In Cheers' March-to-October 2026 API study of 455 businesses, ChatGPT, Gemini and Perplexity described a company in 96% of answers when asked about it by name, but recommended it in 39.7% of answers to buyer questions. The commercial job is to make that second list: the one a homeowner gets when asking who to hire.
Last verified: November 1, 2026.
Typing your company name into ChatGPT is the quick check, but that question hands the engine the business to talk about. Someone whose AC quit tonight hasn't picked anyone yet, and their question is the one that builds a shortlist; the test at the end shows how to ask it about your own town this week.
We asked both kinds of question about the same 455 businesses in 361 cities. Being on that list counts for more this fall, since from Oct. 9 to 31 most of ChatGPT's US local answers in our panel carried at least one phone number, and Google's price check, Gemini's Call for Me test and Meta's Muse beta can each place calls to businesses for people.
Two questions, same business
A by-name question tests whether an engine can find you and say something about you. A buyer question, the kind a homeowner types without naming anyone, tests whether the engine picks you out of everyone it could have named. Our AI visibility benchmarks leave out questions that name the company for this reason, and other research has counted how rarely a location shows up in ChatGPT's recommendations at all. This study measures the gap between the two questions on the same businesses.
Use the by-name answer to check how the engine describes you. Use the buyer answer to measure whether it recommends you. Our free grader makes that distinction so you can see which job each location needs next; the sample and question definitions are in the Method section.
Known by name, left off the list
Cheers AI visibility grader
Known by name, left off the shortlist
- ChatGPT, Gemini and Perplexity: 1,311 of 1,364 by-name answers; 2,166 of 5,456 buyer answers
Cheers AI visibility grader, latest complete run per business, Mar 17 to Oct 30, 2026, Pacific time (developer APIs with web search). Queried Nov 1, 2026.
| Engines | Asked by name | Asked who to hire |
|---|---|---|
| ChatGPT, Gemini and Perplexity | 96.1% | 39.7% |
ChatGPT on its own described the business in 94.9% of by-name answers and named it in 39.3% of buyer answers. Among the 325 home-service companies the gap was wider, 97.5% by name against 35.8% when the question was who to hire, and roofers pull that group down (more on them below).
In late September we changed the wording of the buyer questions and, at the same time, started giving ChatGPT and Perplexity the asker's approximate city. Businesses graded since Sept. 24 were named in 49.4% of their buyer answers, against 38.0% for those graded before, while by-name answers held steady. Gemini, which gets no location hint, moved the least of the three. The companies, dates, prompts and location inputs changed together. This comparison cannot isolate which change explains the difference.
These are two different sets of businesses, not the same companies getting better. The engine that named businesses most often wasn't the same one before and after the change, so don't read any of these rates as a ranking.
107 businesses never made a single list
Cheers AI visibility grader
How many of 12 buyer answers named each business
Businesses
Never named
Median
Named every time
Cheers AI visibility grader (ChatGPT, Gemini and Perplexity), latest complete run per business, Mar 17 to Oct 30, 2026, Pacific time. Queried Nov 1, 2026.
| Times named | Businesses |
|---|---|
| 0 of 12 | 107 (Never named) |
| 1 of 12 | 39 |
| 2 of 12 | 35 |
| 3 of 12 | 29 |
| 4 of 12 | 26 (Median) |
| 5 of 12 | 34 |
| 6 of 12 | 21 |
| 7 of 12 | 26 |
| 8 of 12 | 25 |
| 9 of 12 | 30 |
| 10 of 12 | 23 |
| 11 of 12 | 27 |
| 12 of 12 | 32 (Named every time) |
Of the 455, 107 (23.5%) appeared in none of their 12 buyer answers. The engines weren't strangers to them: at least one engine described all but one of them when asked by name, and all three described 90. Nearly all were graded under the questions we used through Sept. 23. Under the questions we've used since Sept. 24, fewer than 10 of the 68 businesses graded were never named.
As for roofers, they're 94 of the 455 but 39 of the never-named, and the engines named them in 18.3% of buyer answers, against 45.3% for every other business. Roofers were also nearly a quarter of the businesses graded before the late-September change and a much smaller share since. So trade mix explains part of the rise.
Past zero, the bars run close to flat: 39 businesses were named once, then 35, 29, 26, 34, 21, 26, 25, 30, 23, 27 and 32 for two through twelve. The median business turned up in 4 of 12 answers, and 181 of the 454 with all 12 answers (39.9%) turned up in two or fewer.
A by-name screenshot wouldn't catch any of that. If a branch manager or an agency sends you one as proof a market is covered, ask for the buyer answers that go with it, and run the test below to see which bar you're in.
Why the engines skip a business they know
Cheers AI visibility grader
Every shortlist came with a crowd
40
other businesses named, on average
Base: 455 businesses, 12 buyer answers each; median 39
47.5
other names, on average, in answers for businesses never named
37.7 for businesses named at least once
81.1%
of competitor names came from one engine only
18,182 business and competitor pairs
4.3%
came from all three engines
Same 18,182 pairs
Cheers AI visibility grader (ChatGPT, Gemini and Perplexity), latest complete run per business, Mar 17 to Oct 30, 2026, Pacific time. Queried Nov 1, 2026.
| Finding | Value | Base |
|---|---|---|
| other businesses named, on average | 40 | 455 businesses, 12 buyer answers each; median 39 |
| other names, on average, in answers for businesses never named | 47.5 | 37.7 for businesses named at least once |
| of competitor names came from one engine only | 81.1% | 18,182 business and competitor pairs |
| came from all three engines | 4.3% | Same 18,182 pairs |
A buyer answer listed about six businesses on average, the graded one included when it made the list, and one business's 12 buyer answers named about 40 other businesses between them. Describing a company only takes finding material about it; recommending one means picking it out of that crowd. Businesses that were never named faced bigger crowds, with 47.5 other names on average against 37.7 for those named at least once.
The engines rarely agreed on those rivals: 81.1% of competitor names came from one engine only, and 4.3% from all three. When you write down who's named instead of you, the names that turn up on two or three engines are your real competition for that question, and no single engine stands in for the market.
When an answer named the graded business, it cited the business's own website 63.0% of the time, and across all buyer answers the site showed up in 26.5%. Your website is part of the evidence a customer can inspect once you make the shortlist. Make sure the cited page gives a clear reason to choose your branch.
This week, open the sources cited in answers that name a competitor who keeps showing up and compare those pages with yours; our guide to why AI recommends a competitor walks through that source trail, location by location. The lists also change from week to week, which is why one answer can mislead and why the test uses 12.
AI rarely warns people off a contractor
Cheers Market Baseline, October 2026
AI almost never lists a business to warn you off
29
negative tone labels out of 70,624 US business mentions
Base: 60 US metros, answers scheduled Oct 1 to 31, 2026
91.0%
of US mentions labeled positive
Same 70,624 mentions
47 of 90,750
labeled negative across all 100 US and Canadian metros
90.2% positive
Cheers Market Baseline (Cheers' own research panel): ChatGPT, Gemini and Google AI Mode answers to 32 home-service questions asked weekly in 100 metros (60 US, 40 Canadian); answers scheduled Oct 1 to 31, 2026. Queried Nov 1, 2026.
| Finding | Value | Base |
|---|---|---|
| negative tone labels out of 70,624 US business mentions | 29 | 60 US metros, answers scheduled Oct 1 to 31, 2026 |
| of US mentions labeled positive | 91.0% | Same 70,624 mentions |
| labeled negative across all 100 US and Canadian metros | 47 of 90,750 | 90.2% positive |
In the Cheers Market Baseline, a separate panel, 29 of 70,624 business mentions in US answers from ChatGPT, Gemini and Google AI Mode carried a negative automated tone label in October 2026, and 91.0% carried a positive one. Across all 100 US and Canadian metros, it was 47 of 90,750.
Praise is the default because these questions ask which local companies to hire. Engines don't put a business on a list to warn people off it; they leave it off. Treat the negative count as approximate, though, since ChatGPT's Oct. 9 layout change can throw off automated labels.
Complaints do come up when a question names the business and asks what customers say. We didn't measure how often, and most of our by-name questions asked about complaints outright. Track inclusion and sentiment separately. Inclusion tells you whether the branch makes the shortlist; sentiment tells you how an answer describes it. Together they give you a more useful view than either one alone.
Why the shortlist matters more this fall
October changed what that list looks like. Sistrix found ChatGPT switched to a building-block layout overnight on Oct. 8 to 9, 2026, and used it in 97% to 100% of sampled answers from Oct. 23 (Sistrix, Oct. 30, 2026). Sistrix didn't count phone numbers, but our panel did: 12.6% of US ChatGPT local answers in the Cheers Market Baseline carried at least one from Oct. 2 to 8, and 92.5% from Oct. 9 to 31.
Google and Meta also have AI that can make the call itself. Google's price-check help page, read Nov. 1, says "Google will call multiple service providers near you," and if it still can't reach them, "we will send you an email with recommendations of service providers." Its examples are auto repair, salons, groomers, wellness centers and driving schools, with no mention of home repair, though Google's May 19, 2026 I/O post listed home repair.
TechCrunch reported Google's test of Gemini Call for Me on Sept. 24, 2026. It's for US Pixel 11 owners with a Google AI subscription. Google says each call opens by disclosing that it's an AI assistant from Google, and in Google's examples the user names the business. Meta's Muse can place calls to US businesses in a beta that expanded on Sept. 16, per an X post by Ryan Fox that TechCrunch reported the next day.
Customers are already asking. BrightLocal's 2026 survey of 1,002 US adults (Feb. 11, 2026) found 45% had used AI for local business recommendations in the past 12 months, up from 6% in 2025. BrightLocal also found (March 10, 2026) that 97% of those users double-check its picks against reviews at least sometimes.
We expect assistants to handle more of the journey from shortlist to first contact. Earning the recommendation puts your business in position for that next step. Start by measuring the buyer questions that matter in each market, then improve the evidence those answers use.
Run the shortlist test this week
Ask four plain buyer questions without your company name, swapping in your own trade and town: "who's the best water heater company in [town]?", "my AC quit tonight, who should I call in [town]?", "which plumbers in [town] are fair on price?" and "who's best for [specific service] in [town]?" Put each one to ChatGPT, Gemini and Perplexity, count how many of the 12 answers name you, and write down every business named instead, engine by engine.
Report it as one number per market in the monthly review (named in X of 12 buyer answers), and keep each engine's answers visible, since competitor names mostly show up on one engine only. Keep by-name checks for accuracy and tone, and use buyer-question results to judge recommendation coverage.
Lists move week to week, so one run is a snapshot; re-run it monthly and compare the counts. For a brand with dozens of branches, run the test per town and roll it up with a location-by-location audit.
Cheers measures those gaps market by market and turns them into work on the reviews, listings and website pages that support each location.
Trade differences also matter: the roofing-versus-HVAC study compares recommendation rates within its historical measurement window.
Method
Two panels feed this article, so don't compare or subtract their numbers; we pulled grader and tone figures on Nov. 1, 2026.
The Cheers AI visibility grader
The grader asks ChatGPT, Gemini and Perplexity through their developer APIs with web search on, not the consumer apps, which add map cards, phone numbers and business listings. Per business, it asks four buyer questions built from its category and city ("Who are the best [trade] companies in [city]?") and one that names the business and asks what customers say (until early October, what they praise or complain about), each as a fresh request with no sign-in, history or personalization.
Since Sept. 23, ChatGPT and Perplexity also get the business's city as the asker's approximate location, and in late September we reworded the buyer questions. From March 18 to June 8, ChatGPT answered all four buyer questions in one request for about a third of businesses, a method since retired.
A by-name answer counted as describing the business if it named it without saying it couldn't find information.
The sample is the 455 businesses that ran the free grader from March 17 to October 30, 2026, using each one's latest complete run, minus our internal runs. About 7 in 10 are home-service companies, and the median had 285 Google reviews among the 439 with a count.
In all, 1,311 of 1,364 by-name answers described the business (96.1%) and 2,166 of 5,456 buyer answers named it (39.7%). From the 387 businesses graded before Sept. 24 to the 68 since, Gemini went from 44.6% to 47.1%, ChatGPT from 37.4% to 50.0% and Perplexity from 32.0% to 51.1%.
Competitor names match exactly after lowercasing and replacing punctuation with spaces, so across the 18,182 business and competitor pairs two spellings of one company count as two, and true agreement runs somewhat higher.
The Cheers Market Baseline
Cheers' own research panel asks 32 home-service questions weekly in 100 metros (60 US, 40 Canadian), recording answers as ChatGPT, Gemini and Google AI Mode show them, and automated labels mark each business mention positive, neutral, mixed, negative or unknown. We used answers scheduled October 1 to 31, 2026; Perplexity is out: too few of its answers completed. ChatGPT's Oct. 9 layout change can throw off the labels, but no engine had more than 0.2% of its US mentions labeled negative in Oct. 1 to 8, 9 to 16 or 17 to 31.
Phone-number figures cover 1,910 US ChatGPT answers from Oct. 2 to 8 and 6,307 from Oct. 9 to 31.
Sources
- What changed in ChatGPT in October: GPT-6, UTM tracking and sources, Sistrix, October 30, 2026. ChatGPT's Oct. 9 building-block layout and its share from Oct. 23.
- How to check pricing for local businesses with AI in Google Search, Google Search Help, read November 1, 2026. Google calling providers, the email fallback and the example categories.
- A new era for AI Search, Google blog, May 19, 2026. Home repair among Google's calling categories at I/O.
- Google tests letting Gemini call businesses for you, TechCrunch, September 24, 2026. The Call for Me test for US Pixel 11 owners.
- Ask Gemini to handle your everyday phone calls, Gemini Apps Help, read November 1, 2026. The user names the business, and the call discloses it's Google's AI.
- Rival AI agents, Instinct and Meta's Muse, both add the ability to make calls, TechCrunch, September 17, 2026. Muse's beta calls to US businesses.
- Local Consumer Review Survey 2026, BrightLocal, February 11, 2026. 45% used AI for local recommendations, up from 6%.
- Nearly Half of Consumers are Asking AI for Business Recommendations, BrightLocal, March 10, 2026. 97% double-check AI picks against reviews at least sometimes.
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.