Data Journalist Agent: Transforming Data into Verifiable Multimodal Stories

A new multi-agent framework called Data Journalist Agent can transform raw datasets into fully formed, evidence-traceable multimedia news articles, according to research submitted in 2026 [1]. The system, also referred to as Data2Story, orchestrates seven specialized artificial intelligence roles to replicate the workflow of a human newsroom [2]. The framework assigns distinct tasks to a Detective for context hunting, an Analyst for statistics, an Editor for narrative framing, a Designer for visual assets, a Programmer for website creation, and an Auditor for reviewing output [3]. A seventh role, the Inspector, is designed to link every claim, number, and visual element in the final article back to its original source in the data, code, or an external reference [3]. This architecture makes the resulting articles independently verifiable and auditable, a feature the researchers identify as the system's primary advantage over human-produced work [4]. The authors evaluated Data2Story on 18 articles, comparing each against a professionally published piece [2]. A rubric-based assessment with 53 participants measured the output across five dimensions [2]. The study also employed computer-use agents as a cost-saving proxy for reader navigation and used a coding verifier to re-execute statements against the underlying data [2]. The results showed that Data2Story produces competitive multimedia stories, with particular strength in transparency and auditability [2]. Human-authored articles retained a clear edge in editorial angle, creative design, and informative presentation [4]. The project builds on earlier work in agent-based journalism. A separate 2024 study demonstrated that a pipeline of generative agents—an analyst, a reporter, and an editor—could produce tip sheets for investigative data reporting, significantly enhancing newsworthiness and validity compared to a baseline without agents [5]. That earlier system was explicitly positioned as a tool for generating leads, not as a replacement for the reporting process itself [5]. The Data2Story framework extends this concept into full article generation, adding multimodal reasoning that deploys interactive maps, audio, and playable game elements depending on the topic [3]. The researchers position Data2Story as a collaborator for journalists, aiming to enable more evidence-based, transparent, and verifiable reporting [2]. Code and demonstrations for the project are available on the project's website [1].