A new research system called Data2Story can take a plain CSV file — the kind of spreadsheet data that underpins much of modern reporting — and turn it into a finished, interactive news article, complete with graphics and source links.
According to The Decoder, the project comes out of Oxford and Stanford and works much like a small newsroom. Instead of relying on a single model, it splits the job across seven AI agents that collaborate. At the center is a "Data Journalist Agent" that handles the core work of building the article, layering in graphics and web research along the way.
The headline claim is about trustworthiness. The Decoder reports that the system attaches verifiable source links for 93 percent of all statements in the articles it produces — an attempt to address one of the biggest worries about AI-written text, which is that it can state things confidently without any way to check them.
Readers seemed to respond well. In a reader study cited by The Decoder, 74 percent preferred the agent's output.
The details of how the seven agents divide the labor, and how the verification holds up across different datasets, will matter for judging how far this can go. But the basic pitch is clear: rather than a chatbot guessing at facts, this is a structured pipeline that ties its claims back to sources.
Why it matters: if AI tools can reliably turn raw data into readable, source-linked stories, they could speed up data journalism while making automated writing easier to fact-check — a meaningful shift in a field anxious about AI accuracy.