AI visibility · Study
AI assistants change their lists every month — but not whom they leave out
We asked ChatGPT, Gemini, Claude and Perplexity the same 1,728 questions about local businesses in six Swiss cities, a month after the first time. Only a third of the names in a list were the same; who gets named at all hardly changed — and the more Google reviews, the more often named still holds.
When somebody asks an AI assistant for a good bakery or an affordable dentist, which businesses does it name? We measure it every month with the same questions and the same panel of 276 small, independent businesses in six Swiss cities. November is the first time we can compare a month with the one before: the same 1,728 questions to ChatGPT, Gemini, Claude and Perplexity, asked on 1 November, a month after October's run.
-
143 of 162
panel businesses named in October were named again in November
-
1 in 3
of the names in a city and trade were the same in both months
-
1.8×
as often named: the half of a city and trade with more Google reviews, against the other half
How we measure
The panel: 276 independent businesses with a website of their own — up to eight bakeries, dentists, hair salons, plumbers, vets and garages in each of Zurich, Basel, Bern, Geneva, Lausanne and Lugano. 175 were drawn at random from OpenStreetMap; the rest were looked up or found by hand where a city had too few of a trade with a website. Chains are left out.
The questions: four per trade, in the city's language and without naming any business — "What is the best bakery in Lausanne?", "My tooth hurts really badly. Who should I see in Bern?", "Which plumbers in Lugano offer 24/7 emergency service?", "Where can I find an affordable dentist in Zurich if I don't have dental insurance?". Each was asked of each assistant three times, through its API with web search on. Every panel business had 48 chances to be named. The models were the same as in October.
To see what the named businesses have in common, we compare them with 428 more independents from OpenStreetMap that were not drawn into the panel, using their Google reviews, what their websites say and how they are built, and every page the answers cited.
The same four in ten were never named
114 of the 276 businesses (41%) did not appear in a single answer in November — the same number as in October, and 95 of them were never named in either month. 31 were named by all four assistants, against 33 in October. On average, one answer named 9.4% of the panel businesses in its city and trade (9.1% in October).
The same businesses, different lists
Who gets named at all hardly changed in a month. 143 of the 162 panel businesses named in October were named again in November (our comparison of the two months' data). All 33 that every assistant named in October were named again, and so were 70 of the 71 that three or four assistants named. Even of the 53 businesses that only one assistant named in October, 37 came back.
The lists themselves did change. In a city and trade, only about a third of the names were the same in both months — fewer than between two askings a day apart, on 30 September and 1 October. Perplexity's lists changed the most, Claude's no more over a month than from one day to the next. The overall figures barely moved: no assistant's coverage changed by more than a percentage point.
So a single answer, which is what a customer sees, is close to a lottery. Whether a business is in the draw at all is not, and it changes slowly.
Gemini still names the most, Perplexity the fewest
One answer from Gemini named 12.5% of the panel businesses in its city and trade, Claude 10.9%, ChatGPT 8.7% and Perplexity 5.6% — the same order as in October. It still follows the length of the lists: Claude names 7.5 businesses per answer, Gemini 7.0, ChatGPT 4.8 and Perplexity 2.8. Of everyone each assistant named, between 11% and 16% were panel businesses.
Vets and Lugano on top, hair salons and Zurich at the bottom
The ranking of trades and cities is the same as in October. Vets are named most (25.2% coverage), hair salons least (4.0%). Dentists slipped: 31 of the 48 were never named, five more than in October, and in Zurich and in Lausanne a single dentist was named in a single answer all month. Lugano is again the most visible city (13.2%), Zurich the least (5.7%).
Much of this is the size of the market: an answer names three to seven businesses whatever the city and trade, so a hair salon among Zurich's hundreds gets a far smaller share of each answer than a vet among a few dozen. Measured against what a business picked at random from its market would get, the panel is named 1.6 times (hair salons) to 6.0 times (vets) as often. Those ratios are higher than in October, but that is the measure, not the panel: we count the market as at least every business the answers have named there so far, and that count grows with every run. Compare the ratios within a month, not between months.
Google reviews still go with being named
We compare every business only with others of the same trade in the same city, so the size of the market drops out. For the second month, the strongest signal is the number of Google reviews: the half of a city and trade with more reviews was named 1.8 times as often as the other half (1.7 in October), and 2.6 times as often in six answers or more. It holds in every sample we looked at: the panel alone, only its randomly drawn part, without institutions and chains, and the businesses not drawn into it.
The rest of October's picture holds too:
- Being on the pages the assistants cite — local.ch and search.ch lists, guides, booking platforms — goes with being named 2.0 times as often, also with the number of reviews held constant.
- Online booking goes with being named 2.3 times as often, basic local SEO on the website 1.4 times: the city and trade in the page title, a meta description, structured data for the business and its opening hours, an FAQ.
- Prices on the website went with being named 1.4 times as often this month, but not once the number of reviews is held constant: the businesses that publish their prices are the ones with more reviews. Opening hours written as text and an llms.txt file made no difference.
Together, these measures explain about half of why the four assistants agree on whom to name. It is correlation in one month's answers, not proof of cause: businesses with many reviews tend to be busier in every way.
Where the names come from
When an assistant named a panel business, its own website was among the pages the assistant read 53% of the time (55% in October). Which page counts still depends on the question. For a problem — a toothache, a leak, a cake by Saturday — the business's own site was the only page with it in half of the cases. For "who is the best …?", in 45% of the cases only someone else's page had the business: a guide, a ranking, a directory. Questions about a problem again gave every assistant its lowest coverage (3.6% to 9.4%), and the touring club TCS was again the business outside the panel named most, in 50 answers in all six cities.
Perplexity now cites directories more
Most of the pages ChatGPT and Gemini cite are still businesses' own websites — 63% and 55%. Perplexity changed: businesses' websites fell from 54% of the pages it cites in October to 35% in November, while directories rose from 19% to 27% and booking platforms from 6% to 10%, with the same model answering. local.ch remains the most cited site by far, in a third of the answers that cite anything, and search.ch was cited in more than twice as many answers as in October. Claude still cites almost nothing; of the pages its searches returned, 38% were businesses' websites and 19% directories.
Who gets named instead
The most visible businesses of the panel are still institutions: the university animal hospital in Zurich, named in 85% of the answers about vets there, and the university dental centre in Basel (71%). Outside the panel, the names that come back from city to city are the same chains and brands: TCS, the hairdresser chain Gidor, Migros and Coop, Zahnarztzentrum.ch, the dealer groups Emil Frey and AMAG. Of everyone the assistants named, 13% were panel businesses, as in October, and at least 14% were chains — a quarter of the bakeries named.
What a local business can do
- Collect Google reviews. For the second month in a row, it is the strongest single signal of being named.
- Get onto the lists the assistants read. Local guides and media, local.ch and search.ch, and the booking platforms of your trade — Perplexity now leans on them more than on businesses' own sites.
- Do the basics of local SEO on your website. Your city and trade in the page title, a meta description, structured data for your business and its opening hours, an FAQ, and a page for each service people ask about. For a problem, the assistants cite the page that answers it.
- Judge over months, not by one answer. A list changes from one asking to the next; whether you are in the draw at all changes slowly. Our free ChatGPT visibility check shows in a minute whether ChatGPT names your site — check again over time.
The full figures — every city and trade, the comparison with chance, what the named businesses have in common, the most visible businesses, the names outside the panel and the method — are in the November report (PDF). The next run is on 1 December.
The data behind it — every panel business's visibility, the comparison with chance and every test of what goes with being named — is published as CSV files, listed and described in the data file.
Sources
- SMB AI visibility study, November 2026 (PDF) — Quaryan, 1 November 2026
- SMB AI visibility study, November 2026: the data (CSV) — Quaryan, 1 November 2026
- SMB AI visibility study, October 2026 — Quaryan, 10 October 2026
- OpenStreetMap — the panel's businesses, drawn on 30 September 2026