AI engines search and cite: 960 answers measured (tryprefer.com)

In a comprehensive study conducted by Prefer, four AI engines—ChatGPT, Perplexity, Gemini, and Claude—were tested on 80 questions regarding AI search, GEO, and AEO, with a total of 960 answers recorded. The study highlighted significant differences in how each engine utilized web searches and cited sources: Perplexity led with a 100% search rate and cited an average of 19.48 sources per response, whereas Claude had the lowest search rate and cited only 4.58 sources on average. Remarkably, an overwhelming majority (89.7%) of answers came from live web searches, underscoring the engines' heavy reliance on current online content. This study is crucial for the AI/ML community as it reveals the variances in information sourcing and citation practices among leading AI engines, which can influence the reliability and authority of AI-generated answers. The findings indicate that while some engines provide more diverse citations, others rely on a narrower range of sources, potentially impacting the user experience and trust in the information provided. The data is freely accessible under a CC BY 4.0 license, allowing for further analysis and understanding of these AI systems’ behaviors, which is essential for developers and users aiming to refine their interactions with AI-driven content.