Three sites made 215,128 “best software” pages for AI. Perplexity cites them
- ID
- 20907
- Status
- summarized
- Published
- 02 Sep 2026, 9:59 PM
- Fetched
- 04 Sep 2026, 4:03 AM
- Provider
- Hacker News
- Category
- dev-community
- Original URL
- https://trellner.com/reports/manufactured-sources-behind-ai-recommendations/
- Source URL
- https://hnrss.org/best
Summary
- Score
- 8.0
- Created
- 04 Sep 2026, 4:06 AM
- Tags
- Audience
- developersai_agent_usersai_ml_learnerssaas_founders
What happened
A Trellner Research report found that 59.8% of citations behind Perplexity's grounded software recommendations point to domains outside the top 100,000 most-visited websites, and 23.4% to domains not even in the top 1 million. Three sites—none existing before December 2023—published 215,128 machine-generated 'best <category>' pages, with two even labeling their homepage HTML title 'Facts & Grounding Page,' explicitly designed to be read by retrieval models rather than humans.
Why it matters
If you ship AI agents that ground recommendations in web search, assume the retrieval layer is already being gamed at scale—cheap content farms now target model citations directly. SaaS founders should not assume AI-powered recommendation engines surface quality; they surface whatever is structured for retrieval, meaning SEO-for-AI is already a real attack surface that can displace your product in AI answers.
Discussion angle
How should builders of AI agents validate grounding sources—should you filter citations by domain age, Tranco rank, or human-traffic signals before trusting retrieved content, and what does this mean for SaaS companies whose discoverability now depends on being cited by models reading spam?