Publish fake stats, get cited in LLMs
Points and comments are a snapshot, not live.
A researcher seeded 50,000 fake stats online, and ChatGPT still cites them 7 months later.
The author launched stateglobe.com with 50,000 fake statistics generated by GPT-4.1 in one day for ~$50. Google deindexed the site after two updates, but ChatGPT-User bots sent over 600,000 requests in 30 days. The site now has 330+ referring domains (Shopify, DHL, Bluehost), was cited in an academic paper, investment research by a Qatari bank, and an Indian economics publication. LLMs retrieve stats by rewriting prompts into fanout queries, favoring numbers that look fresh and specific. The author warns this enables 'citation laundering' and advises publishing real, verifiable stats with sources and dates.
What commenters are saying
Commenters are alarmed that LLMs can be gamed so easily, with many noting the lack of epistemological grounding in AI outputs. Several point out that fact-checking is technically possible but not implemented due to cost and latency. A camp argues this mirrors existing human problems with fraud and SEO spam, not a new crisis. Others highlight that the problem is worse because LLMs present fake data with authoritative confidence, laundering credibility through citations.