- On October 11, 2026, we asked ChatGPT, Perplexity, and Gemini to recommend a tax advisor in Munich. The three systems named 15 different firms, and only one of them twice.
- None of the firms ranking with their own website on Google's first results page showed up in any of the three answers.
- All three systems drew on the same two directories: a review portal and the city portal muenchen.de.
- ChatGPT and Perplexity find web pages through their own crawlers, which you can allow or block in robots.txt. Gemini and Google's AI features rely on Google Search.
People looking for a provider now also ask ChatGPT, Perplexity, or Gemini. For companies, one question matters: do you get mentioned? So we ran a simple test and asked all three systems the same question.
The results show how differently the systems answer and why a strong Google ranking isn't enough on its own. At the end, we explain how the systems find their sources and what you can take away from this.
Three systems named 15 different firms
On October 11, 2026, we asked ChatGPT, Perplexity, and Gemini: “Which tax advisor in Munich can you recommend?” All three searched the web to answer.
| System | Firms named | Also named by another system |
|---|---|---|
| ChatGPT | 5 | 1 |
| Perplexity | 4 | 0 |
| Gemini | 7 | 1 |
For Gemini, we only count firms; it also listed payroll tax assistance associations.
In total, there were 16 recommendations covering 15 different firms. Only one firm was named by two systems, and none by all three. We are not publishing the names.
Our own query on October 11, 2026, via DataForSEO: ChatGPT with web search via the web interface, Perplexity (Sonar model), and Gemini (3.8 Flash with Google Search) via their APIs. One question at one point in time; answers change from query to query.
None of Google's first-page results made it into an answer
On the same day, we searched Google for “steuerberater münchen” (tax advisor Munich). At the top was the map pack with three firms, followed by seven firm websites in the organic results.
- None of the seven firms with their own website on page one was named by ChatGPT, Perplexity, or Gemini.
- Two of the three law firms from the map view were each named by one of the AI systems.
For ChargeOne, we described the offering and rental model so that AI assistants can pick them up when answering questions about charging infrastructure. From January to mid-October 2025, AI assistants sent 202 visits to the website (see the case study).
All three systems drew on the same two directories
The answers differed, but the sources partly overlapped. All three systems relied on the review portal wer-kennt-den-besten.de and the city portal muenchen.de, plus individual firm websites.
Beyond that, each system had its own focus:
- ChatGPT also drew on the Munich Chamber of Tax Advisors and business directories such as Das Örtliche.
- Perplexity used Gelbe Seiten and Das Örtliche.
- Gemini specifically looked for firm rankings from Focus and Handelsblatt and grouped the firms by client type.
What this means for you: your own website is one source among many. Directory listings, reviews, and mentions in trade media all feed into the same answer.
Each system finds its sources differently
AI systems either answer from what they learned in training, or they search the web first and summarize the pages they find. Only with web search do they name current providers and cite their sources.
| System | How it finds web pages | What you can control |
|---|---|---|
| ChatGPT | Own search crawler: OAI-SearchBot | According to OpenAI, pages that block OAI-SearchBot in robots.txt don't appear in search answers. The training crawler (GPTBot) can be blocked separately. |
| Perplexity | Own crawler: PerplexityBot | According to Perplexity, PerplexityBot surfaces websites in its results and is not used to train AI models. |
| Gemini and Google Search | Google Search | According to Google, AI features in Search are built on its standard ranking and quality systems. No special files or markup are required. |
OpenAI and Perplexity handle fetches triggered directly by a user separately. Perplexity states that its fetch agent, Perplexity-User, generally ignores robots.txt.
We get your company into these answers
Our test shows why, for AI search, we look beyond your own website. Here's how we work for our clients:
- Identify questions: We find out which questions your customers ask AI systems and which providers show up in the answers today.
- Check sources: We identify the pages the answers are based on and whether your website is one of them.
- Build mentions: We strengthen your presence on other sites. Expert articles, mentions, and reviews signal to AI systems that your company is a reliable source.
- Measure mentions: We regularly ask the relevant questions in ChatGPT, Perplexity, and Google and track how often, and how, your company is mentioned.
The ChargeOne case study shows what this looks like for a provider of charging solutions: AI assistants reproduce its unique selling points with 100% accuracy, and the recommendation rate is 62%. Our GEO check is a good place to start.
