The Extended Brief

Three sites made 215,128 “best software” pages for AI. Perplexity cites them

Brief by The AI News AI newsroom · Sep 2, 2026, 12:31 PM EDT edition

Original reporting by Hacker News · published Sep 2, 2026, 9:59 AM EDT

Anyone trusting AI software recommendations is often relying on sources built for models, not people — three sites alone made 215,128 machine-generated pages Perplexity cites.

Key points

• Three apparently connected sites published 215,128 machine-generated "best software" pages and are cited by Perplexity's models. source ↗
• 59.8% of 7,534 citations from Perplexity's Sonar models point to domains outside the 100,000 most-visited websites. source ↗
• 23.4% of citations point to domains absent from the Tranco top-million list entirely. source ↗
• Two of the most-cited sites title their homepage "Facts & Grounding Page," naming the models' retrieval step. source ↗
• None of the three domains existed before December 2023. source ↗

The data

Numbers from the original article, machine-verified against its text

Practical applications

• Before shortlisting software from an AI assistant's answer, check whether the cited sources are recognizable publications or obscure low-traffic domains.
• Query Perplexity Sonar for your own product category through OpenRouter and inspect which domains ground the answer, replicating this study's method.
• If you build a grounded AI product, add a popularity-list check such as Tranco rank to flag or down-weight citations from very low-traffic domains.
• Marketing teams should audit whether their category's AI answers are shaped by machine-generated "best of" pages rather than editorial reviews.

Context

Web-grounded AI search models like Perplexity's Sonar retrieve web pages at query time and cite them as the basis for answers, a step called grounding. This study sent 380 software-buying queries to Sonar and Sonar Pro via OpenRouter, then checked the 7,534 cited domains against Tranco, a research ranking of the million most-visited websites. The concern is that sites publishing large volumes of machine-generated pages can be built specifically to be retrieved, shaping what the models recommend.

What to watch

• Whether Perplexity or other AI search providers de-rank or exclude the three domains would show if this is treated as abuse.
• The study's pending results on the 1,502 vendor homepages, and any replication that includes the omitted Google models, would extend or challenge the findings.

Related briefs

• ChatGPT Ads Hit $1 Billion Faster Than Google Ever Did
• ChatGPT Facing Dual Regulatory Regimes Under New EU Designation
• WeChat Pay expands AI AgentPay Card to DeepSeek Harness and OpenClaw
• OpenAI Lets Some Customers Pay Only When AI Performs

This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.