Snake eating tail: Google’s AI Overviews cites web pages written by AI, study saysResearchers also found that more than half of citations didn’t rank in top 100 for termAI + ML07 Sep 2025|18 (www.theregister.com)

A recent study by Originality.ai revealed that Google’s AI Overviews (AIOs), which summarize search results for high-stakes topics like health and finance, cite web pages generated by other AIs about 10 percent of the time. This recursive citation of AI-written content raises concerns about an echo chamber effect and the risk of "model collapse," where training on AI-generated outputs could degrade future AI understanding and propagate biases. With over half of AIO citations not even appearing in Google’s top 100 search results for a query, the study highlights potential challenges in source reliability and the evolving role of AI in content curation. While Google disputes the findings, arguing that AI detection tools are error-prone and emphasizing that AIOs prioritize relevance over human authorship, the issue remains significant for AI/ML communities. It underscores the complexities of maintaining content quality and originality as AI becomes both a generator and curator of information. Technically, Google’s use of its Gemini LLM in AIOs leverages diverse sources, including PDFs and specialized documents not ranked highly by traditional search, complicating the citation landscape. As AI-driven summaries suppress link clicks to original websites, this shift also threatens the sustainability of human publishers, further accelerating AI’s impact on the information ecosystem.