Skip to content

Applications of Generative AI for Economic Research

Key Ideas

  • Comprehensive guide covering seven application domains of LLMs for economists: ideation and feedback, writing, background research, coding, data analysis, math, and research promotion
  • As of November 2024, all tasks rated at least "useful" with many rated "highly useful," representing significant improvement over earlier versions
  • Coding is identified as the domain with the greatest productivity gains, with tools like Copilot and Cursor enabling rapid high-quality code production
  • LLMs excel at text synthesis, editing, evaluation, translation, and generating titles but still struggle with precise academic citations (though LLM-powered search tools are improving this)
  • Data analysis capabilities include extracting data from text, reformatting data, classifying/scoring text, and sentiment extraction -- all rated highly useful
  • Research promotion is a new category added in Fall 2024, with social media posts, slides, blog posts, interviews, and podcasts all rated highly useful
  • Math capabilities (setting up models, deriving equations, explaining models) remain in the "useful but requires oversight" category

Summary

Anton Korinek's comprehensive guide, a companion website to his Journal of Economic Literature paper, systematically catalogs and evaluates LLM applications across the full economics research workflow. The guide covers seven domains: ideation and feedback, writing, background research, coding, data analysis, mathematical derivations, and research promotion. Each domain includes concrete prompt examples, model responses, and critical analysis of output quality. The rating system ranges from experimental to highly useful, and remarkably, by November 2024 no task falls below "useful."

Writing applications include synthesizing bullet points into prose, editing for grammar and style, evaluating text quality, converting handwritten equations to LaTeX, and generating titles. Background research capabilities span summarization (now excellent given large context windows), literature search (improving via LLM-powered search but still imperfect), translation, and concept explanation. The coding section highlights Python and R proficiency, noting that LLMs benefit from vast GitHub training data. Data analysis applications include text classification, sentiment extraction, and data reformatting.

A notable theme is the rapid pace of improvement. Tasks that were "experimental" in early versions are now "highly useful." The guide also identifies persistent weaknesses: LLMs still hallucinate academic references (though dedicated tools like Elicit help), math capabilities require oversight, and literature synthesis remains imperfect. The Fall 2024 update added research promotion as a category, reflecting new capabilities in generating slides, blog posts, and even podcast content.

Relevance to Economics Research

This is arguably the most comprehensive and authoritative guide to practical LLM use in economics research, published in the field's premier survey journal. It provides economists with a structured framework for identifying where AI tools can enhance productivity and where caution remains warranted. The domain-by-domain rating system offers a practical roadmap for adoption, while the concrete examples lower the barrier to entry. The guide's iterative updates track the rapid evolution of capabilities, making it a living reference for the profession.