Fake stats get cited in LLMs
Archived — this story has rotated out of today’s deck. It is kept here in full.
The gist
GEO researcher Metehan Yesilyurt built a site of 50,000 invented statistics for about $30, and ChatGPT cited them.
Fake numbers reached an academic paper and a bank's research, showing AI search can launder invented data.
Background
Metehan Yesilyurt is a GEO (generative engine optimization — the practice of trying to get content cited by AI chatbots) researcher at Peec AI, whose blog published the experiment write-up. He noticed that as more teams write time-sensitive content with LLMs (large language models), those models look for seemingly reliable data, and statistics are often exactly what they want. So he built stateglobe.com, which claims to offer statistics on SEO, content marketing, social media, e-commerce and web technology from 200 countries.
How it unfolded
- Aug 2025Seer Interactive published a study on AI optimization that inspired the experiment.
- Mar 2026Yesilyurt started building the site for the experiment.
- recentlyHe launched stateglobe.com with 50,000 statistics pages generated in a single day with GPT-4.1, every number invented, at a cost of around $30; 15,000 listicles were later added for about $20 more.
- recentlyTwo Google updates came out right after launch and Google visibility collapsed, with indexed pages bouncing between 5 and 20 impressions.
- recentlyAfter seven months the site had received more than 10,000 AI sessions and more than 330 referring domains, including Shopify, Bluehost and DHL; made-up numbers ended up in an academic paper and in a bank's investment research.
- recentlyIn the last 30 days before writing, the site got more than 600,000 requests from bots, around 72,000 of them from ChatGPT-User.
- Oct 13, 2026Adam Danyal published a LinkedIn post on earning AI citations citing the Peec AI write-up and Yesilyurt's suggestions.
- Oct 15, 2026The University of Richmond published a LibGuides page on AI in academic research.
- Oct 18, 2026Blaksheep Creative published an article on AI citation myths.
- Oct 20, 2026The Peec AI article was submitted to Hacker News, drawing 11 points and 4 comments.
Who’s saying what
- Researcher
- Yesilyurt says Google deindexed the pages but ChatGPT hasn't, and that each ChatGPT-User request is the chatbot fetching a fake statistics page to answer a real question.
- Caution
- The University of Richmond LibGuides says LLMs frequently fabricate data and present falsehoods as truth, and that the citations they generate tend to be even less reliable than their truth claims, so researchers must verify all data and sources.
- Public
- Hacker News commenter khalic called it "fine work" and said more communication to the general public is needed.
- Public
- Hacker News commenter jqpabc123 said "AI doesn't know or care if the data is fake."
Still unverified
Yesilyurt says some pages might fail to load now and then and that he doesn't know; the University of Richmond LibGuides notes that estimates of LLM "hallucination" rates vary.