Brian Chau explains that Effort News uses large language models to query databases containing millions of legal and financial records, while people write the resulting stories. Brian Chau says this changes investigative journalism by making it affordable to pursue many weak leads, discard unsuccessful searches and manually verify the few findings that matter.
Brian Chau reports that current models are highly reliable when extracting clear facts and numbers from source documents, although interpretation and ideological bias still require review. Effort News publishes null results as well as successful investigations, an approach intended to show when broad searches fail to support a hypothesis instead of hiding those dead ends.
Brian Chau identifies important limits to automation. Public-records requests can involve negotiation with officials, and reporting on war or other physical events still requires evidence that databases cannot provide alone. Brian Chau nevertheless sees opportunities in AI-assisted polling, financial auditing and large-scale plagiarism checks, where models can assemble and cross-reference evidence before humans make accountable judgments.
Brian Chau contrasts a three-person AI-assisted publication with large legacy newsrooms and argues that automation radically lowers the cost of document-led reporting. Brian Chau attributes slower adoption in established media partly to an innovator's dilemma: organizations may resist tools that threaten existing jobs and workflows even when those tools expand the scale and presentation quality of investigative work.
Watch on YouTube
