On Liquid Content and LLMs

Edition no 7. Date: 5 May 2026
If you like this newsletter, please forward it to a friend. You can sign up here.

A “new” phrase is floating around in recent online discussions. It’s about “liquid content”, and it has nothing to do with physics. The Reuters Institute for the Study of Journalism describes it as “content or stories that are not static but adapt in real time based on the viewer's context, location, time, or interaction. AI facilitates this by tailoring content to individual preferences. Requires traditional media companies to move away from authoring ‘articles’ towards more flexible atomic objects.”

The idea behind it is to deliver news and articles that are highly personalised to readers. This personalisation can be based on several factors, such as location, language, time of day, and what (and how) the reader has been reading on the given platform.

On paper, it sounds nice, right? You get what you want when you want it, LLMs will help transform it, and publishers will have new opportunities. But by doing this, you miss a few important elements.

First, the context. This is essential to understanding the big picture. If everyone gets the same enthusiastic AI summary about the news, nobody will make sense of the world. You need context to make connections, to process information, to ask yourself: Is this for real? How does this piece of news impact my city, my business, my community?

Second, uniformity. When someone receives the same type of content every day, without challenging views, that person will stop thinking critically and will become a target for manipulation. We already see it on social media platforms: the same trends, challenges, and content structures, where everyone copies everyone and tries to rig the algorithm.

Third, your north star metric. We know from social media platforms how they are built for engagement. For example, Facebook’s North Star metric is “daily active users”, Netflix has “watch time”, and Airbnb has “nights booked”. They will do everything to keep you hooked on their platforms, including letting harmful content to boost engagement, as Facebook did.

All this polarisation and negative conflictual feeds get translated into real life behaviour by attacks on democratic institutions, decreased community involvement and day-to-day trust between people. Liquid content is an intentional decision in controlling what content gets put in front of the reader, and the opportunities for bad actors to shape culture and public perception are high.

Fourth, identity. I’m reading Arwa Mahdawi’s column in The Guardian for her voice, style, wit, and the context she brings, and not for how The Guardian presents me with the content. Her individual voice and style are defined by experiences and backgrounds, but also by the mediums it inhabits. If Arwa Mahdawi makes a podcast, she will structure, organise, and edit differently from the article.
This is true for other brands or newsrooms: the medium is part of the message. This is what will differentiate you from the next competitor.

Fifth, control. When, between you and the reader, a third-party acts as intermediary and transforms your message, you lose control and transparency. An LLM is a prediction machine, trained on opaque datasets and patterns. How can I know if an LLM-generated summary will get information only from the summarised article? That reader will be captured by a walled garden and will not interact with your platform. Even the best LLMs have hallucinations.
In a review study published in 2025, the average hallucination rates for LLMs were between 5 % and 20 %. For a domain where accurate and fact-checked information is paramount, these hallucinations will miss the whole north star metric of a newsroom.

But what can I do?

The LLM personalisation route may seem like the easier path. However, I recommend that news products avoid competing on this front. Instead, they should focus on making their content more discoverable, structured, readable, and engaging by addressing outdated information architecture.

A widely used alternative is the personalisation engine. Gartner defines this as “software that applies context about individual users to select, tailor, and deliver messaging, such as content, offers, and other interactions, through digital channels.” These engines utilise contextual data—including location, language, device, and user insights derived from interactions, transactions, surveys, and input from internal teams—to create a tailored experience.

The key distinction, however, is that the 'actual content' remains unmodified. The editorial team retains control over consumer flows, while readers maintain agency over what information they choose to share. Ultimately, this approach is less about controlling what readers see and more about equipping them with the tools to navigate a complex informational landscape.

Currently reading

I’m finishing the novel “The Sirens” by Emilia Hart. It’s a story about sisters, history and drama, told with the help of three narratives: Lucy in 2019, Jess in 1999, Mary and Eliza in 1800.
Here’s a quote: “Lucy understands the mindset of small towns, the mythology that a community weaves like cloth. A story repeated so many times that every local knows where to draw breath in the telling. Dawes Plain has its own ghosts.”

Recommendations

• An article published last year, that was stating that the use of ChatGPT is beneficial for learning performances, has been retracted. From the retraction note, published on Nature’s website: ”The Editor has decided to retract this paper owing to concerns regarding discrepancies in the meta-analysis. These issues ultimately undermine the confidence the Editor can place in the validity of the analysis and resulting conclusion”.

• The philosopher and social theorist, Zygmunt Bauman, describes today’s culture as something that “seeks no longer to enlighten the people but to seduce them. The function of culture today is not to satisfy existing needs but to create new ones, while simultaneously ensuring that existing needs remain permanently unfulfilled.” From his book “Culture in a Liquid Modern World”.

• About AI Slop: “We don't see it because, well, we're conditioned not to, but slop always arrives on time. Slop is inevitable. Slop is quintessential. Slop is everywhere for those with eyes to see.”

• “So what is software brain? The simplest definition I’ve come up with is that it’s when you see the whole world as a series of databases that can be controlled with the structured language of software code. As I said, this is a powerful way of seeing things. So much of our lives run through databases, and a bunch of important companies have been built around maintaining those databases and providing access to them.” via The Decoder podcast, on why people hate AI, the false marketing problem of AI, jobs and automations.

Let’s work together

I’ve updated my services pages on my website. I’ve added six packages with transparent pricing for a better understanding of how we can work. Of course, if you need something more personalised, just reach out.

Photo by Cemrecan Yurtman on Unsplash