Data tells stories that shape society; the data journalist's job is to turn raw information into stories non-experts can trust. A high-quality news feature takes a newsroom team weeks: hunting for context, running statistics, choosing an angle, and designing visuals. Recent agents handle individual steps well: data-science agents close the analysis loop, while design agents synthesize beautiful websites. But can an agent serve as a data journalist end to end? We introduce Data Journalist Agent (Data2Story), a multi-agent framework that orchestrates specialized roles into a single virtual newsroom. Data2Story contributes two innovations. (i) Claims are evidence-grounded: an Inspector links every number, angle, and asset back to data, code, or an external reference. (ii) Articles are multimodally generative: rather than defaulting to plain text and static charts, Data2Story reasons about what readers will want to see, then deploys multimodal tools, such as interactive maps for geography and audio for music. We evaluate Data2Story on 18 articles, each paired with the originally published expert piece, along four axes: (a) human-agent angle coverage; (b) rubric evaluation with 53 participants across five dimensions; (c) computer-use agents as judges, a cost-saving proxy for how readers navigate interactive articles; and (d) verifiability, where a coding verifier re-executes statements against the data and checks claims against references. Data2Story produces competitive, evidence-traceable multimedia stories, with particular strength in transparency and auditability. Human articles retain an edge in editorial angle, creative design, and presentation. We position Data2Story as a collaborator for journalists, enabling more evidence-based, transparent, and verifiable reporting. Code and demos are available at https://data2story.github.io.
Alex: Welcome to another episode of ResearchPod.
Sam: Today we're looking at a system called Data2Story — a framework designed to turn raw data into interactive news articles automatically. The central challenge it's trying to solve is producing data-driven journalism that is both engaging and verifiable.
Alex: So the paper is asking whether we can build an AI journalist that doesn't just make things up, but provides a receipt for every claim it makes?
Sam: Exactly. The core problem with AI-generated text right now is what researchers call hallucination — where the system confidently states things that aren't true, with no way to check where the claim came from. Data2Story tries to solve this by forcing every claim, statistic, and chart to link back to its original source, whether that's a specific line of code or a data file.
Alex: That's a meaningful shift. Usually, when you read a news article, you just have to trust that the journalist did the math correctly.
Sam: That's the traditional model. Here, the authors introduce what they call an "Inspector" — think of it as a digital auditor. Every time the system makes a claim, the Inspector attaches a record of where that claim came from. In research terms, this is called "provenance," which just means documented origin — like a paper trail.
Alex: So if I'm reading an article and I see a chart, I could click on it and see the actual evidence behind it?
Sam: Precisely. You'd see the exact code or data that produced that visual. It's like having footnotes, except every number has a digital receipt attached that you can follow all the way back to the original source.
Alex: And this isn't just one piece of software doing all of this?
Sam: No — it uses what the paper calls a multi-agent framework. Imagine a virtual newsroom with seven specialists. One agent acts like a detective, gathering background context. Another is the analyst, doing the actual number-crunching. A third is the designer, deciding how to present the visuals. They each handle a different part of the job and pass the work along.
Alex: And the end result isn't just a plain text article?
Sam: Right. The paper calls it "multimodal" — which just means the story uses multiple formats at once. Instead of a static page of text, the system reasons about what the reader actually needs and picks the format that fits the data best. An interactive map for a geography story, a timeline for a historical one.
Alex: How does the reader experience that, practically speaking?
Sam: The articles are built for what the paper calls "scrollytelling." As you scroll down the page, charts and maps animate in response to your movement. The data becomes part of the narrative flow rather than something separate you have to go look up.
Alex: That sounds well-designed, but what happens if the layout breaks — if a chart overlaps with the text?
Sam: That's where a second quality-control agent comes in, called the Auditor. It scans the finished page for structural problems. If it finds something broken, it sends the page back to the programmer agent to fix it. So the system is running two separate checks — the Inspector verifies that every claim is tethered to evidence, and the Auditor verifies that the presentation actually works. It's a closed loop.
Alex: Does this hold up on genuinely complex data, or only on simple, clean datasets?
Sam: The paper tests it on a fairly involved case. The researchers used climate data to reframe the FIFA World Cup schedule — not as a sports calendar, but as a document about heat risk. The system fused weather data with the fixture list and produced an interactive map showing heat-risk levels at specific venues.
Alex: So the AI didn't just report scores — it found a completely different story hiding in the data?
Sam: That's the idea. And because the Inspector is active throughout, every finding in that story is clickable. If the article states that a certain percentage of matches were scheduled at high-risk venues, you can click that figure and see exactly which data source produced it. The abstract statistic becomes something you can verify yourself.
Alex: That does sound like original research, not just summarizing a press release.
Sam: It is closer to analysis than summarization. The paper's argument is that this kind of traceability addresses the "black box" problem — the reader doesn't have to take the AI's word for anything. They can follow the receipt back to the source at every step.
Alex: But is it actually as good as a human journalist?
Sam: The study is honest about the limits. Participants in the evaluation found that professional journalists still hold a clear edge in creative design and editorial framing — the kind of judgment that shapes what a story means, not just what it says. The system is best understood as a tool that handles the heavy analytical work, leaving the human editor free to focus on narrative and meaning.
Alex: So it's not about replacing the journalist's voice. It's about giving that voice a more reliable, evidence-based foundation to work from.
Sam: That's a fair reading of it. The system takes on the data processing, the chart generation, and the verification chain. The human takes on the story. Whether that division of labour proves useful in real newsrooms is still an open question, but as a proof of concept, the paper makes a credible case that automated journalism doesn't have to mean unverifiable journalism.
Alex: That's a useful distinction to end on. Thanks for listening to ResearchPod.