Dashboards claiming to show “millions of AI prompts” are proliferating. The pitch is simple: type in a term and you get back a precise number of times users asked it in ChatGPT, Gemini, or Claude. These numbers are being positioned as if they are observed counts, akin to search volumes in Google Keyword Planner.

The reality is less robust. None of the major LLM vendors provide raw usage logs. What is presented as factual is, in fact, modeled: panel data scaled to population estimates. When these projections are stripped of error margins and displayed as integers, they mislead.