Nieman Journalism Lab: The Guardian's first reader-facing AI product brings narrative to category pages

The Guardian, a major British news organization, launched its first reader-facing AI product in late March 2026. The product is called Storylines, and it is notably conservative for a major publisher — which is partly the point.

What Storylines does

Tag pages on The Guardian list articles related to a given topic (such as “Trump” or “NHS”) in reverse-chronological order. Before Storylines, those pages were essentially archives. The new tool adds a module that reads the most recent 200 articles on a tag and identifies three major storylines across them — surfacing narrative threads that a reader arriving cold might not otherwise perceive.

The AI generates only three subtitle-length descriptions of those storylines. The articles linked from each storyline are human-curated journalism, not generated content. The module is clearly labeled as AI-generated, and the editorial team built in a “very large red button” for immediate deactivation if problems arise.

The design constraints

Chris Moran, head of editorial innovation at The Guardian, described the decision-making behind the product’s cautious scope: “We’re not going to die if we don’t build a chatbot tomorrow.” The organization’s concern was hallucination — the risk of an AI producing a plausible but inaccurate summary of a complex ongoing story.

To mitigate this, the tool feeds only article headlines to the language model, not full article text. That constraint limits what the model can fabricate because it has less material to misinterpret. A team of 20 senior editors spent two weeks reviewing outputs and providing detailed feedback before launch. The product launched on just 10 tag pages, with certain sensitive topic categories excluded entirely.

Why it matters

The Guardian’s approach illustrates a specific decision point that many news organizations face: where to use AI in the reader-facing product and how to contain the risk of errors in coverage areas where accuracy is non-negotiable. Rather than building an AI layer across the entire product, the team isolated a narrow use case — improving the context and navigability of existing archives — where the stakes of a mistake are lower and the editorial benefit is clear. The result is a product that is both modest in scope and instructive as a model for how journalism organizations can introduce AI incrementally without compromising their relationship with readers.