A new study suggests a large share of the sources behind Perplexity's software recommendations are obscure, machine-generated pages rather than established review sites. Trellner Research asked two Perplexity tiers for the best products in 380 software categories, from CRM to museum collection management, and analysed the 7,534 citations the models returned. Nearly 60% pointed to domains ranked outside the 100,000 most-visited websites, and 23.4% were not in the top million at all. The ten most-cited domains accounted for just 17.3% of citations, so the pattern was not a handful of famous sites dominating the answers.
The report highlights three sites, Gitnux, Worldmetrics and WifiTalents, which published 215,128 'best software' pages between them even though none of the domains existed before December 2023. They share registrars, nameservers and page templates, and their homepages carry the HTML title 'Facts and Grounding Page', language aimed at retrieval software rather than human readers. Separately, Guideflow, a vendor selling interactive product demos rather than reviews, was cited 194 times across 96 categories, ahead of Gartner. Trellner also found 17 of 1,502 recommended vendor homepages were unreachable, including picks that now redirect to an online gambling portal and a Monaco hotel group.
The findings are a reminder that AI search is becoming a discovery channel founders can influence, but one whose evidence base is still easy to pollute. Machine-generated listicles built to be read by models can outrank genuine editorial coverage in grounded answers, which matters for founders chasing AI citations and for anyone trusting an AI's 'best software' list. Trellner notes the study covers Perplexity only and is a one-day snapshot, so the exact numbers may shift, but the direction is worth watching as AI-driven buying advice grows.