These days, when people are looking for an optician or a hotel, they no longer just turn to Google—they’re increasingly turning to ChatGPT, Gemini, or Perplexity. A study by Frascati GmbH in Allschwil on visibility in AI models now shows just how unevenly opportunities are distributed between chain stores and independent businesses. The study is based on 1,116 responses regarding 50 Swiss hotels as well as 50 Swiss and 50 German opticians, respectively.

Chain stores dominate; independent local businesses are less common

If you ask Perplexity, Gemini, or ChatGPT to search the web for an optician in a city in German-speaking Switzerland, at least one chain store is mentioned in 87.7 percent of the responses—the figure is exactly the same in Germany. Independent businesses do appear as well: In 83 percent of the 300 Swiss responses, at least one of the five independent opticians examined in the respective city is mentioned; in Germany, the figure is 63.7 percent. On average in Switzerland, however, a single business appears in only 42.9 percent of the responses for its city, compared to 19.7 percent in Germany.

Significant differences are also evident among the models. Perplexity mentions at least one of the opticians tested in 96 percent of its responses, Gemini in 94 percent, while ChatGPT does so in only 59 percent. In Germany, the figures are 76, 86, and 29 percent.

Hotels: One in ten responses refers only to booking sites

When asked where to book a specific hotel online, the models cited the hotel’s own website in 89 percent of 300 responses. However, in 10.3 percent of cases, they referred exclusively to booking portals; with ChatGPT, this figure rose to 20 percent. Booking.com was mentioned in 276 of the 300 responses.

It becomes more difficult for hotels when guests search without specifying a name, for example, with a query like «good hotel in Thun for a weekend.» In 62.2 percent of the 180 responses, one of the five hotels reviewed in the respective location was mentioned. The differences between the cities are striking: In Thun and Winterthur, the reviewed hotels appear in all 18 responses; in Basel, in seven; and in Zurich, in only two. Twelve of the 50 hotels examined do not appear in a single response.

Where the models get their information from

According to Frascati, the models draw their data from directories, portals, and the businesses’ own websites. local.ch is cited as a source in 160 of 318 responses regarding Swiss opticians. Businesses could therefore improve their visibility in AI models by keeping their listings complete and consistent, linking to their own websites, creating a separate page for each location, answering frequently asked questions in plain language, and regularly asking those same questions themselves. However, this does not guarantee that they will be listed.

«An AI’s response isn’t a list of results, but a paragraph with about five names,» says Fabiano Frascati, managing director of Frascati GmbH. «You can determine whether your own business is included by using the same questions that customers ask. No one can promise that, but the foundation lies with the business itself: complete listings, one page per location, and answers in plain language.»

Measurement Method

The data was collected on September 11, 2026, for Switzerland and on September 12, 2026, for Germany, using Perplexity Sonar, ChatGPT (GPT-5.4), and Gemini 2.5 Flash, each using web searches via the OpenRouter API rather than through the apps. The study examined 50 independent businesses with their own websites in each location—five per location—excluding chains: opticians in ten German-speaking Swiss cities and ten German cities, as well as hotels in ten Swiss locations from Basel to Zermatt. For the opticians, five different phrasings of the same question were used per city, along with three questions without a location specified—each asked twice—resulting in 318 responses per country. For the hotels, the booking question was asked for each property (300 responses), and three anonymous guest questions were asked for each location (180 responses). The confidence intervals were calculated using the Wilson method at the 95 percent level. This is not a random sample; the study measures visibility, not revenue.

Further studies and raw data can be found at https://frascati.ch/blog/50-schweizer-hotels-im-ki-test-2026 and https://frascati.ch/blog/50-schweizer-optiker-im-ki-test-2026. Methodology, country comparisons, and licensing are available at https://frascati.ch/messungen documented; the raw data is available under a CC BY 4.0 license at https://frascati.ch/studien/index.json available.

For more information on the measurements conducted by Frascati GmbH, visit https://frascati.ch/messungen retrievable.