Researchers analysed how generative AI systems affect local business visibility by testing eight industries across five Hungarian cities. They ran a total of 8,880 queries and AI runs, producing 161,842 normalized observations. The team compared traditional Google organic and Google Maps results with answers from ChatGPT, Gemini and Google AI Mode.
Key findings
- AI systems frequently recommend specific businesses rather than offering only general guidance: in 95.6% of the 8,400 AI responses examined the researchers could identify at least one specific local business. By system this was 99.9% for ChatGPT, 93.0% for Gemini and 91.0% for Google AI Mode.
- AI recommendations tend to pre-filter options: because AI can take explicit user context into account (price sensitivity, location, special requirements), it often narrows choices rather than merely listing links. As a result, part of the user’s decision may already be made before they click through to a company website.
Conversion impact — a Hungarian case study
In a Hungarian case study within the research, nearly 8% of visits originating from AI recommendations resulted in a purchase, compared with roughly 1.4% from other examined sources. On average about 12.5 AI-originating visits yielded one purchase, versus approximately 71 visits from other sources.
Attention concentration and inequality
- Visibility concentrates more under AI recommendations: the top 10% of domains accounted for 22.0% of appearances in Google organic results, but they represented 42.6% of appearances in ChatGPT.
- Measured by Gini index, the market-average concentration was 0.364 for Google Organic and 0.594 for ChatGPT — roughly a 63% higher concentration for ChatGPT.
Google ranking is not an automatic AI advantage
The data show AI does not simply reproduce Google’s top results. Nearly 70% of the sources ChatGPT used did not come from Google’s top ten organic results. Specifically, ChatGPT sources were 30.2% in Google Top 10, 35.1% in positions 11–100, and 34.6% outside the Top 100. Gemini and Google AI Mode relied more on well-ranked Google sources, but none of the systems were mere copies of Google’s rankings.
Ranking instability
- For the same query, ChatGPT and Google AI Mode produced different first recommendations in 91.1% of cases; ChatGPT and Gemini differed in 86.2% of cases.
- Even within the same model, rephrasing the user need changed the top recommendation: in ChatGPT different phrasings produced a different top company in 27.3% of cases, and in Gemini in 41% of cases. There is therefore no single stable “AI ranking”; different models and different prompt formulations often yield different winners.
Why the danger is hardest to spot when things still work
The authors warn that exclusion is unlikely to appear as a sudden drop in web traffic. A business can keep good Google rankings, receive visitors and grow revenue while simultaneously becoming absent from the AI-driven decision space. The sample contained local companies that appeared in Google’s Top 10 but were not named in any of the AI systems’ responses.
Implications for the future
Online presence may split into two distinct types: the traditional web/SEO presence that makes a company findable on Google, and the narrower AI-driven presence that determines whether a business appears in the AI-generated shortlist shown to a specific buyer. The digital fundamentals (well-structured website, relevant content, external mentions) remain valuable, but they no longer guarantee AI visibility. The shift creates opportunities for businesses that adapt early and risks for those that notice the change too late.
Next steps and further information
Detailed results of the study are published in emlitve.hu’s AI visibility research. The article’s authors, Orosz Kolos (emlitve.hu, co-founder, owner) and Széles Tamás (emlitve.hu, co-founder, owner), will run a 1.5-hour interactive workshop on the topic at the AI & Digital Transformation conference on November 26. Registration and conference details are available through the conference materials.
This article appears in the on the other hand opinion column of Portfolio; the authors’ views do not necessarily reflect the editorial stance of Portfolio.
