Gaps in large language model awareness, usage, and perceptions in the United States: Evidence from a nationally representative longitudinal survey

Large language models (LLMs) have the potential to benefit users in both their work and personal lives, but which groups are quickest to adopt them? To investigate awareness, usage, and perceptions of LLMs among US adults across socio-demographic groups—and to track changes over time—we administered a two-wave survey using a nationally representative, probability-based online panel of 12,224 US residents. Across two survey waves spanning 1 year, we observed marked gaps in usage: groups more likely to use LLMs included men, younger adults, those with college education and higher incomes, individuals in more analytical occupations (e.g. STEM), Democratic-leaning respondents, and those with above-median cognitive ability, internet literacy, and openness to experience. These usage gaps do not appear to be declining and, in many cases, seem to be widening over time. Our analyses indicate that these disparities are associated with differences in both access-related factors (e.g. income, occupation, digital skills) and individual traits and preferences (e.g. openness to experience, political orientation). Overall, our data provide a dynamic picture of the rapidly evolving exposure to, adoption of, and attitudes toward LLMs in the US population.