The liquefaction of journalism: AI is melting news into “liquid content,” and the ad market is recording the damage
Formats dissolve, inventory vanishes, prices spike. A tour of what publishers think survives the melt and the "shrinking" of the open web.
At the Media Voices Publisher Summit in London last week, Reuters Institute researcher Nic Newman gave a demonstration that Press Gazette covered under a phrase worth keeping: liquid content. Newman asked Claude about the UK heatwaves and got the answer back as a Q&A, as charts filterable by region, as Met Office warnings. Same underlying information, rendered on demand in whatever shape the user preferred. “Content is becoming much more fluid, whether we like it or not,” he told the room.
The phrase deserves attention because it names something the traffic-decline conversation keeps missing. For most of the industry’s history, format was the product. The article, the front page, the bulletin, the explainer, the newsletter: these were not neutral containers for journalism, they were the commodity form in which journalism was sold, and the publisher’s control over that form was where the business lived. What Newman is describing is format becoming a rendering decision, made at the moment of consumption, by software the publisher does not operate. The information persists. The commodity form dissolves. And his warning extends past the chat box: as agents begin completing tasks on users’ behalf, anticipating interests and delivering digests unprompted, the rendering moves even further from anything a publisher controls.
The state of things (surprise! — not great)
What this looks like in hard numbers arrived the following day, in benchmarking data from Ozone reported by Digiday. Ozone runs advertising infrastructure for a network of premium publishers including the Guardian, News UK, and the Wall Street Journal, and its dataset tracked roughly 20 billion impressions. The mechanism to hold in mind is simple: when search engines and AI interfaces answer questions in place, fewer people click through to publisher sites; fewer visits means fewer pageviews; and every lost pageview is a lost opportunity to sell an ad. This is not advertisers walking away from news. It is the shelf being removed from the store while demand stands in the aisle.
Between April and June, ad request volumes across Ozone’s network fell 32 to 37 percent year over year in the US and 39 to 41 percent in the UK. Stated plainly: two of every five ad opportunities that existed on these British sites a year ago are gone. Not repriced, not moved to another channel. Gone, because the pageviews that carried them never happened. This is the premium end of the market, the Guardian and the Wall Street Journal, losing a third to two fifths of its sellable surface in a single year. Programmatic spend across the first half of the year was down 30.6 percent.
Prices, meanwhile, moved the other way: with less inventory to bid on, buyers paid roughly 30 percent more per impression in the UK and 7 percent more in the US. Ozone’s own team reads the rising prices as a sign that premium inventory still commands demand; its COO, Danny Spears, reads the larger picture as a structural reset rather than a cycle, telling Digiday that “the open web is shrinking.” Both readings can be true at once. Higher yield is cushioning the fall, and the thing it is cushioning is a contraction of the surface the business was built on. The one channel growing in the dataset was apps, which is a detail worth remembering for what comes next.
Under the hood
The responses now circulating in the industry sort themselves along a spectrum, roughly ordered by time horizon.
At the near end sits yield defense. Some of the supply contraction, INMA’s Gabe Dorosz notes in the Digiday piece, is deliberate: publishers are cutting ad load and culling low-value bid requests, betting that three well-placed ads on an attentive reader are worth more than ten cluttering a page a visitor abandons. The logic inverts the programmatic era’s founding premise. For fifteen years the industry flooded the market with impressions on the theory that supply was effectively infinite, and the predictable result was a long slide in prices. AI-driven traffic loss has now demolished the infinity assumption from the outside, and publishers are discovering, under duress, that scarcity was the asset all along. Hence the strange spectacle of prices climbing through a collapse: this is what it looks like when a market built on abundance is forced to reprice around its opposite.
In the middle sits the retreat to owned surfaces: apps, subscription products, newsletters, logged-in environments. The appeal is control. On an owned surface no intermediary sits between publisher and reader, deciding what gets discovered; the relationship is direct, the data is first-party, and the pricing is the publisher’s to set. The app growth in Ozone’s numbers suggests this is already where the resilient money is. On how long the yield-led open web can hold in the meantime, Madison and Wall’s Luke Stillman offers a precedent from television, where analysts warned about a yield-led market for over a decade while marketers kept paying multiples for shrinking audiences. Buyers, in his reading, leave for reasons of data and measurement and convenience before they leave over price. His prescription is correspondingly unsentimental: take the money AI companies are now paying publishers for access to their content, both as training material for models and as source material those models draw on when answering questions, while that money is on offer; assume the open web ad business keeps deteriorating; and rebuild the P&L around subscriptions and events.
Between retreat and reframing sits a position the doom framing tends to skip: contesting the AI platforms rather than ceding them. Newman himself is explicit that publishers will need to build within these other experiences, on other people’s platforms and with all the risks that entails, much as they once built for search and social. And the licensing deals Stillman treats as harvest money can be read a second way. Structured well, they buy more than a check: they can require attribution, citation, and links, making the publisher the named source inside the answer rather than anonymous feedstock beneath it. And an entire practice is forming around the question of surfacing there.
Readers who have run into chatter about “AI SEO” or generative engine optimization have likely seen its marketing wing: brands trying to get their products named when a chatbot answers a shopping question, the way they once fought for the first page of Google. Publishers are adapting the same toolkit for news, structuring articles and data so that models retrieve them, and measuring how often and how prominently their work appears inside AI answers. The premise underneath the buzz is the one this whole piece has been circling: if the answer layer is where audiences now are, then visibility in the answer layer becomes a currency with advertisers and readers, the way search ranking once was. None of this has yet been shown to replace the economics it is meant to replace. But not every analyst treats the interface layer as surrendered territory, and the publishers negotiating hardest right now are negotiating over the terms of appearing inside it.
At the far end sit the strategic reframings, the attempts to say what journalism sells once format no longer holds. Newman’s survey of 264 news leaders shows an industry consensus forming around one word: distinctiveness. Original investigations and on-the-ground reporting ranked highest among things that will matter more in the AI era, followed by contextual analysis and community building. Service journalism, evergreen content, and general news ranked as mattering less, which is a quiet concession of those categories to the machines. The urgency behind the consensus is generational: 52 percent of 18 to 24 year olds now name social, video, and AI platforms as their main news source, up from 40 percent five years ago. And those categories are converging rather than sitting side by side. The social platforms are baking AI into their own surfaces, from assistants in the feed to AI-generated search and summaries, so the migration to social is increasingly a migration to AI intermediation by another route. Newman’s corollaries follow the same logic: invest in audio, video, and personality-led formats that resist liquefaction better than text, and fold AI’s capabilities into publishers’ own platforms rather than ceding the interactive layer entirely. The Washington Post is already running an on-site chatbot, and Newman noted that Norwegian media group Schibsted intends to roll out similar tools quickly.
Bert Kok’s recent LinkedIn essay pushes that reframing one step further and gives it operational teeth. The wrong question, he argues, is which articles can be turned into apps; that reproduces the content factory in a newer wrapper. Generic application shells will become cheap. What stays valuable sits underneath them: original reporting, inspectable evidence, knowledge kept current, editorial judgment about uncertainty, correction mechanisms, and an institution willing to accept responsibility. His test for a newsroom: “what can the user reliably do because our journalism exists?” Distinctiveness, in this telling, is not a tone or a brand posture. It is a list of verifiable commitments.
Who assembles the news now
Seen from a distance, the three developments describe one movement. For thirty years, the publisher’s page was where the news got put together: reporting, context, charts, and judgment assembled into a form a reader could use, with the ad impression taxing every act of assembly. Model interfaces move that assembly into the chat window, or into an agent’s morning digest. The publisher’s position in the stack shifts accordingly, upstream, toward supplying material to an interface it does not control and is not paid per rendering. The Ozone numbers are what that relocation looks like on an invoice.
What is notable about the current moment is that the diagnosis has consolidated faster than the business model. Nearly everyone now agrees on what survives liquefaction: evidence someone gathered in person, judgment someone stands behind, relationships and habits that attach to people and institutions rather than to formats. What no one has yet demonstrated is how those things pay at the scale the display era paid, and the industry is meanwhile pursuing every point on the spectrum at once: harvesting yield, building apps, selling content to the model makers, negotiating for visibility inside the answers, and reorganizing around distinctiveness. That simultaneity is not confusion so much as an honest reflection of where things stand. The format is melting faster than the answers are setting.