking fixes a stale Yelp page.¶,**,¶¶

The verification loop cuts both ways, too. If AI builds the shortlist and Google verifies it, a business that never makes the shortlist never gets verified. BrightLocal's 2026 research put AI tools third among sources of local recommendations, used by 45 percent of consumers, behind only Google and Facebook.‡‡ Consumers are not choosing AI instead of Google. They are using both, in sequence, and the first step decides who gets considered in the second.

In the early 1980s, plenty of restaurateurs could have dismissed a self-published survey of amateurs as a parlor game. By its second year, the parlor game was selling 18,000 copies.*

What should a local business do first to get recommended by AI?

Start with a baseline prompt grid and a listings cleanup, then change the review ask, build one page per service and pursue credible lists, in that order. Measurement shows where the gaps are, and the cleanup stops the AI from repeating bad data while the slower work compounds.

  • Run a baseline prompt grid of core services, target towns and the three or four AI platforms customers use, and log who gets named and which sources get cited.

  • Fix the surveyors: Google Business Profile, Yelp, Apple Business, Bing Places, BBB, Facebook and the top two industry directories, with identical name, address, phone and hours.

  • Build or rewrite one dedicated page per service that names the towns served, real price ranges, credentials and typical turnaround in plain sentences.

  • Change the review ask to invite specifics, such as what was done and where, with no scripts, no incentives and no filtering of unhappy customers.

  • Inventory every "best of" list the AI cites in the category, then pursue the editorial lists and reader polls with a visible method.

  • Earn two independent mentions a quarter through local press, chamber programs, trade associations, sponsorships or local podcasts.

  • Show up honestly in city subreddits and neighborhood forums under a disclosed account, answering questions instead of pitching.

  • Add "an AI assistant like ChatGPT" as an answer to the lead-source question on web forms and in the call script.

  • Re-run the grid monthly, compare the cited sources against the previous month, and double down on whatever moved.

None of those steps requires a new software subscription, and most require more discipline than budget. The order is the strategy: measure, clean, describe, ask, earn, repeat. Tocobaga's earlier guide, Worth a Detour, covers the Google Business Profile setup that sits underneath all of it.

Tocobaga builds and runs this program for owners who would rather run their business than run 144 prompts a month. Law Smith and Tocobaga have advised more than 1,000 small and mid-sized businesses and executed more than 600 integrated campaigns, and the work here starts the same way every time: with the grid, so the owner sees the real baseline before spending a dollar. An intro call is the place to start, at (813) 534-1261 or hello@solvinghow.com through SolvingHow.com. Flat fees, never a percentage of ad spend, no auto-renewing contracts, and the client owns every listing, account and asset. Tocobaga is not for everyone, and when it is not the right fit, it says so and points the owner somewhere better.

Will AI search replace Google Maps for local businesses?

Not soon, and probably not in the way the headlines imply. AI is becoming the layer in front of the map: Ask Maps puts a conversational assistant inside Google Maps itself, and ChatGPT is adding Yelp's reviews and quote requests to its answers, so the map and the chatbot are merging into one survey.††,¶¶

The next step is already visible. Whitespark's roundup noted that Square and Toast are participating in a program that lets people order through Ask Maps, with Uber Eats slated to join, and argued that this makes saved places a ranking factor for agentic tasks.‖ When the assistant does not just recommend a business but books it, the recommendation is the sale. There is no second page of results to fall back on.

Tim and Nina Zagat never controlled what their surveyors wrote. They controlled the method: ask a lot of people, tally honestly, quote what they said. The restaurants that thrived in the guide did not game it. They gave diners something specific to say, and they kept earning it, because the survey came back every year.

The machine works the same way, minus the annual deadline. It asks the crowd every time someone types a question. It tallies Yelp and Apple Maps and the city magazine and the Reddit thread and all 412 of those Google reviews. Then it quotes what it finds.

ChatGPT is Nina Zagat with a server farm. It can only quote what somebody already wrote.

What is local AI search optimization?

Local AI search optimization is the work of making a local business easy for AI assistants such as ChatGPT, Gemini, Google AI Mode and Ask Maps to recommend by name. It overlaps heavily with traditional local SEO but puts extra weight on third-party evidence: expert-curated "best of" lists, industry directories, review text that names specific services and towns, and independent mentions in news and community forums. Whitespark's 2026 ranking factors were the first to score it as a separate category.

How long does it take for a business to start showing up in ChatGPT recommendations?

There is no published benchmark, and anyone quoting a guaranteed timeline is guessing. Listing fixes on Yelp, Apple Business and Google can show up within weeks as data providers refresh. Review text, list placements and press mentions build over months, because lists publish on annual cycles and reviews accumulate job by job. A monthly prompt grid is the only dependable way to see movement for a specific business, service and city.

What is Ask Maps in Google Maps?

Ask Maps is a Gemini-powered conversational feature that Google launched in March 2026 in the U.S. and India on Android and iOS. It answers multi-part questions about places using information on more than 300 million places and reviews from more than 500 million contributors. It effectively replaced the question-and-answer section on Google Business Profiles, which Google wound down after retiring the Q&A API on November 3, 2025. Review text now does the answering.

Are paid "best of" awards worth buying for AI visibility?

Usually not. Whitespark's 2026 experts ranked presence on expert-curated "best of" lists as the top AI visibility factor, and the key word is curated. A list with a published method, an editor and independent criteria works like an independent surveyor. An award sold to any business that pays has no stated criteria and says nothing specific. There is no public evidence that AI engines reward pay-to-play plaques, so the budget usually does more good elsewhere.

Can a service-area business without a storefront get recommended by AI?

Yes, and AI search may treat it better than the map pack does. Whitespark's 2026 study ranked a visible address 7th for local pack rankings but 54th for AI visibility. The catch is that AI tools need text to place the business: in one 2026 test, Ask Maps could not locate a service-area business at all. Naming the specific towns and neighborhoods served on service pages, on listings and in reviews closes most of that gap.

How many reviews does a local business need to get recommended by AI tools?

No study sets a minimum for AI recommendations, but consumer thresholds offer a guide. BrightLocal's 2026 survey found 47 percent of consumers will not use a business with fewer than 20 reviews, 74 percent care only about reviews from the last three months, and 31 percent require a rating of 4.5 stars or higher. For AI, recency and specificity matter as much as raw count, because the engine summarizes and quotes what reviewers actually wrote.

Should a business track AI assistants as a lead source?

Yes. Adding "an AI assistant like ChatGPT" as an answer to the lead-source question on web forms and in phone intake scripts is the cheapest attribution available. It will undercount, because a Yext survey found most consumers verify AI suggestions through Google, websites and reviews before buying, so many will simply answer "Google." Pairing that field with a monthly prompt grid and Search Console's AI Overviews and AI Mode reporting gives a workable picture.

Sources
*
Brennan, Carol. "Zagat, Tim and Nina." Newsmakers 2004 Cumulation, Encyclopedia.com, accessed October 26, 2026. Source for the Downtown Wine Tasting Association dinners, the complaint about The New York Times critic, Tim's survey idea, the first survey of about 100 friends, the poor ratings for Luchow's and Sign of the Dove, the three free editions, about 600 contributors by 1982, sales of 7,500 and 18,000 copies in the first two years, the Zagats' 1965 marriage and 1966 Yale Law degrees, Tim's law career to 1987, and about 200,000 participants across roughly 70 cities. https://www.encyclopedia.com/books/culture-magazines/zagat-tim-and-nina
†
Begley, Sarah. "The Survey: An Interview with Nina Zagat." Vassar Quarterly, 2013, accessed October 26, 2026. Nina Zagat on starting in 1979 with the couple's wine-tasting group, the one-page Paris guide, the firms where she and Tim practiced, the 30-point food scale with separate ratings for decor, service and cost, and editors who "pull together quotes from lots of different people based on the numerical ratings received by a place." https://vq.vassar.edu/issues/2013/02/eat/nina-zagat.html
‡
Shaw, Steven. "The Zagat Effect." Commentary, accessed October 26, 2026. Describes the Zagats as two lawyers who assembled the opinions of 200 acquaintances into a self-published survey of New York restaurants. https://www.commentary.org/articles/steven-shaw/the-zagat-effect/
§
Coldewey, Devin. "Report Puts Google's Zagat Purchase At $151 Million." TechCrunch, October 26, 2011. Google paid $151 million in cash for Zagat, per an SEC filing, and the deal fit Google's local strategy and Zagat's "vast store of reviews and local data." https://techcrunch.com/2011/10/26/report-puts-googles-zagat-purchase-at-151-million
‖
Ellis, Miriam. "22 Local SEO Developments You Need To Know About From Q3 2026." Whitespark, October 8, 2026. Reports the OppAlerts study of 403,000 prompts across 100 industries and 1,100 personas (being in search results the top AI visibility factor, homepage keyword relevance second, domain page rank third), Yelp's 330 million reviews licensed to ChatGPT, Google guidance that generative AI features pull from Business Profiles and Merchant Center feeds, Ask Maps failing to locate a service-area business, Ask Maps ordering with Square and Toast, Search Console AI reporting and Aleyda Solis's comment, Cyrus Shepard's peer-resource advice, a Yext survey of nearly 4,000 consumers finding 5 percent use AI as a direct purchase path, and Local Services Ads tests in AI Mode. https://whitespark.ca/blog/22-local-seo-developments-you-need-to-know-about-from-q3-2026/
Shaw, Darren. "Whitespark's Official 2026 Local Search Ranking Factors Report." Whitespark, November 6, 2025. 47 experts scored 187 factors and added AI search visibility for the first time; "In AI SEO, mentions (citations) are the new link"; scannable content ranked 10th for AI and 128th for the local pack, a visible address 7th and 54th, open at time of search 5th and 104th; review requests should go beyond Google. https://whitespark.ca/local-search-ranking-factors/
**
Ellis, Miriam. "2026 Playbook: How to Optimize Citations for AI Visibility." Whitespark, September 10, 2026. Ranks expert-curated "best of" lists first, industry-relevant domains third, unstructured citation quality fourth, third-party review site authority fifth, unstructured citation quantity seventh, high Google ratings eighth, review site diversity 11th and unlinked brand mentions 12th among AI visibility factors, with citation sources dominating the top 20. https://whitespark.ca/blog/2026-playbook-how-to-optimize-citations-for-ai-visibility/
††
Daniel, Miriam. "How We're Reimagining Maps with Gemini." Google, March 12, 2026. Introduces Ask Maps as "a new conversational experience that answers complex, real-world questions a map could never answer before," drawing on more than 300 million places and reviews from more than 500 million contributors, personalized by searched and saved places, with sample prompts and a launch in the U.S. and India. https://blog.google/products-and-platforms/products/maps/ask-maps-immersive-navigation/
‡‡
Murphy, Rosie. "Half of Consumers Are Asking AI for Business Recommendations." BrightLocal, March 10, 2026. AI tools rank third for local recommendations behind Google and Facebook, used by 45 percent of consumers; ChatGPT is used by 31 percent and Google AI Mode by 23 percent; 64 percent of adults 30 to 44 and 24 percent of those over 60 asked AI for a business recommendation; 97 percent of AI users sometimes double-check AI recommendations against real reviews. https://www.brightlocal.com/research/lcrs-ai-trust/
§§
Schwartz, Barry. "ChatGPT Enables Location Sharing For More Localized Near Me Results." Search Engine Roundtable, March 31, 2026. Device location sharing is optional and off by default; Glenn Gabe's "best steakhouses near me" test returned restaurants about 45 minutes away. https://www.seroundtable.com/chatgpt-location-sharing-41128.html
‖‖
Rudzki, Tomek. "ChatGPT Search Result Providers." Peec AI, September 2026. Finds ChatGPT uses its own index plus at least eight external providers, including scraped Google and Google Maps results, licensed Yelp and TripAdvisor data and Microsoft's Web IQ grounding platform, and concludes "Bing is not a proxy." https://docs.peec.ai/research/chatgpt-search-result-providers
¶¶
Mills, Madison. "Exclusive: Yelp Partners with ChatGPT to Surface Reviews." Axios, July 23, 2026. Yelp licenses reviews, ratings, photos and business details to OpenAI with Yelp branding and links in ChatGPT, a "Request a Quote" feature is coming, and Yelp already supplies data to Apple Maps. https://axios.com/2026/07/23/yelp-reviews-chatgpt-geo-partnership
***
Rollison, Damian. "How Does ChatGPT Conduct Local Searches?" Search Engine Land, accessed October 26, 2026. In a documented test, ChatGPT scanned the first 20 to 30 Bing results, examined five to eight sources and chose three to five "verifiable and linkable" sources with star ratings, without seeing Bing Places profile data. https://searchengineland.com/how-does-chatgpt-conduct-local-searches-454894
†††
Paget, Sammy. "Uncovering ChatGPT Search Sources." BrightLocal, December 12, 2024. 800 local searches across 20 industries and 20 cities in November 2024: business websites 58 percent of sources, business mentions 27 percent, directories 15 percent; Three Best Rated 24 percent and Expertise 18 percent of directory citations; no forum results. https://www.brightlocal.com/research/uncovering-chatgpt-search-sources/
‡‡‡
Harmo. Summary of Whitespark's 2026 Local Search Ranking Factors, accessed October 26, 2026. Reports that "Dedicated Page for Each Service" ranked first for local organic and second for AI visibility. https://harmo.me/en/blog/local-seo-ranking-factors-2026-whitespark
§§§
Murphy, Rosie. "Local Consumer Review Survey 2026." BrightLocal, February 11, 2026. Survey of 1,002 U.S. adults: Apple Maps rose from 14 to 27 percent and Google fell from 83 to 71 percent as review sources; local news sites fell from 48 to 29 percent; consumers use six review sites on average, with Facebook, Tripadvisor, BBB, Trustpilot, Healthgrades, Yellow Pages and Angi rising; 82 percent read AI review summaries and 23 percent would rely on them alone; 56 percent seek consistent sentiment; 74 percent care only about reviews from the last three months; 47 percent avoid businesses with fewer than 20 reviews; 31 percent require 4.5 stars, up from 17 percent. https://www.brightlocal.com/research/local-consumer-review-survey/
‖‖‖
Ellis, Miriam. "Want to Rank in ChatGPT? Focus on These Review Sites (New Research)." Whitespark, September 15, 2025. 153 queries across 17 categories in nine cities: Facebook appeared on nearly 1.5 times as many Bing listings as Yelp, Yelp more than twice as often as TripAdvisor, Porch led garage door repair, and the Bing-to-ChatGPT link was unconfirmed. https://whitespark.ca/blog/want-to-rank-in-chatgpt-focus-on-these-review-sites-new-research/
¶¶¶
Murphy, Rosie. Guide to Google Business Profile Q&A and Ask Maps. BrightLocal, last updated April 2026. Google retired the Q&A API on November 3, 2025, and replaced public Q&A with Ask Maps. https://www.brightlocal.com/learn/google-q-and-a/
****
Apple. "Introducing Apple Business: A New All-in-One Platform for Businesses of All Sizes." Apple Newsroom, March 24, 2026. Reports that Apple Business, free in the U.S. and more than 200 countries and regions from April 14, 2026, replaces Apple Business Connect, Apple Business Manager and Apple Business Essentials, and lets businesses customize Apple Maps place cards with photos, location details and hours, add Showcases and action links such as order or reserve, and reach local customers across Apple Maps, Mail, Wallet and Siri. https://www.apple.com/newsroom/2026/03/introducing-apple-business-a-new-all-in-one-platform-for-businesses-of-all-sizes/
††††
Chapekis, Athena, and Anna Lieb. "Google Users Are Less Likely to Click on Links When an AI Summary Appears in the Results." Pew Research Center, July 22, 2025. 900 adults and 68,879 searches: AI summaries on 18 percent of searches; a traditional result clicked in 8 percent of visits with a summary versus 15 percent without; Wikipedia, YouTube and Reddit made up 15 percent of sources cited in AI summaries. https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/
‡‡‡‡
Tong, Anna, Echo Wang, and Martin Coulter. "Reddit in AI Content Licensing Deal with Google, Sources Say." Reuters, February 22, 2024, excerpted by Nieman Lab. The contract is worth about $60 million per year, according to one source. https://www.niemanlab.org/reading/reddit-in-ai-content-licensing-deal-with-google-sources-say/
Previous
Previous

A Sign at Every Exit: Google Maps vs AI Overviews for Local Business Leads

Next
Next

Recommended by Duncan Hines: How to Get More Google Reviews Without Annoying Customers