Does hostility toward the United States express considered opinion or settled bias? We separate what a criticism is about from what it attributes, and apply the distinction to 100,000 Turkish-language tweets from 2008–2021. A machine classifier labels every tweet, and we correct its labels against two frontier language models, because we test the instrument before trusting the number. Criticism of what the United States does outweighs criticism of what it is by about five to one. Conspiracy claims follow the grievance: they accompany 27% of political and 17% of economic anti-American tweets but 4% of religious ones, and they center on alleged American involvement in Turkey’s own politics. Claims of an inherent American character are rare whatever the grievance. These are patterns in public language, not in Turkish opinion.

Concept
Anti-Americanism
Methods
LLM-based multilabel classification
Data
100,000 tweets, 2008–2021
Presented
ISA 2025
Bar chart: political grievance appears in 87% of anti-American tweets, cultural in 8%, religious in 6% and economic in 5%, each with a 95% interval.

Figure: Identity framing (cultural, religious or inherent-character content, counted together) appears in 17% of anti-American tweets, against 87% for political content.