A decade ago, the concept of liquid content emerged as a response to the fragmentation of devices, platforms, and audience consumption habits. The idea was that publishers could create modular or atomic stories that could be versioned cheaply for multiple channels, increasing reach.  

Except that it did increase workload, and the work was too manual for publishers to sustain. But the idea may have just been a decade ahead of its time.

Liquid content—sometimes called structured content, modular content, or atomic content—is the practice of breaking stories into reusable components that can be reassembled for different formats, audiences, and platforms. The concept first gained traction a decade ago as publishers struggled to meet rising distribution demands across web, mobile, and social channels. Back then, the tools weren’t mature: versioning content for every platform created more work, not less. Today, generative AI, improved personalization, and cross-platform automation are reviving the strategy. With AI now able to help structure, maintain, and re-render content, publishers are revisiting liquid content not only to serve current channels more efficiently, but to future-proof their operations for distribution platforms and channels that don’t yet exist. 

For David Cohn, senior director, generative AI innovation for content and the newsroom at Advance Digital, the concept itself isn’t new, just newly practical. From 2021-2014, he ran Circa , which was a mobile news app that restructured news into atomic units like events, statistics, quotes, and images that could be resurfaced and reused. The problem was scale: even with a globally distributed team working 24/7, Circa could barely keep up with the biggest stories of the day because every atom of news had to be created by humans. It wasn’t sustainable. 

“Now, we just need more compute,” Cohn says. “And, while compute is expensive, it’s a lot cheaper and a lot more doable.” So, he says, it’s time for media execs to revisit “those ideas around structured content and see if they are more attainable and more worthy of our attention.”  

The data layer: from story to query 

Cohn’s current work at Advance Digital, while still in the thinking and testing phase, treats liquid content less as format differentiations, and more like infrastructure. Instead of a one‑off story, an article is framed as a query across a structured datastore of atomic units like facts, quotes, statistics, dates, and other fields extracted from reporting.  

An article then becomes more like a SQL query that joins pieces of verified information and uses an AI layer to render them as a traditional story, a video script, or another format altogether.​ In other words: it is rendered whatever way people prefer to have their content delivered . 

“But, what we’re in charge of is, I’ll use the word purity, of the data: how well it’s structured, how well it’s maintained, how much we add to it,” Cohn says. “That can be a proprietary data set. It is in our ability to create those proprietary data sets that we all of a sudden have value over large, giant technology corporations.” 

Cohn is quick to flag that structured, liquid systems excel when audiences want clarity and up-to-date facts but are a poor fit for narrative non-fiction like the classic Gay Talese’s Frank Sinatra Has a Cold. “This type of structured content would be terrible to apply to that story,” he explains. “It’s about knowing what type of news and information this is great for and what types it is not. There are certain types of stories that are best told as a narrative. It’s challenge to make sure it’s applied in the right way.” 

But, the upside, according to Cohn, is creating a massive system of highly adaptable content. “Once you start to do this at scale, you end up where the parts are more valuable,” he says. “There’s a number of ways that we can imagine that that kind of information becomes useful once it becomes a database that you can query.” 

Digital media companies’ vast archives may become useful datasets in this framework. At SXSW this year, Lee Enterprises CIO Virginia Fletcher suggested that media outlets get their archives in order, wrote Alex Mahadevan for Poynter.  With a structured database of articles, media companies can convert what they have into different mediums.  

This sentiment was echoed by Jane Barrett, head of Reuters AI strategy, this year at the Nordic AI in Media Summit, noting their vast archive has inconsistent metadata. “Generative AI gives us the possibility to classify and structure our archive more effectively,” she said in a panel discussion. This moment is pushing us to think about our content as data. As Barrett put it, “it’s kind of giving us a good challenge to think now about our content as data and what needs to be true to turn our content into data that then can flow into these liquid experiences.” 

The knowledge layer: encoding journalistic value 

While Cohn’s focus on liquid content is its data system, researcher Sannuta Raghu, who heads Scroll’s AI Lab, pushes the idea one layer deeper, to what she calls the knowledge layer of journalism.  

During her ICFJ Knight Fellowship in 2024, she mapped the structural layer of journalism, and compiled her findings in what she calls the Directory of Liquid Content. The directory is a scalable and modular taxonomy designed to map, describe and standardize how digital news is structured, styled and surfaced.  

“Liquid content is not necessarily moving something from one format to another,” she says. “It’s moving something, or it’s using information such that it can be contained in any container on demand.” 

Liquid content can move between various containers – from an article to a podcast, a calculator, or a decision tree, she says. Within these, it can also be formatted in various styles, as an inverted pyramid, an interview or an explainer for short TikTok videos.  

“So, it’s making sure that you are able to have information that is dynamic enough that can be adapted to various containers, with high fidelity to the source where it came from,” she says. 

 Months into that fellowship, she was granted another Fellowship at the Reuters Institute, where she developed the News Atom to codify journalism’s epistemic layer. Raghu explains when AI parses a news article, “what they’re essentially doing is they’re looking for statements which can be converted into causal relations and subject object predicate,” she says.  

When a news article is used as raw material to answer questions, AI will remove the intentionality of the signals a journalist has added. These can include the temporal aspect, words that describe when something happened, complex quotes and nested quotes.  

“There are many subtle examples,” Raghu says. “Words like likely, reportedly, allegedly, very intuitive to us as journalists, but are considered extra when it comes to form.”  

The News Atom provides a way for the news industry to structure, define and embed its own meaning rather than having that meaning imposed by external systems. Raghu believes if publishers don’t define and encode their own values, like trust, attribution, temporal precision, into the data they produce, those values won’t exist for AI models.  

Liquid content in 2025 and beyond, then, is about survival in an ecosystem and consumption experience of on-demand form. This on-demand form transformation is already happening around the world, Raghu points out. Take for example, Google’s NotebookLM, which converts PDFs into mindmaps and podcasts. 

“For us, it’s about making sure that you are codifying some of the practices that are intuitive to us. Like the inverted pyramid, for example, is a very journalism format. But, it needs to be codified and taught to a model in a very deterministic way so that it can be replicated and converted into multiple formats.” 

For publishers, the question is whether to codify their practices and values into these systems, or whether those definitions will be written by others. As Cohn puts it: “We’re in a world where generative AI feels like the beginnings of the internet again, where we can fundamentally rethink assumptions.” 

“We want to both understand it and have a solid foothold in it, rather than just be steamrolled by it,” Cohn says. “Has this fundamentally changed everything for Advance? No, but what we’re doing is to prepare if that does become the case, it’s on our terms.” 

The shift toward liquid content is not simply a tactical experiment. As AI systems increasingly generate fast, unverified content in every format, publishers’ defensible advantage is the integrity and structure of their own data. Pursuing liquid content at scale may require rethinking stories as more than finished containers and embedding core journalistic values — attribution, verification, temporality, precision — into the underlying layers where content is created, stored, and reused. 

This approach also raises the stakes for content architecture. Unstructured archives function as low-value files that can be scraped or summarized by AI. Structured, verified, and queryable archives become high-value assets, built from components that cannot be easily replicated with the same fidelity. Liquid content, in this context, is not a mandate but a strategic shift: a way for publishers to strengthen the foundations of their work and ensure it can adapt to the formats and distribution environments still to come.