{
  "id": 1689511,
  "title": "We still don’t know how people are really using AI",
  "url": "https://urgent.news/2026/08/18/we-still-dont-know-how-people-are-really-using-ai",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-18T10:06:43.000Z",
  "source": {
    "name": "MIT Technology Review",
    "slug": "mit-technology-review",
    "url": "https://www.technologyreview.com/2026/08/18/1142226/how-people-use-ai/"
  },
  "original_language": "en",
  "account": "AI companies such as Anthropic and OpenAI publish reports on the usage of their products like Claude and ChatGPT, but these reports are not fully representative of actual usage, according to AI researchers. Anka Reuel, a computer science PhD candidate at Stanford Trustworthy AI Research (STAIR) Lab and co-lead of the AI Observatory, explains that there is no independent source to corroborate the data provided by these companies. The AI Observatory is a public platform that aggregates and analyzes real AI conversations with popular models, collected with users' consent through seven existing datasets. The aim is to provide independent sources of information for researchers and policymakers to understand how people are using generative AI.\n\nThe research conducted by the AI Observatory reveals that AI usage varies significantly across different models and has changed over time. It also found that many sensitive behaviors are not captured in the data provided by major AI companies, as they focus more on work-related uses of AI, such as productivity-focused tasks. For instance, when applying Anthropic's methods to the AI Observatory's dataset, nearly half of the conversations (48%) would have been filtered out, focusing only on work-related topics. These filtered-out conversations were more likely to include health and relationship topics, adult or illicit content, harassment and hate, and sexual content, which were found in 44.2%, 7.9%, 27.5%, and 16.7% of conversations, respectively.\n\nFurthermore, the AI Observatory's findings indicate that the way people use AI has evolved over time. Conversations within WildChat, one of the largest datasets included in the study, became longer and more elaborate over time, with more small talk and less disclosure from AI assistants that they were chatbots. Additionally, exchanges labeled as sensitive use—potentially harmful or restricted content—dropped, possibly due to improved safeguards. The AI Observatory also discovered that AI usage differs depending on the model, with users employing Grok and Gemini more frequently for information retrieval, particularly on news and politics, while Anthropic was preferred for coding, and ChatGPT for homework assistance. Even within the same model, there were variations in user behavior depending on the version of the model. For example, people tended to have shorter conversations with ChatGPT using GPT-3.5, but longer and more iterative ones with GPT-4o.",
  "summary": "AI companies like Anthropic and OpenAI regularly publish reports on how people are using products like Claude and ChatGPT, but they only release the data they want us to see, AI researchers say. “There is no independent source to corroborate it,” says Anka Reuel, a Computer Science PhD candidate at the Stanford Trustworthy AI Research…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}