Data2Story turns a CSV file into a verified interactive news article using seven AI agents

2026-06-20 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Emerging Technologies & Innovation · Depth: Intermediate, medium

Summary

The "Data Journalist Agent" (Data2Story), developed by Oxford and Stanford researchers, is a Claude Code skill that transforms CSV files into interactive, verifiable online news articles. This AI pipeline automates data journalism, generating research context, statistics, and graphics. A core "Inspector" feature links 93 percent of all visible statements, charts, and interactive elements to their evidence, such as code, data sources, or external URLs, significantly surpassing the 25 percent verifiability of human-written articles. Data2Story operates using seven specialized AI agents, including a Detective for context, an Analyst for calculations, and an Editor for narrative. In a study with 53 readers, agent-generated articles were preferred by 74 percent over human originals across five categories, with a +1.49 lead in transparency. While excelling in data-heavy content, Data2Story currently falls short in editorial perspective, creative design, and dense single graphics compared to human journalists.

Key takeaway

For data journalists or newsroom teams struggling with capacity for data-heavy investigations, Data2Story offers a powerful automation solution. You can significantly boost article verifiability and production speed for niche datasets. Consider integrating multi-agent AI systems to handle computational tasks and graphic generation, freeing human journalists to focus on critical editorial perspective, creative design, and complex narrative "why" explanations that AI currently cannot provide.

Key insights

Data2Story automates verifiable data journalism using a multi-agent AI system, significantly improving transparency and efficiency.

Principles

• Verifiability through source linking is crucial for AI-generated content.
• Specialized AI agents can mimic complex editorial workflows.
• AI excels at data processing but struggles with human editorial perspective.

Method

Data2Story employs a "virtual newsroom" of seven agents (Detective, Analyst, Editor, Designer, Programmer, Auditor, Inspector) to research, analyze, narrate, design, build, check, and source an article from a CSV.

In practice

• Automate data-heavy reports from niche datasets.
• Implement an "Inspector" panel for claim traceability.
• Integrate multi-modal models for diverse content generation.

Topics

• Data Journalism Automation
• Multi-Agent AI Systems
• Content Verifiability
• Claude Code
• AI-Generated Content
• Newsroom Technology

Code references

• rfordatascience/tidytuesday
• QinghongLin/data2story-skill

Best for: Research Scientist, AI Product Manager, Product Manager, AI Scientist, AI Engineer, Tech Journalist

Related on AIssential

• CNN, The New York Times, Reuters, and Hacks/Hackers on AI in the Newsroom: What’s Working, What’s Next, and What’s at Stake — Major news organizations are actively integrating AI, balancing innovation with ethical considerations.
• Why Your Agents Can’t Read Enterprise Documents — and How to Fix It — Accurate document processing is foundational for trustworthy agentic workflows, especially with complex enterprise documents.
• joeseesun / qiaomu-anything-to-notebooklm — This tool converts diverse content into various formats using AI, featuring robust paywall circumvention and multi-source integration.
• I Built a Fully Automated AI Publication in a Few Hours. Here’s Exactly How It Works. — Transparent AI-generated content, with distinct bot personas, can create engaging and cost-effective digital publications.
• ‘News’ Site Keeps Hallucinating EFF Staffers — AI-generated "news" sites are fabricating sources and quotes, damaging trust and spreading misinformation.
• How The San Francisco Standard is Reinventing the News App: In Conversation with Kevin Delaney & Jim Friedlich — AI can transform news consumption by personalizing content and synthesizing atomic reporting units into dynamic user experiences.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.