AI Local Ranking Factors 2026: What 34,742 Answers Show

The AI engines we measured named local businesses with strong ratings and deep review histories, then built their answers from different evidence: ChatGPT and Perplexity leaned on directories and "best of" lists, while Google AI Mode leaned on Google's own data, business websites and booking pages. In the Cheers Market Baseline, 34,742 answers from ChatGPT, Gemini, Perplexity and Google AI Mode to 32 home-services questions in 100 US and Canadian metros between August 31 and September 20, 2026, only 3.7% of Google AI Mode answers named a business under 4.0 stars, ChatGPT cited a directory, review platform or best-of list in 84% of its answers against 38% for Google AI Mode, and Google AI Mode linked an online booking page in 31% of its answers while ChatGPT never did. For a brand with 40 locations, that means the rating has to hold on every engine, and most of the remaining work splits by engine.

Last verified: September 28, 2026.

Most ranking-factor lists for AI search come from surveys and small tests. Whitespark's 2026 Local Search Ranking Factors report, published November 6, 2025, asked 47 local search experts to score 187 factors and added an AI search visibility category for the first time; BrightLocal's July 2025 study ran 20 searches across 10 industries on four AI platforms. This page measures what the engines did instead, three weeks of it, factor by factor, with a link to the full analysis of each. The matrix at the top puts six of the measures side by side, and if you'd rather see where one of your own locations stands first, the free AI Visibility Grader runs a check across ChatGPT, Gemini and Perplexity.

Ratings: a floor almost nobody went under

Google AI Mode showed a star rating for 89.6% of the businesses it named, ChatGPT for 85.3% during the days we captured its map (September 18 to 22) and Gemini for 53.3% (September 14 to 22); we captured no rating cards from Perplexity. The named field was highly rated. Of the businesses Google AI Mode named with a rating, 83% sat at 4.8 stars or above and 1.0% under 4.0, and only 3.7% of its answers named anyone under 4.0. We never see the businesses it passed over, so that isn't proof of a filter, but it shows what the shortlist looks like.

The top of the scale was flat. Among businesses with 100 or more reviews, 14.5% of those at a perfect 5.0 were named in all three weekly checks against 15.8% at 4.9, while businesses rated 4.0 to 4.4 managed 6.7%. Protect the rating at every location before you worry about the last tenth of a star; do Google reviews affect AI recommendations breaks the review picture out by engine.

Review depth: more reviews went with first place and staying power

The median business Google AI Mode named had 226 Google reviews; 8% had fewer than 20 and 21% had 1,000 or more. Depth was a softer floor than rating: almost a quarter of Google AI Mode answers (23.5%) named at least one business with fewer than 20 reviews, so a new or small location isn't locked out.

Depth showed up in order and repetition. In answers naming three or more rated businesses, the first one named had a median of 342 reviews, the second 237 and the third 195, and the most-reviewed business came first 36% of the time against about 25% by chance. Of the businesses Google AI Mode named at least once for a question and city, those showing 1,000 or more reviews came back in at least two of three weekly checks 43% of the time, against 26% for those under 20. Review count also tracks a company's size and age, so treat this as a pattern, not a lever with a known payoff. Medians by trade and a pace plan per branch are in how many Google reviews you need for AI recommendations.

Sources: lists for ChatGPT and Perplexity, websites for Google AI Mode

The engines split hardest on sources. ChatGPT cited a directory, review platform or ranked best-of list in 84.2% of the answers where it cited anything and Perplexity in 93.5%, against 38.3% for Google AI Mode and 55.8% for Gemini (in the answers where we captured its citations). ChatGPT cited Expertise.com in 32.1% of its answers and Angi or HomeAdvisor in 29.4%. Google AI Mode cited Google's own properties in 95.5% of its answers and a business's own website in 92.5%; Perplexity cited business websites in 89.0% of its answers and Gemini in 88.3% of those with captured citations, while ChatGPT did in 44.2%. A location can look strong on Google AI Mode through its profile and website and still be missing from the lists ChatGPT reads.

That sharpens the experts' top pick. Whitespark's panel ranked "Presence of Business on Expert Curated Best of and Similar Lists" first among its AI search visibility factors. In our data, ranked best-of list sites appeared in 69.6% of Perplexity's answers with citations and 64.2% of ChatGPT's, but in 30.1% of Gemini's and 4.8% of Google AI Mode's, so the experts' pick held on two of the four engines.

A citation shows which lists an engine read, not whether being on one got a business named, so treat each list as a record to check: is the location on it, with the right phone number, hours and service area? The border changes the list, too. Angi and HomeAdvisor appeared in 1 of 3,775 Canadian ChatGPT answers while HomeStars appeared in 39.9% of them, and Expertise.com appeared only in US answers, so a brand with branches in Denver and Calgary is managing two directory lists. The full breakdown is in where ChatGPT gets its information, the ChatGPT playbook is how to get ChatGPT to recommend your business, and the Google side is what Google AI Mode shows local businesses.

Yelp: rarely cited by ChatGPT, often on its map

BrightLocal found Yelp among the sources in 33% of its 20 searches. At 34,742 answers, the Yelp picture depends on the engine and on where you look: as a cited source, Yelp appeared in 19.7% of Google AI Mode's answers, 13.9% of Perplexity's and 0.6% of ChatGPT's. ChatGPT's map told a different story. During the five days we captured it, September 18 to 22, 34% of its rated business cards linked to Yelp and 72% of its answers with rated cards showed at least one, 46.4% of cards in US metros against 12.8% in Canadian ones.

For 1,624 businesses we could match in the same city, ChatGPT's Yelp cards averaged 4.3 stars against 4.9 on Google AI Mode, with a median of 31 reviews against 469, while ChatGPT's other cards matched Google AI Mode's rating within 0.1 star 98% of the time. When a ChatGPT answer showed any business under 4.0 stars, a Yelp card was involved 98% of the time (617 of 630 answers). The fixes that fit Yelp's own rules are in why ChatGPT shows a lower star rating than Google.

Stability: the list moves every week, and the engines disagree

A ranking factor implies a stable ranking, and these answers weren't stable. Ask Google AI Mode the same question for the same city a week later and 41.2% of the businesses it named come back; Gemini brought back 42.6% and ChatGPT 57.7%. Over three weekly checks, Google AI Mode and ChatGPT each named a median of 8 different businesses for one question and city, and only 12% of Google AI Mode's businesses and 25% of ChatGPT's appeared all three weeks. Perplexity is left out here, because a collection outage left it too few week pairs.

The engines keep separate shortlists as well. When ChatGPT and Google AI Mode answered the same question for the same city in the same week, they shared about one business on average (1.04), and across all four engines 77% of the businesses named on the same occasion were named by only one engine. One screenshot is one draw, so measure the share of repeated checks that name each branch, by engine. Why AI recommends different businesses every time has the full analysis, and what a good AI visibility score looks like covers turning repeated checks into one number.

Website pages: homepages and service pages, almost never blogs

Of 74,969 citations to contractor websites, 43.1% went to homepages and 48% to service, location or service-plus-city pages (17.1%, 16.9% and 14.0%); blog posts got 1.5% and review or testimonial pages 0.7%. Google AI Mode sent 66.9% of its website citations to homepages, which we think arrive through the website button on each Business Profile (an inference the data can't confirm), while Perplexity cited service-plus-city pages more than any other type, at 28.6%. A Tucson plumbing branch with a real page for drain cleaning in Tucson gives Perplexity and Gemini something to cite; a blog post on five signs of a clogged drain mostly doesn't. The page spec is in what location pages should include for AI search.

Booking links: a Google AI Mode factor

Google AI Mode linked an online booking page in 31.2% of its answers (2,947 of 9,435), meaning scheduling pages on field-service software: Housecall Pro in 20.5% of its answers, ServiceTitan in 10.6% and Jobber in 3.1%. ChatGPT never did, Gemini didn't in any of the answers where we captured its citations, and Perplexity did in 0.2%. The rate followed the trade: 39.4% of plumbing answers, 38.9% of HVAC, 33.3% of electrical and 1.8% of roofing.

Our read, which the data can't confirm, is that these links come from the booking link on each location's Google Business Profile. The check is quick: press the booking button on every branch profile and confirm it opens that branch's scheduler, not a corporate page with the wrong city selected. The garage door guide shows the split on a single question, with booking links in 31.3% of Google AI Mode's answers and none of ChatGPT's.

Map cards and emergency language

Google AI Mode showed rated business cards in 90% of its answers, and the share barely moved with the question: 89.7% for "best X for Y" questions, 91.6% for emergencies and 93.3% for deals and quotes. ChatGPT showed cards in about 9 in 10 of its answers in the days we captured them.

The language moved instead. On emergency questions ("right now", "stopped working", "available today"), 98.8% of ChatGPT's answers, 98.2% of Perplexity's, 97.7% of Gemini's and 92.8% of Google AI Mode's mentioned 24/7 or after-hours service, against 34% to 61% of answers to "best X for Y" questions, and on deal and quote questions ChatGPT put a dollar figure in 62% of its answers and Gemini in 68%. Mentions include general advice, so this shows what answers talk about, not why a company was picked; it still tells a branch what to state plainly, from real after-hours coverage to prices, financing and warranty terms. The HVAC guide splits this question by question, and the electrician guide shows emergency and planned work pulling apart.

National brands: in a quarter of answers, a small share of names

National franchise brands appeared in 23.8% of the answers that named a business but filled 7.5% of the named slots and were named first in 8.3% of answers. ChatGPT included one in 21.1% of its answers and gave them 5.0% of its slots; Google AI Mode included one in 27.2%, Perplexity in 24.0% and Gemini in 22.9%. The trade decided most of it: only 0.9% of roofing answers named a national brand, against 9.3% of HVAC answers, 13.3% of electrical, 36.4% of plumbing, 60.1% on our pest control question and 69.4% on our water damage restoration question. We count brands as a category and never name one, and the shares lean high because some answers mention chains only to say avoid them. The engines name the location, not the logo, which is the argument of how franchise brands get recommended by ChatGPT.

Reddit: a small and very local source

Reddit made up about 1.2% of all citations. Perplexity cited it in 10.8% of its answers, Google AI Mode in 9.1%, ChatGPT in 4.9% and Gemini in 1.2% of the answers with captured citations, and 93% of the 2,836 Reddit links came from city or regional subreddits, against 4% from trade or homeowner communities. A branch that wants a place in those threads needs a real person answering questions in their own city's subreddit, affiliation disclosed, which is slow work with a low ceiling; what AI search cites from Reddit has the rules.

Trades: the same engines set a different bar

HVAC and plumbing rest on nine questions each in our set, electrical on six and roofing on two. HVAC had the deepest review histories: the median business Google AI Mode named had 449 reviews, and only 11.6% of HVAC answers named a business with fewer than 20. Plumbing sat close behind at 364 and 16.9%, while electrical and roofing had medians of 113 and 130, and 40.8% of electrical answers and 23.6% of roofing answers named someone with fewer than 20 reviews.

ChatGPT cited a contractor's own site in 53.7% of plumbing answers, 53.6% of electrical and 44.0% of HVAC, but in 17.8% of roofing answers, where it leaned on directories and lists. On our single water damage question, ChatGPT cited the BBB in 61.0% of its answers. The trade guides go question by question for HVAC, plumbing, electricians, roofing, pest control, garage doors and water damage restoration.

What we did not measure

A factor list is only as honest as its gaps, and these are the ones I'd want a reader to know before quoting any number above:

• Google Maps rank against AI answers. We did not compare where a business ranks in the local pack with whether an engine named it; that comparison needs more weeks of matched data than these three.
• The businesses that were not named. We see only the businesses the engines put in their answers, so we can't say what kept anyone else out.
• Cause. None of this was an experiment, and review count in particular also tracks a company's size and age.
• Leads and revenue. Being named is not a phone call, and none of these figures measure calls, bookings or sales.
• Listing consistency, schema and llms.txt. We did not measure whether a business's name, phone and address agree across the web, or what markup its site carries. Google says "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add."
• Other engines and other questions. Claude, Copilot and Apple's assistants are outside this sample, as is every question beyond our 32.

What to do this week

Pick five locations and two questions a customer would ask about each, with the city written into the question. Run both on ChatGPT and Google AI Mode twice this week, on different days, and write down every business named. By Friday you'll have 40 answers, and for each location you can check four things:

• The star rating and review count each engine shows, including whether ChatGPT's map card for the location links to a Yelp page with a lower number.
• Whether the location appears, with the right phone number, hours and service area, on the directories and best-of lists ChatGPT cited.
• Whether the booking button on the location's Google Business Profile opens that location's own scheduler.
• Whether the location has a crawlable service page for the job in the question, in its city.

Fix them in that order, because the rating shows on every engine and the rest map to the engine they move.

Methodology

Every week, Cheers asks ChatGPT, Gemini, Perplexity and Google AI Mode the same 32 home-services questions in 100 metro areas across the United States and Canada, and records every business each engine names, with its star rating, review count and map position where the engine shows one, and every source it cites. This analysis covers the three weeks from August 31 to September 20, 2026: 34,742 usable answers (9,434 from ChatGPT, 9,435 from Google AI Mode, 9,364 from Gemini and 6,509 from Perplexity) across 60 US and 40 Canadian metros. We excluded empty or cut-off answers, most of them from a Perplexity collection outage. Each question was asked once per city per week, so week-to-week changes include normal answer-to-answer variation. Business names were extracted automatically; a hand check of 40 answers found the extraction caught about three in four of the businesses a reader would count, so counts are lower bounds. Star ratings and review counts are the figures each engine displayed; for ChatGPT and Gemini we captured them from September 18 and September 14 onward. Findings describe associations in this sample, not how any engine ranks businesses.

ChatGPT card figures come from 11,875 rated cards captured September 18 to 22, when about 23 of the 32 questions ran; a card counts as Yelp-sourced when it links to a Yelp business page, and the rating comparison matches 1,624 businesses by name within a metro. Directory and list shares use answers with at least one citation (9,365 ChatGPT, 9,390 Google AI Mode, 6,509 Perplexity and 4,141 Gemini answers), and every Gemini citation share uses only the 4,141 Gemini answers where we captured citations. Source groups come from hand-built domain lists, so the business-website share is a lower bound. Repeat figures compare consecutive weeks and exclude Perplexity. National brands are an internal category list, and no business, franchise location or Cheers customer is named. Pest control, garage door and water damage figures each come from one question.

Sources

Checked September 28, 2026.

• 2026 Local Search Ranking Factors. Whitespark, by Darren Shaw, published November 6, 2025. Survey of 47 local search experts scoring 187 factors, including the new AI search visibility category and its top factor, presence on expert-curated best-of lists.
• BrightLocal's study of the sources AI search uses for local businesses. BrightLocal blog, by Kate Herbert-Smith, July 22, 2025. Twenty searches across 10 industries on Google AI Mode, Gemini, Perplexity and ChatGPT, with Yelp among the sources in 33% of searches.
• Optimizing your website for generative AI features on Google Search. Google Search Central, last updated July 10, 2026. Source of the quoted line on structured data.
• Cheers Market Baseline, August 31 to September 20, 2026: 34,742 answers from ChatGPT, Gemini, Perplexity and Google AI Mode to 32 home-services questions in 100 US and Canadian metros, aggregate only. Research methodology. Source of every rating, review, source, Yelp, stability, page-type, booking, card, language, brand, Reddit and trade figure on this page.

Dylan Allen-Arnegård is the CEO & Co-Founder of Cheers, the local search platform for multi-location service businesses.