What a New Study Reveals About How AI Engines Choose Who to Cite

What the Study Did

The researchers at Peak Answer wanted to understand which sites AI engines actually cite when someone asks a practical buying question. They chose a specific category: startup brand identity design. They wrote 25 questions that a founder might realistically type into a chat window. Questions like "who are the best brand identity designers for startups" or "which agencies should I consider for a startup rebrand." Then they ran those questions through ChatGPT six times each. That produced 150 answers to analyze. They also ran the same questions through Google's AI Mode and through standard Google search, so they could compare the results.

The methodology is straightforward. The findings are not.

The Key Findings

ChatGPT cited 302 unique websites across the 150 answers. Sixty seven of those sites appeared on two or more different questions. But the top of the list is dominated by a small number of platforms. Upwork was cited more than any other site, appearing in 40 of the 150 answers and on 10 of the 25 questions. Dribbble came second, appearing in 30 answers and on 9 questions. The rest of the list includes a mix of agency sites, portfolio platforms, directories, and resource pages.

The most striking finding is how different ChatGPT's sources are from Google's. Google relied heavily on Reddit, which appeared in its organic results for 23 of the 24 questions that returned results. ChatGPT cited Reddit on only one question. Of the 302 sites that ChatGPT cited, only 39 appeared anywhere in Google's results. That is a small overlap. The two systems are looking at different parts of the web.

What This Means for AI as a Back Web Presence

This divergence points to something important. AI engines are not simply reading the same web that humans read. They are building a parallel map of it. They cite sources that a human searcher might never see. They weight platforms like Upwork and Dribbble because those platforms contain structured lists, service descriptions, and named providers. They ignore Reddit, which Google loves, because Reddit threads are conversational and unstructured. The result is a hidden layer of visibility. A site can be invisible to Google and still be cited by ChatGPT. A site can rank well on Google and never appear in an AI answer.

This is what we mean by a back web presence. It is a presence that exists for machines, not for human readers. It is built from the structured data that AI models can extract and repeat. It rewards clarity, consistency, and specific named entities. It punishes ambiguity, marketing language, and pages that are hard to parse.

Why This Matters for Anyone Trying to Be Found

If you are a designer, a founder, or anyone who depends on being discovered, the study is a useful map. It shows you where AI engines are actually looking. The full breakdown of cited domains is in the study, and it is worth reading closely: https://peakanswer.com/research/startup-brand-identity-design

The study does not offer a magic solution. It does not say that every designer should join Upwork or Dribbble. But it does show that AI visibility is not random. It follows specific sources. It rewards specific kinds of content. It is a new kind of web presence, one that operates behind the chat window, and it is already shaping who gets recommended and who does not.

The old game was about ranking. The new game is about being legible. The study is one of the first attempts to measure that legibility in a real category. You can read it here: https://peakanswer.com/research/startup-brand-identity-design

‍