Perplexity cites 215,128 auto-made 'best software' pages
A Trellner study of 7,534 Perplexity citations found 60% point to sites ranked below #100k — and one demo vendor's blog outranked Gartner.
Copy markdown
Three sites, 215,128 pages, all built for the bots
Trellner Research traced 215,128 templated /best/<category>-software/ pages to three linked sites — worldmetrics.org, gitnux.org and wifitalents.com — that share Cloudflare nameservers and were all registered within six months of each other in 2023-24. Perplexity's Sonar models cited them 181 times across the test.
Most of what AI cites lives in the internet's basement
Across 380 software categories and 7,534 collected citations, 59.8% pointed to domains ranked worse than #100,000 on the Tranco list and 23.4% weren't ranked at all. When authoritative sources thin out in a niche, retrieval grabs whatever filled the gap.
A demo startup's blog outranked Gartner
Guideflow, a product-demo vendor's marketing blog, landed as the third-most-cited domain overall — surfaced across 96+ categories it has no real expertise in, ahead of Gartner. Not a spam farm, just volume: the quieter failure mode where marketing content out-reaches analysts.
The tell: pages that introduce themselves to machines
The farm pages label themselves a 'Facts & Grounding Page' — a 'machine-readable record' addressed to retrievers, not humans. It's the GEO (generative-engine-optimization) playbook in the open: structured, fact-shaped pages exploiting retrieval-layer gaps that classic search spent two decades hardening.
What this means for you
If you lean on Perplexity or ChatGPT to pick tools, treat 'best X' answers as leads, not verdicts, and click through to primary sources. If you ship a dev tool, the uncomfortable flip side is that structured fact pages are demonstrably what's getting cited right now.