AI-Powered Search Engines Rely on “Less Popular” Sources, Researchers Find

Why It Matters

The reliance on obscure or outdated sources raises concerns about the credibility, bias, and timeliness of AI‑generated answers, prompting the industry to develop new evaluation standards and influencing how businesses and users trust generative search tools.

Summary

Researchers from Ruhr University and the Max Planck Institute compared Google’s AI Overviews, Gemini‑2.5‑Flash, GPT‑4o and GPT‑4o with Search Tool to traditional Google organic results and found that AI‑powered searches cite far less popular domains, often falling outside the top 1,000 or even top 1,000,000 sites tracked by Tranco. More than half of the sources used by Google’s AI Overviews do not appear in Google’s top‑10 links for the same query, and 40% are absent from the top‑100. While GPT‑based searches favor corporate and encyclopedia sites and avoid social media, they provide comparable conceptual coverage but tend to compress information, missing secondary or ambiguous details, especially for ambiguous queries. Gemini in particular shows a strong bias toward low‑popularity domains, and GPT‑4o with Search Tool frequently relies on internal knowledge, limiting its ability to fetch up‑to‑date information.