AI local search is changing fast, and service businesses need a current view of where customers are being sent. All eight source statistics we published in September moved when we measured the same metros again from October 9 to 31, 2026. This edition gives operators twelve dated measures they can use to prioritize profiles, phone numbers, website evidence and booking routes.
Last verified: November 1, 2026.
So we turned the measurement into an index. The Cheers Local AI Answer Index measures twelve things about local answers from ChatGPT, Gemini and Google AI Mode in this dated edition, across the 60 US metro areas of our research panel with the same 32 home-service questions each time, and it prints the window on every number. Edition one covers Oct. 9 to 31, 2026, the first stretch after ChatGPT's Oct. 9 layout change, so treat it as a baseline for a format that was still moving. Cite it as "Cheers Local AI Answer Index, edition one (US, Oct. 9 to 31, 2026)."
This week, open your sales deck, pitch page or board update and find every AI search statistic in it. Label anything about ChatGPT from before Oct. 9, 2026 with its dates, or swap in a number from Oct. 9 or later that carries its own window.
If you run a home-service company rather than report on one, do the listing and booking-page checks under "Where the answers send people" first, or run Cheers' free AI visibility check on your own business.
The index, Oct. 9 to 31, 2026, 60 US metros
Edition one, US, Oct. 9 to 31, 2026
Four numbers from the Local AI Answer Index
95.7%
of US ChatGPT answers with sources cited a directory, review platform or best-in-city list site
Base: 6,304 answers with captured citations
92.5%
of US ChatGPT answers printed at least one phone number
6,307 answers
29.3%
of US Google AI Mode answers linked a booking or scheduling page
1,847 of 6,307 answers
18.4%
of US Google AI Mode map cards named another town
4,532 of 24,681 cards
Cheers Market Baseline, 60 US metro areas, 32 home-service questions, answers from Oct. 9 to 31, 2026. Re-run Nov. 1, 2026.
| Finding | Value | Base |
|---|---|---|
| of US ChatGPT answers with sources cited a directory, review platform or best-in-city list site | 95.7% | 6,304 answers with captured citations |
| of US ChatGPT answers printed at least one phone number | 92.5% | 6,307 answers |
| of US Google AI Mode answers linked a booking or scheduling page | 29.3% | 1,847 of 6,307 answers |
| of US Google AI Mode map cards named another town | 18.4% | 4,532 of 24,681 cards |
The engines didn't move the same way in those weeks. ChatGPT used its building-block layout the whole time, though its citations kept shifting. Google announced five changes to how sources appear in AI Mode and AI Overviews on Oct. 15, yet our panel saw far smaller moves on AI Mode than on ChatGPT; our guide to what that means for local leads covers the announcement. Gemini is the hardest to read: in the answers our panel captured, 30.2% of its window answers showed place cards, and most of its cited answers came from Oct. 19 to 31, so its citation cells mostly describe late October.
Phone numbers, links and length
ChatGPT put at least one phone number in 92.5% of its US answers, against 14.5% for Gemini and 11.1% for Google AI Mode.
Printing a number doesn't mean it's the one on the business's own site. We compared the numbers ChatGPT printed in October with each business's own website, and 17.6% of the US numbers we could check weren't on the site in the pages we checked. We found no single-digit or transposition match among the confirmed nonmatches. We did not dial the numbers.
ChatGPT's answers also ran longest, at a median of 544 words and an average of 23.2 citation links. Gemini's ran 404 words and 2.5 links, while AI Mode's ran 262 words and 3.3 links.
How many businesses make the list
Map cards are the business listings that come with an answer's map. When they showed, ChatGPT averaged 4.22 per answer and Google AI Mode 4.21. We don't index Gemini on cards this edition, because in the answers our panel captured its place cards faded during the window and were gone by the week of Oct. 26.
Urgent questions shorten the list, though only two of them by much. Asked about a furnace that stopped working, ChatGPT named 3.22 businesses on average, against 4.97 on other HVAC questions; for a burst or frozen pipe it was 3.02 against 5.08 on other plumbing questions. The four "best emergency or same-day" questions sat within about 0.2 of their trade, which pins the short list on the two breakdown questions rather than on urgency in general.
The index counts how many businesses make the list, not which ones make it from week to week. That's in why AI recommends different businesses every time.
Where the answers send people
ChatGPT cited a directory, review platform or best-in-city list site in 95.7% of its US answers with sources, Google AI Mode in 49.3% and Gemini in 57.1%. A blended "AI" number hides a gap that size, so quote each engine on its own.
ChatGPT's ten most-cited sites were Angi or HomeAdvisor (76.3%), Expertise.com (69.8%), google.com (63.4%, mostly Google Maps links that came with its business listings), the review site ConsumerAffairs (46.7%), BestProsInTown (42.5%), the review site Birdeye (38.6%), ServiceAgent.ai (37.0%), Thumbtack (35.7%), Yelp (34.4%) and BBB (22.8%). BestProsInTown and ServiceAgent.ai are best-in-city list sites. No contractor or franchise website made the top 16.
Outside ChatGPT, Yelp turned up in 26.5% of AI Mode's cited answers and 0.0% of Gemini's. Reddit appeared in 6.6% of ChatGPT's, 6.5% of AI Mode's and 1.3% of Gemini's.
Google AI Mode linked a booking or scheduling page in 29.3% of its US answers (1,847 of 6,307), most often a Housecall Pro page (22.5% of answers; ServiceTitan 7.1%). ChatGPT did in 0.1% of its answers and Gemini in 0.0%.
This week, confirm the phone number, hours and service area on each branch's Google Business Profile and location page. Then do the same on any page Angi or HomeAdvisor (we count the two as one site) or Expertise.com already has for the branch, since those were ChatGPT's two most-cited sites.
If branches book through Housecall Pro, ServiceTitan, Jobber or Workiz, open each branch's public booking page yourself. Make sure it loads, lists the right services and lands on the right location.
Addresses on the map cards
ChatGPT showed no street address on 9.4% of its US map cards and AI Mode on 11.0%, which fits service-area businesses that hide theirs. More cards named another town: 18.4% of AI Mode's (4,532 of 24,681) and 14.9% of ChatGPT's 20,808. A card counts when its address doesn't contain the city in the question, so suburbs and big-city neighborhoods with their own mailing names are included.
So the card won't always tell a customer where a branch works, and the branch's service-area pages have to, as our service-area guide explains.
Whose website gets cited
In 78.1% of its US answers, ChatGPT cited the website of at least one business on that answer's own map cards, and AI Mode did in 44.1%. That squares with the top-16 list above: no single contractor or franchise site gets cited often enough to rank, yet most ChatGPT answers cite the site of a business they show.
A footnote on tone: of 70,624 business mentions our automated tone labels scored in US answers in October 2026, 29 were negative.
October answers differed from the September sample
Shelf life
All eight ChatGPT source shares we published in September were higher from Oct. 9 to 31
- google.com (mostly Google Maps links): 0.1% to 66.3%, mostly Google Maps links that came with ChatGPT's business listings, Oct. 9 to 29. On the same nine of our 32 questions, US answers with citations linking google.com fell from 85.7% (Oct. 16 to 17) to 2.2% (Oct. 30 to 31).
- Angi or HomeAdvisor: 29.4% to 45.9%
- Expertise.com: 32.1% to 41.9%
- BBB: 14.5% to 28.7%
- Yelp: 0.6% to 26.6% as a cited source; ChatGPT's map cards already linked to Yelp, 46.4% of rated US cards on Sept. 18 to 22
- HomeStars: 16.0% to 25.5%, cited in Canadian metros only
- Thumbtack: 5.0% to 21.5%, and already rising in September: 3.6 links per 100 runs in the week of Aug. 31, 7.6 in the week of Sept. 21
- Reddit: 4.9% to 9.6%
Cheers Market Baseline. Aug. 31 to Sept. 20, 2026: 9,434 ChatGPT answers, published Sept. 28, 2026. Oct. 9 to 31, 2026: 10,507 ChatGPT answers, re-run Nov. 1, 2026.
| Site | Aug. 31 to Sept. 20 | Oct. 9 to 31 |
|---|---|---|
| google.com (mostly Google Maps links) | 0.1% | 66.3% |
| Angi or HomeAdvisor | 29.4% | 45.9% |
| Expertise.com | 32.1% | 41.9% |
| BBB | 14.5% | 28.7% |
| Yelp | 0.6% | 26.6% |
| HomeStars | 16% | 25.5% |
| Thumbtack | 5% | 21.5% |
| 4.9% | 9.6% |
All eight ChatGPT shares in our study of where ChatGPT gets its local information, measured Aug. 31 to Sept. 20, were higher from Oct. 9 to 31 across the same 100 US and Canadian metros. The google.com share went from 0.1% of ChatGPT answers to 66.3%, mostly through Google Maps links. Angi or HomeAdvisor went from 29.4% to 45.9%, Expertise.com from 32.1% to 41.9%, BBB from 14.5% to 28.7%, Yelp from 0.6% to 26.6%, HomeStars (Canadian metros only) from 16.0% to 25.5%, Thumbtack from 5.0% to 21.5% and Reddit from 4.9% to 9.6%.
Some of that is arithmetic. ChatGPT's answers in our panel carried about 23 citation links each from Oct. 9 to 31, against 4.6 in the window we published, and more links lift every share. Sistrix's general sample rose only from 3.3 linked pages per answer to 4.1, and its author says he can't be sure the new model is the reason, so a jump that size showed up in our local panel and not in Sistrix's general one.
Two rows had a head start. Thumbtack climbed inside September, from 3.6 links per 100 ChatGPT runs in the week of Aug. 31 to 7.6 in the week of Sept. 21, while links per answer stayed flat. Yelp's September share counted citations only, and we'd already reported in September that 46.4% of rated US ChatGPT map cards linked to Yelp from Sept. 18 to 22.
Google AI Mode held steadier. Across all 100 metros its Yelp share sat at 19.7% in both windows and Housecall Pro went from 20.5% to 19.0%, though Angi or HomeAdvisor rose from 9.3% to 15.9%. For the same September window across four engines, Perplexity included, see our AI local ranking factors study of 34,742 answers.
The answers themselves changed shape between the two windows. By Sistrix's account, GPT-6 had been rolling out since Oct. 7, paying users first, and ChatGPT switched to a building-block layout overnight on Oct. 8 to 9, using it in 97% to 100% of sampled answers from Oct. 23. In our panel, 80.7% of US ChatGPT answers from Oct. 9 to 23 carried a Google Maps link, mostly directions links that came with the new per-business layout. Gemini changed too: in the answers our panel captured, its US place-card share went from 80.9% (Oct. 2 to 4) to 0.0% in the week of Oct. 26.
Nor did the citations settle once the new layout arrived. On the same nine of our 32 questions, a week apart, the share of cited US ChatGPT answers linking google.com went from 85.7% (Oct. 16 to 17) to 54.3% (Oct. 23 to 24) and then 2.2% (Oct. 30 to 31), and Angi or HomeAdvisor from 81.8% to 63.7% to 43.6%. AI Mode moved less, with its Yelp share at 22.9%, 25.4%, then 18.2%.
Week to week
ChatGPT's Angi or HomeAdvisor links more than doubled, then fell back
Angi or HomeAdvisor links per 100 ChatGPT runs
Cheers Market Baseline, 100 US and Canadian metro areas, weeks starting Aug. 31 to Oct. 26, 2026 (2,699 to 3,288 ChatGPT runs a week). Re-run Nov. 1, 2026.
| Week starting | Links per 100 runs |
|---|---|
| Aug 31 | 37.8 |
| Sep 7 | 37.2 |
| Sep 14 | 41 |
| Sep 21 | 44 |
| Sep 28 | 36.8 |
| Oct 5 | 52.3 |
| Oct 12 | 89 |
| Oct 19 | 76.6 |
| Oct 26 | 46.5 (Oct. 26 to 31 only) |
Pooled across all 100 metros, Angi or HomeAdvisor ran between 36.8 and 44.0 links per 100 ChatGPT runs in each of the five weeks starting Aug. 31, then jumped to 52.3 in the week of Oct. 5 and 89.0 in the week of Oct. 12 before falling back to 76.6 and, over Oct. 26 to 31, 46.5. That last week is also when Expertise.com dropped to 23.7 links per 100 runs, below every September week (25.5 to 33.5).
A monthly number brings those weekly movements together. Compare it with the weekly detail when deciding which source changes deserve attention. Other trackers have watched numbers move this fast: Promptwatch reported on Aug. 18, 2026 that Reddit's share of ChatGPT citations fell from 3.83% (July 18 to Aug. 7) to 0.52% (Aug. 14 to 17), and it called the size of that drop provisional.
How to use and cite the index
Cite the engine, the measure, the edition and the window together, and keep the base. Don't set these numbers beside our Home Services AI Visibility Index 2026: that index counts share of citations and this one counts share of answers, so their directory figures aren't comparable.
For a portfolio or multi-location brand, use matching windows and definitions when comparing a company's results with the index. Keep the October label on this edition and use fresh branch checks to guide current work.
This page records the October 2026 edition, with the measurement definitions below so readers can interpret and reproduce the comparisons.
Why we measure this
In this Cheers ad, I explain what we believe: AI tools will become the default way people choose their home-service providers. As assistants take on more of the research and coordination, a business needs to be ready to be found, evaluated and contacted through them.
That conviction is why we do this research. The index shows how the channel works today, so operators can act now and improve their position as it develops.
Watch · 12 seconds
Why we measure AI answers every week.
Dylan on AI becoming the default way people choose a home-service company.
Method
The Cheers Market Baseline is our own research panel: every week it asks ChatGPT, Google AI Mode and Gemini the same 32 home-service questions, once each, in 100 metro areas, 60 in the US and 40 in Canada. It holds no customer data. The questions span ten trades (HVAC, plumbing, electrical, roofing, water damage restoration, duct cleaning, garage door, pest control, windows and doors, lawn care). Edition one uses the 60 US metros from Oct. 9 to 31, 2026: 6,307 answers per engine, the base for every answer-level measure. We left out the Canadian metros and Oct. 2 to 8, when ChatGPT still used its old layout.
Edition one doesn't cover Perplexity, Google's AI Overviews or other assistants, and it records what the answers did, not why. The Canadian metros appear only in the September comparison and the weekly series, which pool both countries. The smallest US city in the panel has about 300,000 people, so nothing here speaks for small towns. Gemini's card and own-site measures stay blank because, in the answers our panel captured, its place cards faded during the window, its card answers carried no citations and its cited answers carried no cards.
The twelve measures, defined so later editions count the same way. Citation measures (6 to 8) count only answers with a captured citation: 6,304 for ChatGPT, 6,283 for AI Mode and 4,219 for Gemini, 2,891 of them from Oct. 19 to 31.
Phone numbers (1) is the share of all answers matching a North American phone number pattern; citation links (2) is the mean number of citation links per answer, across all answers; and length (3) is median words per answer. Cards (4) counts map or place cards per answer, in answers that showed cards. The urgent shortlist (5) compares cards per answer on six urgent questions with the rest of HVAC, plumbing and electrical, plus the businesses ChatGPT named per answer on the two narrowest.
Directories (6) is the share of cited answers citing a site on a fixed list of about 90 directory, review and best-in-city list domains (version one), and Yelp (7) and Reddit (8) are that same share for each of those sites. Booking (9) is the share of all answers linking a booking or scheduling page on a fixed list of field-service platforms and AI scheduling tools, or on a booking or scheduling subdomain (version one). Measures 10 to 12 work from the cards: no street address (10) is the share of map cards without one, another town (11) the share whose address doesn't contain the city in the question, counting suburbs and big-city neighborhoods with their own mailing names, and own website (12) the share of all answers citing the website of a business on their own map cards, a floor because only cards carrying a domain can match.
Urgent-question cards span Oct. 2 to 31, the one cell outside the window: ChatGPT showed 4.13 per answer on the six urgent questions against 4.48 on the rest of those trades, and AI Mode 3.88 against 4.18.
Comparisons inside the window use the same nine questions a week apart, or full weeks, because each day of the panel runs a different set of questions. In date order, the same-question shares rest on 537, 540 and 539 cited ChatGPT answers and 533, 539 and 539 AI Mode answers. The Maps-link share uses 4,087 US ChatGPT answers, and Gemini's two place-card shares 837 and 1,620 answers.
The September comparison sets the shares we published on Sept. 28 (9,434 ChatGPT and 9,435 AI Mode answers, Aug. 31 to Sept. 20) against 10,507 answers per engine from Oct. 9 to 31, all 100 metros. The weekly series counts links per 100 runs (2,699 to 3,288 ChatGPT runs a week), so one answer can count more than once, and its last week covers six days. The phone check compared 900 numbers from ChatGPT answers between Oct. 2 and 31 with each business's website on Nov. 1; the US share is 66 of the 374 we could check.
We re-ran every index cell and the weekly series on Nov. 1, 2026.
Sources
- More ways to connect with the web from Search, Google, October 15, 2026. Google's Oct. 15 changes to links in AI Mode and AI Overviews.
- What changed in ChatGPT in October: GPT-6, UTM tracking and sources, Sistrix, October 30, 2026. The GPT-6 rollout, the Oct. 8 to 9 layout switch and its general sample's links per answer.
- Reddit Citations Are Dropping in ChatGPT, Promptwatch, August 18, 2026. Reddit's mid-August drop in ChatGPT citations, called provisional.
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.