[DSRP Evidence](https://dsrpevidence.org/)

# Evaluating Context in LLM Prompts for Causal Inference in Empirical Sustainability Studies

Zhang and Fan, 2026, Environmental Research Communications — Engineering

Patterns: [Distinctions](https://dsrpevidence.org/pattern/distinctions), [Relationships](https://dsrpevidence.org/pattern/relationships)

## In short

An AI system can correctly identify causal relationships and their direction yet fail at the accompanying distinction task of bounding which connections genuinely belong, revealing that inferring a relationship and correctly delimiting it are separable competencies.

## What they found (results)

Benchmarking GPT-5 against five peer-reviewed sustainability studies, the authors found the model recovers core causal edges with moderate accuracy (0.33-0.85) and assigns direction reliably, but shows a strong tendency toward over-connecting variables with spurious causal links.

These researchers were not testing DSRP. The finding is theirs; the correspondence to DSRP is drawn by this site.

[Source](https://doi.org/10.1088/2515-7620/ae9d05)