About seven months ago, a website publishing 50,000 fabricated statistics went live and has since been cited extensively by AI models such as ChatGPT. Despite Google deindexing the site, the fake data has driven over 10,000 AI sessions and attracted backlinks from more than 330 domains, including Shopify, Bluehost, and DHL, according to peec.ai.
The site was created as an experiment by peec.ai to explore how large language models (LLMs) source and use data. The founder noticed that LLMs often seek seemingly reliable statistics to fill knowledge gaps when generating content. By seeding the site with fake statistics, the experiment tested whether these numbers would be incorporated into AI-generated answers and cited externally, including in an academic paper and a bank's investment research.
This case highlights challenges in AI content reliability and the risks of unverified data propagation. While Google removed the site from its index, AI models like ChatGPT continue to reference the fabricated statistics, demonstrating the difficulty of controlling misinformation in AI outputs. The experiment underscores the importance of grounding AI responses in verified data to prevent the spread of false information.
The website’s fake statistics have influenced over 330 referring domains and have been cited in formal research contexts, illustrating the broad impact of AI-generated misinformation. The experiment’s findings were detailed in a peec.ai blog post published recently.