We Still Don’t Know How People Are Really Using AI

A report from MIT Technology Review highlights a fundamental problem in the AI industry: we still don't have independent, reliable data on how people are actually using AI tools like ChatGPT and Claude. AI companies such as Anthropic and OpenAI regularly publish usage reports, but researchers argue these are selective and only release data the companies want the public to see. There is no independent source to corroborate these claims, says Anka Reuel, a PhD candidate quoted in the report. This lack of transparency makes it difficult to understand the true societal impact of these technologies. Without unbiased data, policymakers, researchers, and the public are left to rely on corporate narratives that may be shaped by marketing goals. For example, a company might highlight impressive user numbers while downplaying harmful uses or the frequency of errors. This information asymmetry is particularly concerning as AI tools become integrated into education, healthcare, and employment decisions. Claims about productivity gains or user satisfaction cannot be properly verified, leading to a distorted picture of AI's real-world effects. The report calls for more independent research and third-party audits of AI usage. This could involve partnerships with academic institutions, the release of anonymized datasets, or the creation of public observatories to track AI adoption. Without such measures, we risk making policy and investment decisions based on incomplete or biased information. The authors argue that understanding how people really use AI is essential for addressing issues like misinformation, job displacement, and algorithmic bias. Until then, the true scale and nature of AI usage remains a mystery, hidden behind corporate gatekeeping.