It has been revealed that three websites have mass-produced over 210,000 'best software' pages targeting AI search engines, with Perplexity using them as the basis for its answers.
Imagine this: you are starting a new project and ask Perplexity, an AI-based answer engine, what the ‘best task management tool’ is. The AI instantly presents a very plausible-sounding recommendation list. We usually trust that AI answers based on thoroughly verified data, so we often blindly trust and choose from these recommendations. But what if the ‘best software’ information you are reading is actually just one of hundreds of thousands of auto-generated pages containing zero human experience?
A shocking report from the technology research firm Trellner was recently released. It revealed that just three websites mass-produced a staggering 215,128 ‘best software’ recommendation pages specifically designed to rank at the top of AI search engines [Source: 215128 ‘bestsoftware’ pages for AI search: Trellner report].
Why is this happening?
AI search has completely changed how we find information. However, this incident suggests that the answers AI provides might not always be based on human ‘intelligence.’ In fact, about 60% of the information Perplexity and similar AI platforms cite when recommending software comes from obscure websites with very low visitor counts [Source: 215128 ‘bestsoftware’ pages for AI search: Trellner report].
Simply put, there is a high probability that the AI we use daily is not listening to the opinions of skilled experts, but is instead treating data from ‘automated factories’ that churn out hundreds of thousands of pages to exploit search engine flaws as if it were fact. This warns us that the quality of information recommended by AI can be much more opaque than we expect.
A Library Analogy
Let’s compare this to a library. Suppose you are looking for the ‘best cookbook’ in a library. What if 6 out of the 10 books on the shelf were actually ‘fake books’ written by people who have never cooked? The librarian (AI) would be recommending them to you simply because the titles sound plausible and they have many pages, without checking whether the content is genuine.
These ‘made-for-AI pages,’ which abuse ‘Search Engine Optimization (SEO)’ techniques to stuff keywords indiscriminately, contain absolutely no human experience or insight [Source: Measure ChatGPT Visibility: Track LLM SEO Performance]. We may have hired AI as a smart assistant, but that assistant might just be reading the nameless ‘billboards’ of the internet world and feeding them to us.
Current Responses and Limitations
Each AI search model has slightly different criteria for handling data [Source: Measure ChatGPT Visibility: Track LLM SEO Performance]. Currently, Perplexity constructs answers based on broad web data, and in this process, information for over 380 software categories is cited from sites outside the top 100,000 in visitor traffic [Source: Hacker News – Telegram].
The good news is that AI platforms are continuously upgrading their technology to identify this kind of ‘junk data.’ Pages that lack real human experience or value and are merely keyword-stuffed are increasingly likely to be considered unreliable by AI platforms and not adopted as the basis for answers [Source: Measure ChatGPT Visibility: Track LLM SEO Performance].
How to Use AI Smartly
Going forward, the source of information—’who wrote this’—will become even more critical. Beyond just ranking at the top of search results, only information containing real user reviews or analysis from verified experts will survive in the AI search answer window.
As users, rather than blindly trusting what AI recommends, it is essential to develop the habit of clicking the ‘citations’ at the bottom of an answer to check which site provided the information. Because AI is merely a tool, the final responsibility for judging whether the result that tool brought back is true still rests with us.
References
- 215128 ‘bestsoftware’ pages for AI search: Trellner report
- Measure ChatGPT Visibility: Track LLM SEO Performance
- Hacker News – Telegram
- Hacker News Mirror
- progscrape: trellner.com
-
[Natural 20 — AI News in Real-Time The Bloomberg Terminal for AI](https://natural20.com/c/2px2gq)
- About 100,000
- About 215,128
- About 500,000
- About 20%
- About 60%
- About 90%
- Automatically generated content lacking human value
- High-quality news articles
- Well-known blog posts