Summary

GIJN's March 2026 guide is one of the clearest recent explanations of where AI is actually useful in investigative journalism and where it is dangerous. Rather than selling AI as a universal newsroom upgrade, it narrows the useful cases to leak triage, pattern detection, PDF and scanned-record extraction, timeline building, entity organization, and targeted image review, all under strong human oversight. It also explicitly warns that prompts are not methodology, that cloud tools can expose sensitive data, and that closed commercial systems can create legal, privacy, and dependency problems for investigative teams.

Why It Matters

This is a strong direct journalism workflow record because it focuses on investigative practice rather than generic publishing automation.

  • it identifies which investigative tasks benefit most from AI: large document troves, hidden-pattern detection, extraction from unstructured records, and scalable organization
  • it gives named examples from ICIJ and other investigative groups instead of abstract claims
  • it emphasizes that AI should be problem-driven and tightly scoped rather than used as a general reporting substitute
  • it treats verification, methodology disclosure, and source protection as workflow requirements, not afterthoughts

Investigator Workflow

The clearest investigator tasks here are closed-corpus document triage, timeline generation, entity-list building, and identity-document review inside large case or leak archives. The workflow maturity is `advanced workflow` because the useful pattern is not a single chatbot prompt but a structured, human-supervised pipeline for narrowing large collections to the subset worth manual review. The journalism workflows are source-stated; the private-investigator adaptation is an internal inference, but it is direct because the same tasks map to civil-case file review, fraud inquiries, locate work, and evidence-sensitive OSINT.

What the Source Says

The guide says journalists increasingly want practical help using AI for analyzing large document troves, connecting disparate datasets, detecting patterns, mapping networks, and triaging large leaks without exposing sensitive data to third-party platforms. It cites ICIJ's Implant Files investigation, where machine learning helped analyze more than eight million health records and identify over 2,100 previously obscured patient deaths after journalists fact-checked the results. It lists concrete present-day AI uses in investigative journalism, including retrieval-augmented querying of curated document collections, extracting names and tables from PDFs and scans, building timelines and entity lists from court records and contracts, and locating specific visual items such as passports in leak archives. The guide also stresses recurring failure modes: hallucinations, non-reproducible prompt behavior, legal and privacy risks from commercial platforms, and global capacity gaps between well-funded and smaller newsrooms.