Without web search, AI names none of 20 businesses; with web search, six

The result in brief

Without web search, meaning from the models’ trained knowledge, none of the three AI engines named any of the 20 businesses: 0 of 960 queries. With web search they named 6 of 20 businesses at least once, in 28 of 240 queries (11.7 percent). On purchase-intent questions like “Best dentist in Munich”, the businesses we tested therefore mostly went unnamed.

14 of 20 businesses went unnamed in all 12 queries, even with web search. The most frequently named was a real estate agent in Stuttgart, in 9 of 12 queries. We recorded which pages the web search returned for Claude only, as the list of search results, not as sources the answer relied on. Directories, rankings and review portals appeared most often in these search results: kennstdueinen.de in 35 of 80 queries, dasoertliche.de in 34, golocal.de and threebestrated.de in 33 each.

For comparison: the first measurement in July 2026 had found one mention in 480 queries without web search (0.2 percent, 19 of 20 businesses unnamed). It used namedly’s looser name matching, which already counts a matching part of a name as a hit. Under that rule the October measurement without web search would also have produced two hits, under the strict rule none. The July individual results were not saved and cannot be verified. That is why we re-measured in October and logged every query.

How we tested

The sample: 20 real, independent businesses, no chains, no directories. Spread across 10 industries (dentist, hair salon, tax advisor, restaurant, carpenter, real estate agent, physiotherapy, veterinarian, driving school, bakery/cake shop) and 10 cities (Munich, Hamburg, Nuremberg, Cologne, Berlin, Stuttgart, Vienna, Frankfurt, Duesseldorf, Dresden).

We measured with a script built on our tool namedly. Without web search: three AI engines (ChatGPT, Claude, Gemini, each through the provider’s API) times eight purchase-intent prompts in two runs, which is 48 queries per business and 960 in total. With web search: the same engines with the first four prompts in one run, 12 queries per business and 240 in total. A fixed instruction before each question asked for only a list of 3 to 5 recommended providers, without explanation. Almost all answers (1,154 of 1,200) named 4 or 5 providers, so a business had to make it onto a short list. A hit only counts if the distinctive part of the business name appears in full in the answer. A similar-sounding name does not count.

Models as reported by the providers: GPT-4o mini (gpt-4o-mini-2024-07-18), Claude Sonnet 5.5 (claude-sonnet-5-5) and Gemini 2.5 Flash (gemini-2.5-flash). We measured on 5 October 2026. Without web search, the models answer from their trained knowledge alone. With web search, that instruction also asked the model to research the web before answering, and the model demonstrably searched in all 240 queries. Business names stay anonymous for fairness; results are broken down only by industry and city. As of October 2026.

The anonymised individual results of all 1,200 queries are available as a CSV file, one row per query. The prompts were asked in German; numbers 0 to 7 translate as “Best [industry] in [city]”, “[industry] [city] recommendation”, “Which [industry] in [city] can you recommend?”, “Good [industry] near [city]”, “[industry] [city] with good reviews”, “Who is the best [industry] in [city] and the surrounding area?”, “Looking for a [industry] in [city]. Which provider do you recommend?” and “Top [industry] [city]”. The full log with the raw answers contains business names and therefore stays internal. In the CSV file, bakery and cake shop are listed as separate industries; the table below combines them. The file is UTF-8 encoded and comma-separated; in Excel, open it via “Data” and “From Text/CSV”.

The numbers by industry

The columns show in how many queries one of an industry’s businesses was named. With two businesses per industry, no industry pattern can be derived from this. The engines differ with web search: Claude and Gemini each named one of the businesses in 11 of 80 queries, ChatGPT in 6 of 80.

What web search changes, and what it doesn’t

Without web search the models answer from their trained knowledge, and none of the 20 businesses came up there as a recommendation. With web search the engines draw on current sources, and some of the businesses show up. What can be found online about a business therefore plays a part.

Still, 14 of 20 businesses went unnamed even with web search. We did not measure how well these businesses’ websites rank on Google. We only measured whether the AI names the business.

What actually drives AI visibility

AI visibility has three layers. The first is the technical base: clean schema, crawlable for AI bots, clear answers in the first sentences. Necessary, but not sufficient on its own.

The second is citable content: original numbers, real answers to real questions, dense paragraphs that fully answer a question. The third, and this is the real lever, is mentions through independent sources: industry directories, trade media, reviews, consistent brand signals. A ranking alone isn’t enough.

What a business can do now

• Check whether your own site is even reachable for AI bots like OAI-SearchBot, PerplexityBot and ClaudeBot. Many businesses block them by accident.
• Sharpen the structure: Schema.org data, FAQ blocks with real questions, answers in the first two sentences.
• Build independent presence: a complete Google Business Profile, consistent listings in industry directories, reviews, trade articles.

The technical base is done within days. The independent presence takes weeks to work, because AI models first need to pick up those sources. Anyone starting now is early: hardly any local business has a documented GEO strategy.

Frequently asked questions

Cite this study

Writing about AI visibility or GEO and need solid numbers? The graphic and citation text are free to use, with a link back to the source.

Suggested citation

bytebrise (October 2026): AI Visibility of Local Businesses in the German-Speaking Region. Queried via API with GPT-4o mini, Claude Sonnet 5.5 and Gemini 2.5 Flash: without web search none of the three models named any of 20 tested businesses (0 of 960 queries), with web search 6 of 20 were named (28 of 240 queries, 11.7%). https://bytebrise.com/en/ai-visibility-study.html

Embed code for your page

<img src="https://www.bytebrise.com/studie-ki-sichtbarkeit-infografik-en.svg" alt="bytebrise study: without web search AI names 0 of 20 businesses, with web search 6 of 20" width="480" height="480" />
<p>Source: <a href="https://www.bytebrise.com/en/ai-visibility-study.html">bytebrise AI Visibility Study</a></p>

Copy code

For logos, screenshots and further press material, see our media kit.