Study finds AI users credit themselves for AI-generated answers, risking overconfidence in work they didn't actually do

Researchers at arxiv.org published a study (paper 2604.14807) documenting that people using large language models (AI systems that generate text) routinely mistake output produced by the AI for their own knowledge or skill, without recognizing the AI did the work.

The gap matters because users who feel competent without actually doing the thinking may make confident decisions on topics they don't understand — a programmer might submit AI-written code they can't debug, or a business analyst might present AI analysis they can't defend in detail.

For knowledge workers relying on AI tools daily (coders, writers, analysts), this is a practical warning: if you're using ChatGPT, Claude, or similar tools without actively verifying and understanding their output, you may be building a false sense of expertise that fails when the AI is unavailable or wrong — affecting your actual job performance and credibility.