AI slop sites cited by Perplexity
Archived — this story has rotated out of today’s deck. It is kept here in full.
The gist
Perplexity's AI cited 215,128 machine-generated 'best' pages from three sites. 59.8% of citations point to low-traffic domains, raising questions about AI answer reliability.
Background
A report by Trellner Research analyzed 7,534 citations from Perplexity's AI models across 380 software categories. It found that 59.8% of cited domains rank below #100,000 in traffic, and three sites, apparently under common control, published 215,128 machine-generated 'best' pages. These sites are designed to be read by AI models, not humans, and are cited frequently, raising concerns about the quality and trustworthiness of AI-generated recommendations.
How it unfolded
- Dec 2023 - May 2024Three sites (wifitalents.com, worldmetrics.org, gitnux.org) are registered through NameCheap, sharing similar templates and nameservers.
- Sep 2, 2026Trellner Research runs 760 queries to Perplexity models across 380 software categories, collecting 7,534 citations.
- recentlyReport published revealing that 59.8% of citations point to domains ranked worse than #100,000, and that three sites produced 215,128 'best' pages.
Who’s saying what
- Analysts
- The prevalence of low-quality, machine-generated content in AI citations undermines the reliability of AI-powered recommendations.
- Public
- Users may unknowingly rely on AI answers that are based on content created solely to be cited by AI, not to inform humans.
Still unverified
The claim that the three sites are under common control is based on circumstantial evidence (shared nameservers, templates, and blog posts) but not proven.