Publish fake stats, get cited in LLMs

The experiment ran on stateglobe.com, a domain registered in March, where GPT-4.1 generated 50,000 statistics pages in a day. The author never checked a single figure. GA4 data shows more than 10,000 ChatGPT sessions to the site, and over 330 domains link to it. On October 7, the site ranked second for visibility across a few prompts tracked in Peec's tool, behind only Statista. Peec's GEO research team later repeated the test with real data, and this time they checked and cited every stat. The pull toward pages like this has a mechanical cause: LLMs add modifiers such as "statistics" to their sub-queries, then fetch pages without verifying the numbers. Rerankers score relevance rather than truth, so a specific figure looks more quotable than a vague claim. For a team building an AI product, a citation in a model answer does not confirm that a number is accurate. Retrieval pipelines that ground answers on web pages inherit this risk, so source checks belong before figures reach users. The author's advice for publishers, to place a source and year beside each number, also applies to any team that publishes its own data.