AI search systems are becoming information gatekeepers — and a new study suggests that some of the sources influencing their recommendations come from obscure websites built as much for machines to read as for people.
WHAT’S HAPPENING
Trellner Research tested Perplexity’s Sonar and Sonar Pro models across 380 software categories, collecting 7,534 citations from 2,055 different domains.
The researchers found that 59.8% of citations came from websites outside the 100,000 most-visited domains, while 23.4% came from sites outside the top million entirely.
Several little-known sites appeared repeatedly.
Three sites identified by the researchers contained more than 215,000 generated “best software” pages, while another company’s marketing blog was cited more often than Gartner.
Wikipedia appeared only three times in the entire dataset.
WHY IT MATTERS
Traditional SEO was built around one goal:
Get humans to find your website through search engines.
AI changes that equation.
If AI systems increasingly answer questions directly, businesses and publishers may start designing information specifically to become part of the material AI retrieves when forming an answer.
The competition may shift from:
“How do we rank on Google?”
to:
“How do we become a source the AI trusts?”
WHO BENEFITS
Smaller publishers and businesses may have a new way to influence discovery without becoming one of the internet’s most heavily visited websites.
Companies that structure useful, machine-readable information could gain visibility when AI systems assemble recommendations.
WHO LOSES
Established publishers and review platforms could lose some of their traditional influence if AI retrieval systems increasingly pull information from thousands of lesser-known sources.
Consumers could also have difficulty knowing why an AI recommended one product over another — or which sources shaped that answer.
WHAT HAPPENS NEXT
This study examined Perplexity’s retrieval system only. It does not establish that ChatGPT, Gemini, Claude or other AI platforms behave the same way.
It also does not prove these websites manipulated Perplexity or caused incorrect recommendations.
What it does expose is a new battleground.
The internet was built around getting human attention.
The next web may increasingly compete for AI attention first.
