Panel: Using LLMs for text analysis
Governments around the world articulate their approaches to Artificial Intelligence (AI) through national AI strategies. While framing theory provides a useful lens for analysing these documents, methods for measuring political frames vary and remain contested in the literature. This article asks how governments frame AI and why these frames vary across countries. To answer these questions, I develop a systematic, inductive method to measure frame configurations at scale. The method operationalises Entman’s classical notion of frames as configurations of problem definitions, causal attributions, moral evaluations, and treatment recommendations. Using large language models (LLMs) and comprehensive human validation, I inductively extract frame configurations and calculate frame salience scores – the distribution of emphasis across co-present frames – for a multilingual corpus of all available national AI strategies published between 2017 and 2026, covering 86 countries. I then relate these frame salience scores to country-level characteristics, including income, region, regime type, and AI capacity, to explain cross-national variation in government framing. Preliminary analysis suggests that national AI strategies typically combine multiple co-present frames rather than relying on a single dominant frame. These frame salience distributions cluster into recurring configurations, with capacity-building emphasis as the most common, while sovereignty and foreign-dependence frames are concentrated among lower-income and middle-power states. The article makes three contributions. Theoretically, it conceptualizes frames as configurations, or distributions of emphasis across co-present frames, extending Entman's framing theory to the comparative study of technology policy. Methodologically, it offers a fast, reproducible, and transparent instrument for measuring frames at scale, validated against human judgement, addressing long-standing concerns about the reliability of frame measurement. Empirically, it provides the first systematic account of how, and why, governments frame AI differently across the world.
