Definitions, Label Names, or Demonstrations? Auditing What Governs LLM Annotation

Viraj Mishra, Shubham Gupta, Dhruv Gupta, Yuehong Cassandra Tai, Qunhua Li

Large language models (LLMs) are increasingly used to annotate social-science texts, but high accuracy does not establish that they apply researcher-supplied category definitions: predictions may instead reflect prior associations with category labels or labeled demonstrations. We introduce a counterfactual audit that reassigns definitions among output labels, distinguishing adherence to supplied definitions from preservation of a dataset's original mapping. Across three political-text datasets, models rarely follow reassigned definitions, and standard fine-tuning strengthens the original mapping without reliably increasing definition adherence. We then study retrieval-augmented annotation. Retrieving the nearest training examples generally outperforms retrieving examples separately within each category despite using fewer demonstrations. On the two tasks with substantial gains, direct \(k\)-nearest-neighbor classification recovers 74--93\% of the improvement, indicating that neighborhood labels provide much of the predictive signal. Relabeling fixed demonstrations redirects predictions, whereas aligning demonstrations with reassigned definitions substantially increases adherence…