Google loves you, but ChatGPT doesn’t: why?
Generative models such as ChatGPT, Gemini, and Perplexity are increasingly becoming the entry point for information. Companies assume that a good ranking on Google will automatically translate into frequent appearances in LLM responses. A new analysis by Search Atlas clearly shows: this connection is much weaker than we think.
Google ≠ LLM
The study compared which domains and URLs Google displays in top positions versus those extracted by the three largest models. The results are surprising:
- Perplexity covers approximately 25–30% of the same domains as Google.
- ChatGPT matches Google in only 10–15% of domains, and even less for URLs.
- Gemini is the most selective – it shares only about 4% of domains.
Even if a website is number one on Google, it does not mean an LLM will mention it at all, let alone cite it.
Why such a difference?
LLMs do not “rank” content like Google. They do not operate based on classic SEO signals, but rather:
- select sources that appear in their training data or trusted lists,
- prioritize simply structured content,
- often ignore commercial or sales pages,
- sometimes do not cite sources at all (ChatGPT) or filter them heavily (Gemini).
SEO optimization alone is therefore not enough for visibility in generative responses.
What does this mean for marketing?
Online presence is shifting from search engines to conversational interfaces. If you want models to recognize you as a relevant source, you need a different logic for content creation:
1. More emphasis on authority, less on classic SEO tricks
LLMs prefer to cite authoritative, expert, and well-structured content rather than aggressively optimized commercial pages.
2. Topics that answer questions
Because models attempt to simulate an “answer,” articles written as clear explanations or guides – without unnecessary marketing jargon – have an advantage.
3. Transparent sources and references
Clearly citing sources and research increases the likelihood that content will be recognized as trustworthy and thus included in LLM responses.
4. Structure that models can easily digest
Short paragraphs, clear statements, concise subheadings. “LLM-friendliness” is more important here than mere readability.
5. Strengthening presence across multiple channels
Blogs, interviews, reports, expert studies, collaborations — all of this helps content find its way into training sets and external references more frequently.
A paradigm shift
SEO hasn’t disappeared, but it has become only half of the equation. With the rise of generative AI tools, a parallel ecosystem is emerging — one where algorithms do not rank the most relevant results for the user, but instead generate a single answer based on their own assessment of trust.
The gap between Google positions and LLM citations will likely continue to grow. The winners will be those who understand the dynamics of both worlds in time.
Sources
Sources
- Search Engine Journal – New Data Finds Gap Between Google Rankings and LLM Citations
- SparkToro – LLMs Are Not Search Engines (Rand Fishkin)
- Moz – Google vs. Generative AI: How Search Is Changing
- Perplexity AI Hub – How Perplexity selects and cites sources (official documentation)
- OpenAI – ChatGPT: principles, limitations, and use of sources
- Nature – Large Language Models Don’t “Know” Things the Way Search Engines Do
