Answers are grounded in this paper and cite their sources.
Large language models (LLMs) have clear potential to contribute to psychological research methods. However, as with any new research method, LLM-based methods require rigorous evaluation. Towards this end, Austin and colleagues evaluated the ability of two LLMs (GPT-4o and Gemini 2.0 Flash) to facilitate scale validation by identifying relevant criterion measures and generating validity hypotheses for assessing convergent and discriminant validity. We argue that their study has several key limitations that limit the conclusions that can be drawn about the proposed method's suitability for use in validation research. We situate our critique within three recommendations for incorporating LLMs into research: (a) avoid essentialisation and overgeneralisation about 'AI', (b) use application programming interfaces when evaluating LLMs in research contexts and (c) evaluate LLM-based methods using clearly specified criteria. Finally, we consider whether adopting the proposed method might reinforce or amplify existing limitations in the psychological literature, namely, the proliferation of overlapping constructs and inadequately validated measures.
No takes yet. Share an insight, caveat, or question.
Higgins et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: