How AI picks software recommendations

When asked for top software products across 380 categories, Perplexity relied heavily on obscure, low-traffic websites to supply evidence.

“Of the 7,534 citations that came back, 59.8% point at domains ranked worse than #100,000 in the Tranco top-1M list”

The rise of obscure sources

Over 23% of citations used to ground AI answers came from websites outside the top million most popular domains. Wikipedia was cited just three times.

“Wikipedia, for comparison, was cited three times in 7,534.”

Marketing blogs beating industry leaders

A software vendor's content marketing blog became the third most-cited domain overall, outranking established research giants like Gartner.

“Its blog was nonetheless cited 194 times across 96 of our 380 categories — a quarter of them — placing it third overall and ahead of Gartner.”

Synthetic content on a massive scale

Three connected websites published over 215,000 automated software comparison pages in under three years to flood search indexes.

“They and a third site under apparently common control have published 215,128 machine-generated best <category> pages between them”

Built for bots instead of humans

These automated networks titled their homepages 'Facts & Grounding Page' to directly target the machine retrieval steps used by AI engines.

“Grounding is not a term buyers use. It is the name of the step in which a retrieval system fetches documents to condition an answer on.”

Contradictions and fake bylines

The synthetic sites cited fictional experts and contained unrendered template code. They gave contradictory rankings for identical software queries.

“All three carry an unrendered template variable in the byline line, reading 'Within the next 26 days' on two of them”

Bad links and gambling redirects

Over one percent of vendor links supplied by the AI were completely unreachable. Other links redirected users to online gambling portals.

A shared engine under the hood

Perplexity's standard and pro models returned identical citations in 289 categories, revealing they rely on the exact same underlying retrieval system.

“so the Perplexity tiers share a retrieval layer and should be read as one search stack sampled twice.”

A new era of AI optimization

Spammers are moving beyond traditional human-focused search optimization to target the hidden machine retrieval pipelines powering modern AI.

“What we measured is which documents the evidence base is made of.”