Report: Perplexity AI Cites 215,128 Machine-Generated 'Best Software' Pages

Perplexity AI Citation Quality Under Scrutiny

A Trellner report published on September 2, 2026, reveals that Perplexity AI's citation-grounded product recommendations frequently originate from a network of 215,128 machine-generated 'best software' pages, raising concerns about its source quality compared to established platforms like Wikipedia or Google. For broader context, explore our AI News.

Understanding the Research Methodology

Trellner's investigation involved making 760 calls to Perplexity's perplexity/sonar and perplexity/sonar-pro models. These calls covered 380 distinct buyer-intent categories, resulting in the collection of 7,534 citations from 2,055 unique domains. The researchers then analyzed the Tranco rank of these cited domains to assess their authority.

The Prevalence of Low-Ranked and Machine-Generated Sources

The report's findings highlight a significant reliance on less authoritative sources. Specifically, 59.8% of the citations came from domains ranked worse than #100,000 on the Tranco top-1M list, and 23.4% were from domains not even within the top million. The median Tranco rank for all ranked citations was 71,611. In contrast, Wikipedia, a widely recognized authoritative source, was cited only three times out of the 7,534 collected citations.

A notable discovery was the presence of a network of machine-generated content. Three specific sites — wifitalents.com, worldmetrics.org, and gitnux.org, which were not registered before December 2023, have collectively published 215,128 machine-generated 'best software' pages. These sites reportedly cross-link each other and a fourth brand, with two of them labeling their homepages as 'Facts & Grounding Page'. The report suggests these sites are optimized for retrieval by AI search engines rather than for informing human buyers. Furthermore, worldmetrics.org offers custom research starting from €5,000 and vendor selection from €2,500.

The study also found that guideflow.com, a demo-software vendor, was the third most-cited source, appearing in 96 of the 380 categories examined.

Conflicting Information and SEO Tactics

The report identified conflicting top-five rankings for the same questions across the three machine-generated content brands (wifitalents.com, worldmetrics.org, and gitnux.org). This inconsistency raises concerns about the accuracy and reliability of information derived from such sources. The researchers suggest that these findings indicate AI assistants' evidence bases are being 'farmed' through link-farm-like SEO tactics, potentially compromising the integrity of the information presented to users.

Comparison: Perplexity AI vs. Traditional Search and Knowledge Bases

While Perplexity AI aims to provide direct, citation-grounded answers, the report by Trellner indicates a potential vulnerability to content designed to manipulate AI search algorithms. Traditional search engines like Google continuously evolve their algorithms to identify and penalize low-quality and machine-generated content, striving to present authoritative sources. Wikipedia, as a collaborative knowledge base, relies on community vetting and editorial processes to maintain accuracy and prevent the proliferation of unreliable information.

Implications for Users and AI Development

The findings suggest that users relying on Perplexity AI for product recommendations or factual information should exercise caution and consider cross-referencing information, especially when dealing with commercial or critical decisions. For AI developers, this report underscores the ongoing challenge of ensuring the quality and trustworthiness of the data sources that ground AI models. The 'farming' of evidence bases by SEO tactics represents an evolving threat to the reliability of AI-generated content and recommendations.

Conclusion

The Trellner report highlights a critical aspect of AI search: the quality of its underlying citations. While Perplexity AI offers a novel approach to information retrieval, its reliance on a significant number of low-ranked and machine-generated sources, as detailed in the study, presents a challenge to its claim of providing authoritative, grounded answers. As AI search continues to develop, the mechanisms for evaluating and prioritizing source authority will be crucial for maintaining user trust and delivering genuinely reliable information.

Sources

• GitHub - HuestonCo/perplexity-citations-study: Comprehensive analysis of 23,936 Perplexity citations revealing how AI search engines evaluate authority. Dataset includes 6,606 domains across 23 verticals showing YouTube's 3.3x dominance and why position #1 no longer matters in AI search.
• perplexity-citations-study/README.md at main · HuestonCo/perplexity-citations-study · GitHub
• Perplexity AI Citations · Issue #996 · pydantic/pydantic-ai · GitHub
• GitHub - mishamyrt/perplexity-web-api-mcp: 🔍 Perplexity AI MCP without API key