Claude Code Series (Part 9): Creating .bib Files and Automating Article Retrieval

  • Author/Source: Scott Cunningham (Baylor), via Substack ("Causal Inference")

  • Original: https://causalinf.substack.com/p/claude-code-series-part-9-creating

  • Key Ideas

  • Claude Code can automate the full literature-review pipeline: generate a .bib file of seminal papers on a topic, retrieve the PDFs from the web, organize them locally, create a LaTeX narrative document citing them, and produce a Beamer deck storyboarding the literature.
  • The entire pipeline (25 papers on economics of abortion access regulation) took about 50 minutes, including iterative deck refinement.
  • The workflow reflects how Cunningham processes research: piecing together the narrative of a literature is the cognitive work; Claude Code handles the mechanical retrieval, formatting, and visualization.

  • Deck iteration is part of the process -- repeatedly tinkering with slides until they communicate the story effectively is how the researcher internalizes the literature.

  • Summary

Cunningham demonstrates a four-step literature-review workflow using Claude Code on his HB2 abortion project. First, he asked Claude to create a .bib file containing the top 25 seminal economics papers on regulating abortion access (demand-side and supply-side), covering a range of outcomes. Second, Claude went online, retrieved the PDFs, and placed them in a local folder. Third, Claude produced a LaTeX document citing and organizing the 25 papers into a coherent narrative. Fourth, Claude built a Beamer deck storyboarding the literature visually.

The session took about 50 minutes, with a significant portion spent iterating on the deck -- tinkering with layout, narrative flow, and aesthetics until Cunningham was satisfied. He frames this as integral to his workflow: the deck is not a final product but a tool for processing and internalizing the literature. The bibliography is not exhaustive (the task was narrowly scoped to high-citation economics papers on access regulation), but it provides a working foundation and a .bib file ready for use in the manuscript.

  • Relevance to Economics Research

Addresses one of the most common complaints about ChatGPT -- its unreliability for literature reviews -- by showing how Claude Code's filesystem access and web retrieval change the equation. The ability to produce a .bib file, retrieve actual PDFs, and generate both a narrative LaTeX document and a visual deck in under an hour represents a major productivity gain for the literature-review stage of any empirical project. The hallucination risk remains (citations should be verified), but the mechanical labor of assembling bibliographies and organizing papers is largely automated.