Metehan Yesilyurt generated 50,000 pages of invented statistics and watched them acquire real citations. His account at Peec AI describes a cheap way to pollute AI-assisted research: publish precise-looking numbers, then let other publications give them borrowed authority.
According to his account:
- The initial pages and another 15,000 listicles cost about $50 to generate.
- Google visibility collapsed, but ChatGPT kept retrieving pages and sent more than 10,000 referral sessions over seven months.
- More than 330 domains linked to the site. Screenshots show citations in an academic paper and a bank’s investment research.
These are the author’s observations, supported by screenshots rather than an independent audit. Retrieval requests and backlinks do not establish how often false numbers reached an AI answer. Later contamination of model training data remains a possibility he raises.
Following a citation to another page is only the first check. Follow the number back to the original measurement, and check who measured it and how.
What the thread adds
The 100-comment thread on Hacker News adds historical context and an ethical objection:
- acadapter — calls the older circular-citation problem “citogenesis,” a reminder that repeated publication can manufacture apparent corroboration without AI.
- KronisLV — objects to deliberately creating misinformation because it can feed real decisions; khalic instead welcomes public demonstrations of the weakness.
- cyral — recalls low-quality articles gaming search before LLMs, challenging the idea that this pollution began with generative AI.
An unanswered question: hmokiguess asks whether the site ranks on Bing; Hitton asks why Google demoted it. The article’s Google-versus-ChatGPT comparison leaves those mechanisms unresolved.
Handles are pseudonymous; HN publishes no per-comment scores, and the sampled comments follow HN’s ranking. This is a slice of the thread, not a consensus.