Three sites made 215,128 "best software" pages for AI. Perplexity cites them (trellner.com)

A recent analysis reveals that three newly established websites collectively generated a staggering 215,128 "best software" pages using AI, significantly influencing online software recommendations. Conducted by leveraging web-grounded models like perplexity/sonar, the study reviewed 380 software categories, highlighting that over half of the cited domains ranked poorly or were outside the top million in the Tranco rankings. The implications of this study are profound for the AI/ML community as it underscores the reliance on questionable sources for AI-generated recommendations, reflecting potential biases in how AI systems retrieve and prioritize information. The findings also emphasize the need for transparency in AI recommendations, as many of these sites operate similarly and are often marketed through content that lacks genuine review or comparison. The analysis noted that a significant portion of the citations came from lesser-known domains, some of which are linked to marketing content rather than independent reviews. This raises concerns about the quality and reliability of AI-generated outputs and challenges the community to scrutinize the sourcing of data that informs AI models, urging developers and users to consider the implications of relying on possibly engineered recommendation systems.