ChatGPT cites dated best-of lists far more often when the question names a big city. In the Cheers Market Baseline, our own research panel, it cited one in 54.0% of October 2026 answers about the ten biggest US cities and 29.9% about ten mid-size US cities. Build the source plan for each market you serve. The two-name test below shows how the core city's answers differ from those for a neighboring town.
Last verified: November 1, 2026.
Our smallest US city has about 315,000 people, so the small-town numbers are Canadian, from towns under 100,000. In the US the share fell from the 10 biggest cities, each with more than a million people, to 29.9% in 10 mid-size cities of about 315,000 to 422,000; in Canada it ran 48.9% in cities over 500,000.
The shortlist stays about the same length: ChatGPT named 5.04 businesses per answer in the 10 biggest US cities, 4.95 in the mid-size cities and 4.57 in Canadian towns under 100,000. The sources are what shift: in smaller places ChatGPT cites BBB and Angi more often, and Google AI Mode cites the businesses' own websites more.
Our window, October 9 to 31, 2026, starts after ChatGPT's layout change: Sistrix found ChatGPT switched to a building-block layout overnight on October 8 to 9, 2026, and used it in 97% to 100% of sampled answers from October 23 (October 30, 2026). This study compares source patterns across the cities and towns in our panel after that layout change.
Big-city answers cite dated best-of lists
For this study a dated list is any cited page with "best" or "top" plus 2025 or 2026 in its title or URL, the "best [trade] in [city] 2026" kind of page. Between the top 10 and the mid-size group, ranks 11 to 25 ran 50.6% and ranks 26 to 50 ran 47.5%. Canada's 12 cities of 100,000 to 500,000 came in at 26.2%, and its small-town share rose to 16.8% in the window's last week.
Cheers Market Baseline, October 2026
ChatGPT cites dated best-of lists far more in big cities
Share of ChatGPT answers citing a dated best-of list
US, panel's 10 biggest cities
US, panel ranks 11 to 25
US, panel ranks 26 to 50
US, 10 cities of 315,000 to 422,000
Canada, over 500,000 people
Canada, 100,000 to 500,000
Canada, under 100,000
Cheers Market Baseline: 10,507 ChatGPT answers to 32 home-service questions in 100 US and Canadian cities, October 9 to 31, 2026. Cut November 1, 2026.
| City size band | Lowest city | All answers | Highest city | All answers, without ConsumerAffairs' lists |
|---|---|---|---|---|
| US, panel's 10 biggest cities | 45.3% | 54.0% | 61.9% | 43.6% |
| US, panel ranks 11 to 25 | 40.0% | 50.6% | 68.6% | 36.8% |
| US, panel ranks 26 to 50 | 30.5% | 47.5% | 61.0% | 30.1% |
| US, 10 cities of 315,000 to 422,000 | 14.2% | 29.9% | 50.5% | 26.2% |
| Canada, over 500,000 people | 23.1% | 48.9% | 60.4% | 48.9% |
| Canada, 100,000 to 500,000 | 2.9% | 26.2% | 57.1% | 26.2% |
| Canada, under 100,000 | 1.9% | 11.4% | 36.2% | 11.4% |
The gap holds city by city: every one of the top 10 came in at 45% or more and 10 of the 18 Canadian towns sat under 10%, while the US mid-size group ran from 14.2% (Santa Ana) to 50.5% (Bakersfield, the outlier) and Canada's middle group from 2.9% to 57.1%. With about 105 answers behind each city, one city's share can move about 10 points either way, so treat the band as the finding.
One publisher accounts for part of the US gap. Dated lists from ConsumerAffairs, a review site, appeared in 28% to 29% of answers about the panel's top 50 US cities by rank and 6.4% about the mid-size group. Take them out and the slope narrows but holds: 43.6% in the top 10, 36.8% and 30.1% in the next two groups and 26.2% in the mid-size group. Canadian answers cited no ConsumerAffairs dated list, so Canada's numbers don't change.
US and Canada together, 84.4% of the ChatGPT answers that cited a dated list had it on a directory or best-in-city list site. My read is that list publishers cover the city names people search most; we didn't count the lists themselves.
Google AI Mode rarely cites a dated list in any band (5.3% to 9.6% of US answers, under 4% in Canada), and Gemini shows ChatGPT's slope at lower levels (21.7% in the top 10, 14.0% in the mid-size group).
Ask about the town next door and the lists often thin out
The sharpest gaps sit inside single metros, clearest in Alberta. ChatGPT cited a dated list in 55.2% of Calgary answers against 10.2% for its commuter towns of Airdrie, Cochrane and Okotoks, and Edmonton ran 60.4% against 14.8% for Leduc, Sherwood Park, Spruce Grove and St. Albert; both gaps held in each half of the window. At 12.8%, the seven commuter towns looked like the panel's 11 other small Canadian towns (10.5%), not like the big city next door.
Bigger neighboring cities split. Five sat 20 points or more below their core city: Santa Ana, the mid-size group's low, and Anaheim (26.9%) against Los Angeles (49.5%), Surrey (23.1%) and Burnaby (26.7%) against Vancouver (53.8%), and Aurora (26.7%) against Denver (49.1%). Long Beach (39.0%), Brampton (43.8%, against Toronto's 50.5%) and Henderson (37.1%, against Las Vegas' 51.4%) landed in between. Mississauga (49.5%) held level with Toronto and Mesa (42.5%) with Phoenix (45.3%), while Oakland (46.7% against San Francisco's 41.0%) and Arlington (54.3% against Dallas' 46.7%) came in a few points above their core cities, close enough that I wouldn't read either as a rise.
Same metro, different city name
Commuter towns drew far fewer dated lists; bigger neighbors split
Share of ChatGPT answers citing a dated best-of list
Left: Core city. Right: Neighbor named in the question. Every panel uses the same scale.
Calgary to Airdrie, Cochrane and Okotoks
105 and 315 answers
Edmonton to Leduc, Sherwood Park, Spruce Grove and St. Albert
106 and 419 answers
Los Angeles to Santa Ana
105 and 106 answers
Vancouver to Surrey
106 and 104 answers
Vancouver to Burnaby
106 and 105 answers
Los Angeles to Anaheim
105 and 104 answers
Denver to Aurora
106 and 105 answers
Las Vegas to Henderson
Reversed in the final week
Los Angeles to Long Beach
105 and 105 answers
Toronto to Brampton
105 and 105 answers
Phoenix to Mesa
106 and 106 answers
Toronto to Mississauga
105 and 105 answers
San Francisco to Oakland
105 and 105 answers
Dallas to Arlington
105 and 105 answers
Cheers Market Baseline: ChatGPT answers to 32 home-service questions, October 9 to 31, 2026. Pairs share a US Census or Statistics Canada metro area; Okotoks sits in the Calgary Metropolitan Region just outside it. Cut November 1, 2026.
| Core city to neighbor | Core city | Neighbor named in the question | Answers (core city and neighbor), or note |
|---|---|---|---|
| Calgary to Airdrie, Cochrane and Okotoks | 55.2% | 10.2% | 105 and 315 answers |
| Edmonton to Leduc, Sherwood Park, Spruce Grove and St. Albert | 60.4% | 14.8% | 106 and 419 answers |
| Los Angeles to Santa Ana | 49.5% | 14.2% | 105 and 106 answers |
| Vancouver to Surrey | 53.8% | 23.1% | 106 and 104 answers |
| Vancouver to Burnaby | 53.8% | 26.7% | 106 and 105 answers |
| Los Angeles to Anaheim | 49.5% | 26.9% | 105 and 104 answers |
| Denver to Aurora | 49.1% | 26.7% | 106 and 105 answers |
| Las Vegas to Henderson | 51.4% | 37.1% | Reversed in the final week |
| Los Angeles to Long Beach | 49.5% | 39.0% | 105 and 105 answers |
| Toronto to Brampton | 50.5% | 43.8% | 105 and 105 answers |
| Phoenix to Mesa | 45.3% | 42.5% | 106 and 106 answers |
| Toronto to Mississauga | 50.5% | 49.5% | 105 and 105 answers |
| San Francisco to Oakland | 41.0% | 46.7% | 105 and 105 answers |
| Dallas to Arlington | 46.7% | 54.3% | 105 and 105 answers |
The clip is our own silent ad, "The Neighborhood."
Watch · 17 seconds
More places to be found than SEO covers.
A silent animated town showing where customers look besides search: Maps, reviews, ChatGPT and local pages.
In smaller places, Google AI Mode cites the business's own site more
For Google AI Mode, the number that moves is how often it cites the website of a business on its map cards. That share rose band by band, from 37.3% of answers in the 10 biggest US cities to 47.7% in the mid-size group, and in Canada from 47.6% in cities over 500,000 to 59.9% in towns under 100,000. Only domains matching one on that answer's cards count, so treat these shares as a floor.
Phone numbers move mostly at the small end. AI Mode printed one in 10.3% of answers in the 10 biggest US cities, about the same in the next two groups, and 14.1% in the mid-size group, while in Canada the share climbed from 10.5% in cities over 500,000 to 24.6% in towns under 100,000.
ChatGPT is flat on both, so the own-site shift is a Google AI Mode result. In every group ChatGPT cited the site of a business on its map cards in 76.9% to 83.3% of answers and printed at least one phone number in 92.2% to 95.9%, which doesn't mean every business it named got one.
Google AI Mode, October 2026
In smaller places, AI Mode cites the business's own site more
Share of Google AI Mode answers
Left: Biggest places. Right: Smaller places. Every panel uses the same scale.
Cites a carded business's own website, Canada
Cites a carded business's own website, US
Prints a phone number, Canada
Prints a phone number, US
Cheers Market Baseline: 10,507 Google AI Mode answers to 32 home-service questions in 100 US and Canadian cities, October 9 to 31, 2026. Cut November 1, 2026.
| City size band | Cites a carded business's own website | Prints a phone number |
|---|---|---|
| US, panel's 10 biggest cities | 37.3% | 10.3% |
| US, panel ranks 11 to 25 | 42.6% | 10.1% |
| US, panel ranks 26 to 50 | 46.3% | 10.9% |
| US, 10 cities of 315,000 to 422,000 | 47.7% | 14.1% |
| Canada, over 500,000 people | 47.6% | 10.5% |
| Canada, 100,000 to 500,000 | 50.3% | 14.7% |
| Canada, under 100,000 | 59.9% | 24.6% |
ChatGPT's directory mix changes with size. BBB rose a little in the US, from 19.0% of top-10 answers to 23.8% in the mid-size group, and climbed in Canada from 28.9% in cities over 500,000 to 35.3% and then 44.0% in towns under 100,000. Angi or HomeAdvisor went from 69.6% to 81.0% across the same two US groups, while Yelp ran the other way, cited most in the top 10 at 39.3%.
Which page gets cited? In an earlier window of the same panel, 66.9% of AI Mode's citations to business websites went to homepages (AI local ranking factors 2026). We think they arrive through each Business Profile's website button, which the data can't confirm, so that linked page most needs the town detail.
Nearly one AI Mode map card in five names another town, and the share swings by metro
A card counts here when the asked city's name isn't in its address, so suburbs count, and so do big-city neighborhoods with their own mailing names, like Brooklyn. By that test Google AI Mode put another town's name on 18.4% of US map cards and ChatGPT on 14.9%; 95.0% and 96.2% of those cards named a town in the same state.
The share swings hard by city, from 0.3% in Bakersfield to 58.2% in Boston. Cleveland ran 53.1%, Detroit 48.9%, Milwaukee 42.6% and Minneapolis 39.4%, and in all four the 12 most common towns on those cards were separate municipalities, such as Parma outside Cleveland and Ferndale outside Detroit.
Google AI Mode map cards, US, October 2026
Cards naming another town ran from 0.3% to 58.2% by city
Share of map cards whose address names a town other than the one asked
Includes Boston neighborhoods such as Dorchester; about 46% of cards named a separate town
Cheers Market Baseline: 24,681 Google AI Mode map cards in 60 US cities, October 9 to 31, 2026. Cut November 1, 2026.
| City asked about | Cards naming another town (%) |
|---|---|
| Boston | 58.2% (Includes Boston neighborhoods such as Dorchester; about 46% of cards named a separate town) |
| Cleveland | 53.1% |
| Detroit | 48.9% |
| Milwaukee | 42.6% |
| Minneapolis | 39.4% |
| All 60 US cities | 18.4% |
| Colorado Springs | 2.2% |
| Corpus Christi | 2% |
| Tucson | 1.5% |
| Albuquerque | 0.9% |
| Bakersfield | 0.3% |
Boston's share includes in-city neighborhoods such as Dorchester; about 46% of its cards named a separate town, such as Quincy. Four more cities ran high but sit outside the chart, three of them mostly on places inside the city.
- New York, 42.3%: 147 of 156 were Brooklyn, Queens, Staten Island or Manhattan addresses, leaving at most 2.4% of its cards outside the city.
- Los Angeles, 40.7%: mostly city neighborhoods such as North Hollywood; 14.0% of all its cards named another city, such as Burbank.
- Honolulu, 33.4%: almost all Oahu places inside the City and County of Honolulu.
- New Orleans, 33.1%: its 12 most common card towns, such as Metairie, all sat outside the city.
My read, and we didn't measure it, is that older core cities ringed by separate suburbs sit near the top, while cities whose limits take in much of their metro sit near the bottom: Colorado Springs (2.2%), Corpus Christi (2.0%), Tucson (1.5%), Albuquerque (0.9%) and Bakersfield.
Google's rule, not ranking, settles whether to show an address: if customers don't come to it, Google says to remove it and set up to 20 service areas by city, postal code or another area. Whether a branch should hide its address, and how to show the towns it covers, is in how service-area businesses show coverage.
Franchise territories and lead routing
A franchise territory is a line in an agreement; the map card shows whatever town is in the business's address, and we didn't measure how many other-town cards cross a territory. That address has to be a real place your team works from. Google treats virtual offices and mail drops as ineligible, so a suburb profile opened just to get its name on a card can cost you the profile.
The part you control is what happens when a call lands at the wrong location, so write it down on one page that answers five things:
- Who answers when the call reaches a location that doesn't serve that town.
- How fast the handoff happens, in minutes, and who confirms the right location picked it up.
- Who books the job, so the customer never has to call a second number.
- How the lead source gets logged.
- How the receiving location gets credit, so nobody has a reason to sit on a transferred call.
Messaging needs an owner too. Search Engine Roundtable reported on October 28 that Google asked some businesses for an EIN and a privacy-policy link when they updated parts of their Business Profile (one commenter hit it while setting up the message button), and Google hasn't published a help page on it. If franchisees own their profiles, agree on who handles that request before anyone turns messaging on.
All of this is operating advice, not legal advice on territory clauses.
Our AI visibility grader, a separate tool that queries ChatGPT, Gemini and Perplexity through their developer APIs, found national franchise and big-brand names in 2.9% of the 30,759 competitor names in its answers (repeats counted, March 17 to October 30, 2026). It measures a different surface, so read it on its own (setup in Method).
What to check this week
Big-city locations
Ask ChatGPT your top job question for the city, open the sources and work through every dated "best of 2026" list for your trade, correcting anything wrong it prints about you, like your phone number or the towns you serve. If a list says how it picks businesses (review scores, an editor, a submission form or a fee), read that before you contact anyone.
Prioritize accurate inclusion on the lists customers and assistants already consult. If a publisher offers paid placement, assess the placement, disclosure and referral value separately from its appearance in AI citations.
Suburb and small-town locations
Make the page your Business Profile links to say which towns that branch serves, with real jobs, the crew and response times. A town page only helps if it says something true about that town; otherwise a service-area hub does more, the line building city pages without doorway risk draws. Check your BBB profile too, especially in Canada, where ChatGPT's BBB citations climbed as towns got smaller.
Every location
Check that the phone number on the profile and the site rings at that branch, since both engines print numbers and AI Mode prints more in smaller places.
Then run the two-name test. Ask ChatGPT and Google AI Mode your two most valuable job questions, once naming the core city and once a town next door you serve, with every other word the same: "best water heater installer in Calgary, AB," then "best water heater installer in Airdrie, AB." Log the dated lists, directories and business sites each answer cites and the town on each map card, one row per location (or per town if you run one shop). Where the city answer cites dated lists and the town answer doesn't, that location needs both blocks above: the list fixes for the city, and the page and BBB fixes for the town.
Method
The Cheers Market Baseline
The Cheers Market Baseline is our own research panel. Every week it asks the same 32 home-service questions in 100 cities, 60 in the US and 40 in Canada, and each question names the city. This cut, made November 1, covers October 9 to 31, 2026: 10,507 answers each from ChatGPT and Google AI Mode (6,307 US, 4,200 Canadian), with Gemini as a supporting line and no Perplexity.
US bands follow panel rank: the 10 biggest cities, ranks 11 to 25, ranks 26 to 50 and ranks 91 to 100 (Anaheim, Aurora, Bakersfield, Cleveland, Corpus Christi, Henderson, Honolulu, New Orleans, Riverside and Santa Ana, about 315,000 to 422,000 people each as of July 1, 2025). They're ranks, not clean size tiers: Wichita (rank 50) has about 401,000 people. Canadian bands use 2021 Census population: over 500,000 (10 cities), 100,000 to 500,000 (12) and under 100,000 (18).
Pairs. A neighbor sits in the core city's US Census or Statistics Canada metro area, plus Okotoks, just outside Statistics Canada's Calgary metro in the Calgary Metropolitan Region. Paired cities rest on 104 to 106 answers each and the Calgary and Edmonton town groups on 315 and 419. Riverside (its own metro) and Fort Worth (a million-person city itself) aren't paired.
Bases. Card shares rest on 24,681 AI Mode and 20,808 ChatGPT US cards, and businesses named per answer are averaged over answers where ChatGPT presented businesses as entries, 82% to 88% of each group.
Stability. The US slope held in both halves of the window, 54.2% against 29.6% from October 9 to 23 and 53.8% against 30.5% after, while Henderson went from 29.4% to 51.4% and Las Vegas from 54.4% to 45.9%. From October 2 to 8, before the layout change, it already ran 44.5% against 26.6%.
The 66.9% homepage share comes from the panel's August 31 to September 20, 2026 window (34,742 answers on ChatGPT, Gemini, Perplexity and Google AI Mode).
Limits. Bands hold 10 to 25 cities, and the small-town tier is Canadian and mostly Albertan. The dated-list test can miss lists without a year and catch "top-rated" pages that aren't lists, and a pattern by size is not proof that size causes it.
The Cheers AI visibility grader
The Cheers AI visibility grader sends its questions to ChatGPT, Gemini and Perplexity through each company's developer API, not through the consumer apps. Web search is on for every question: ChatGPT runs its web search tool, Gemini grounds its answer in Google Search, and Perplexity searches the web by design. Each engine gets the same short instruction to answer the way a helpful assistant would for a real buyer, to name only real businesses and to say when the evidence is thin. For each business we ask four buyer questions built from its category and city (for example, "Who are the best [trade] companies in [city]?") and one question that names the business and asks what customers say about it; until early October, that question asked what customers praise or complain about. Every question goes out as a fresh request with no signed-in user, chat history or personalization, and we store the answer text and the links it cites exactly as returned. Since September 23, ChatGPT and Perplexity also receive the business's city as the asker's approximate location, and in late September we changed the wording of the buyer questions. Between March 18 and June 8, ChatGPT answered the four buyer questions for about a third of businesses in one combined request; we retired that method. Answers from developer APIs can differ from what a signed-in person sees in the apps, which add map cards, phone numbers and business listings that the APIs don't return. Engine models change over time; this sample spans March 17 to October 30, 2026, Pacific time.
The 2.9% covers buyer-question answers about 455 graded businesses (5,456 answers), matched against an incomplete list of about 100 national names.
Sources
- What changed in ChatGPT in October: GPT-6, UTM tracking and sources, Sistrix, October 30, 2026. ChatGPT's October 8 to 9 layout change.
- AI Citations, User Locations, & Query Context, Yext Research, October 9, 2025. City-size research listed as planned, not published.
- Manage your service areas for service-area & hybrid businesses, Google Business Profile Help, undated. The address rule and up to 20 service areas.
- Google Business Profiles Asking Businesses For EIN, Search Engine Roundtable, October 28, 2026. The EIN and privacy-policy request.
Sources read November 1, 2026.
Dylan Allen-Arnegård is the CEO & Co-Founder of Cheers, the done-for-you local visibility partner for established home-service companies and PE-backed portfolios.