How does ChatGPT choose its sources?

They no longer get a list, but an answer. How does the AI decide which businesses appear in that answer? That’s exactly what GEO is about.

What is a generative engine?

Researchers at Princeton University and IIT Delhi call AI-powered search systems “generative engines” [1]. They search the web, combine information from several sources and write their own answer from it. For users, that’s convenient. For businesses, it has a downside: anyone who publishes content has little control over whether and how it appears in the answers [1].

A study presented at the SIGIR 2026 conference sums up the core problem: AI answer engines access many pages but cite only a few of them. Visibility therefore no longer depends only on ranking, but on being cited [2].

How often does AI already answer instead of Google?

A long-term measurement of more than 55,000 search queries in spring 2026 found an AI Overview above the search results for 13.7% of all queries. For searches phrased as a question, the figure was 64.7% [3]. Another study with 11,500 representative user queries arrived at 51.5% [4]. The figures differ because the studies looked at different queries.

The direction is the same: anyone asking a specific question increasingly gets an AI answer.

That answer often replaces the click. In an analysis of the browsing behavior of 900 US adults, the sources named in AI Overviews were clicked on in only about 1% of visits [5]. An analysis based on Semrush data puts the drop in click-through rate for top positions at 34.5% when an AI Overview appears [6].

Does the AI simply take the top Google results?

No. In the long-term measurement mentioned above, almost 30% of the domains cited by Google AI Overviews did not come from the first page of results [3]. The authors see this as a separate selection mechanism that differs from classic ranking.

Comparing Google Search, AI Overviews and Gemini, researchers found hardly any overlap in the sources used [4]. The AI systems also behave differently from one another: Perplexity and Google cite more sources on average, ChatGPT fewer. In return, those few sources shape ChatGPT’s answer much more strongly [7]. Being included by one system doesn’t automatically mean being included by the next.

A study of 11,000 real search queries speaks of “answer bubbles”: the same question leads to different answers depending on the system, because each system prefers certain sources [8].

A disadvantage for smaller businesses

A comparison of AI search and classic Google search shows a clear tendency: AI systems rely disproportionately on third-party sources such as trade media and news coverage, and much less on a company’s own website. The authors explicitly speak of an advantage for big brands over smaller niche providers [9].

For a trades business or a small service company, this means: a good website alone doesn’t guarantee that the AI will mention you. A study on the changing nature of online visibility puts it this way: some sources are included in answers, others become practically invisible, regardless of how well they ranked before [10].

So are Google rankings unimportant now?

No. A good ranking remains an important foundation. Research also warns against declaring SEO dead: the data is more nuanced than the loud voices on both sides suggest [6]. Classic SEO isn’t outdated, but it has to be thought through together with visibility in AI answers [11].

But ranking is losing its exclusive role. It now answers only one of two questions. The first is: will I be found? The second is becoming more and more important: will I be named?

Is there a checklist for getting mentioned in AI answers?

There’s no universal one. Research into source selection is still in its early days, and the systems keep changing. The Princeton study already showed that effects differ by industry [1]. AI Overviews also don’t always answer the same query the same way on two runs [4].

That’s why it pays to look at the individual case: how do ChatGPT, Perplexity, Gemini or Claude describe your business today? Is it mentioned at all – and if not, who is mentioned instead? This is exactly where my AI visibility for businesses comes in.

Conclusion

Search is changing from a list of results to a finished answer. AI systems read many sources but name only a few, and they choose according to their own patterns, which don’t match the Google ranking. Besides ranking, there’s therefore a second question for businesses: am I one of the sources the AI relies on? If not, it recommends someone else to your customers.

Sources

1. Aggarwal et al. (2023/KDD 2024): GEO: Generative Engine Optimization
2. Vishwakarma et al. (2026, SIGIR): What Gets Cited: Competitive GEO in AI Answer Engines
3. Xu et al. (2026): Measuring Google AI Overviews: Activation, Source Quality, Claim Fidelity, and Publisher Impact
4. Grossman et al. (2026, SIGIR): How Generative AI Disrupts Search: An Empirical Study of Google Search, Gemini, and AI Overviews
5. Chapekis et al. (2026): Investigating Click Behaviors On Google Search Result Pages That Produce an AI Overview
6. Pagadala (2026, SSRN): The Disruption of Search Engine Optimization by Large Language Models
7. Zhang et al. (2026): From Citation Selection to Citation Absorption
8. Huang et al. (2026): Answer Bubbles: Information Exposure in AI-Mediated Search
9. Chen et al. (2025): Generative Engine Optimization: How to Dominate AI Search
10. Blanco-Guzmán et al. (2026, MIPRO): From Search Rankings to Answer Rankings
11. Palonna et al. (2025): SEO-Strategies Transforming under the AI Influence

FAQ