The information ecosystem is being redrawn by AI. That might be good news

On a flight from London to Boston last year, I was sat next to an orthodox priest who spent most of the journey talking to Grok. I have no idea what the conversation was about. But whatever it was, it held him completely.

There is no punchline coming.

I spent much of that flight wondering what needs of his were being served. Informational? Emotional? Confessional? And then, what signals was Grok getting back from him? What an orthodox priest flying to Boston actually wants, worries about, is trying to work through? 

If you work in news, that second question should unsettle you more than the first.

Four shifts and a funeral

The dominant conversation around AI and journalism is loud, fast-paced, and largely tactical: which tools to adopt, whether to license content or litigate; how to optimise for AI search; whether to block AI crawlers or court them. 

These are real questions which need answering. But they are about optimising for the current ecosystem, and the risk is that they crowd out the larger picture.

The larger picture is not journalism-specific. The same dynamics are reshaping any sector where the core product is knowledge, but journalism is among the most exposed. I see four shifts redrawing the economics of information. They cascade, each creating the conditions for the next, and together they demand not optimisation but redesign. 

1. From scarcity to abundance

Information used to be expensive and difficult to produce. It no longer is; and the proliferation of LLM-assisted writing across every sector is now well documented.

When the marginal cost of production approaches zero, the ability to publish ceases to be a competitive advantage. Publishers using AI to produce more efficiently may find themselves accelerating the very commoditisation that threatens them.

2. From human audience to machine audience

Abundance almost inevitably creates the conditions for this second shift. Decades of rising information oversupply have made it impossible for any human to navigate what’s available. Machines become necessary intermediaries. Publishers once served audiences directly. Then platforms mediated that relationship. Now AI systems are inserting themselves as active intermediaries in what I have called B2A2C.

It is tempting to see this as another platform shift: Google, then Facebook, now ChatGPT. But the mechanism is fundamentally different. 

A platform ranked and distributed content. An AI intermediary parses it, interprets it, decides what it means, and recomposes it into an answer the user never needs to leave. It might take one or two sentences from a 2,000-word essay, or nothing at all, just the meaning. It has, in a functional sense, editorial agency. 

The Reuters Institute’s Trends and Predictions report for 2026 puts numbers to the shift: Google search traffic to publishers fell by a third globally in the year to November 2025, and news executives now expect a further 43% decline over the next three years.

3. From artefacts to liquid information

Once machines are the primary audience, the artefacts we have built around information (the article, the bulletin, the documentary) become structurally optional. 

Fragmentation was the old problem. Now information can exist without a container at all. And when information becomes liquid, the market bifurcates. Economic value concentrates at two extremes of a ‘barbell’; premium brands competing on differentiation and direct audience relationships; and commodity infrastructure operating at massive scale and razor thin margins. The middle, where most publishers currently sit, hollows out.

4. From attention to intention

When content is liquid and an intermediary controls the interface, something more consequential than format changes. The currency does. Clicks, pageviews, time on site; these were always crude proxies for what we actually wanted to know: what do people need? Now there are systems that can infer the answer directly. I have written elsewhere about this shift from attention to intention as the defining transition of this era. 

Here’s the short version: a query to an AI chatbot isn’t just a request for data. The system infers what isn’t explicitly stated and responds with something closer to counsel than retrieval. The signs of this shift are early but visible: AI assistants are growing fastest not as search replacements but as conversational tools, with users returning for longer, more personal sessions than they would ever conduct through a search bar.

…And the funeral 

Taken together, these shifts describe a funeral for the economic model that has sustained most journalism for decades. Production is no longer a moat. The audience is increasingly someone else’s customer. Content dissolves into raw material. And the richest signal about what people actually need, the intention signal, is captured by the intermediary, not the publisher.

A bull case for journalism

Everything I have just described probably sounds horrifying for most media leaders, who might read this as an extinction-level threat.

I think they are wrong; or at least, only half right. The current model is dying. But I believe in the new model, the market for what journalism does will not shrink. It might expand, in terms of both who needs it and what it can be.

A larger market

Every information revolution in history has expanded who could participate: printing reached the literate, radio the illiterate, mobile connected five billion people. AI could be the next jump, but its distinctive contribution is not access. It is comprehension. Millions of people can technically reach a public health study or a municipal budget today; very few can meaningfully engage with either. AI intermediaries can bridge that gap and more, translating, contextualising, adapting information to what a specific person actually needs to know. If mobile brought five billion people into the information ecosystem, AI could make that ecosystem legible to all of them. At the same time, AI makes previously uneconomic coverage viable: beats too expensive to staff, languages too niche to serve, questions too latent for anyone to commission.

The scale is already visible. ChatGPT alone has 800 million weekly active users, with seeking information and advice as its primary activity. OpenAI's own research, published as an NBER working paper, found that the largest single category of use  (nearly a third of all conversations) was ‘practical guidance’, including personalised tutoring, customised advice, creative problem-solving. 

Demand for trustworthy information has always existed but that no publisher was structured to meet it. Some of that is a substitute for traditional news consumption – and history indeed suggests that incumbents rarely capture the growth when an information revolution reshapes their market. But much of it, a priest's seven-hour conversation, a first-generation student navigating financial aid, a farmer parsing climate projections, is entirely new. The addressable market for trustworthy information is almost certainly not contracting. It is possibly larger than it has ever been.

New products

The question, then, is what journalism builds for a market this size. When journalistic artefacts become commoditised, the value shifts upstream, to the processes that produce them. Detection, verification, sense-making, anticipation: these can be disaggregated from the article and delivered as products in their own right.

Consider iTromsø, a Norwegian newspaper in the Arctic Circle with an editorial staff of 25, which is nudging at what this might look like. Their recent Airbnb project used econometric analysis to ask whether short-term rentals drive housing prices, and published the uncertainty alongside the findings. Confidence intervals, model assumptions, and the limits of what the data could prove are what their Head of AI, Lars Adrian Giske, and I have discussed as “probabilistic news”: journalism that leads with what the evidence can support, and is honest about where it stops.

That analysis was designed for iTromsø's readers. But push it to the absolute limit of scale. What if the same methodology ran continuously everywhere the question matters? Not as a one-off investigation but as a living analytical layer: a policymaker in Lisbon, a renter in Barcelona, an AI assistant answering “should I buy in this neighbourhood?”, all drawing on the same verified, uncertainty-aware analysis. 

That could be journalism as informational infrastructure, deployed at scale to support decision-making wherever it happens, in whatever form users require. It could not have existed before AI, an AI intermediary cannot easily build it alone, and it serves the expanded market rather than retreating from it.

This specific product may never exist. But the discipline of asking what any innovation looks like at its most radical scale, and whether the answer serves the larger market or retreats from it, is probably a more fitting test for the moment than most newsrooms are currently applying. Journalism has the expertise, the access, and the verification methods to build for this market, but only if it is willing to deliver in forms and through channels that look nothing like a newspaper or a website.

What’s next?

This bull case describes a market opportunity, not an inevitability. The difficulty is that the journalistic experiments most likely to capture value in the new market are precisely the ones that feel least compatible with the current business. That tension between protecting what exists and building what might replace it is not something a commentary can resolve.

I believe the bull case is there. But reaching it probably means letting go of some familiar things, formats, assumptions, perhaps institutions, and holding on tightly to the functions that matter: verification, accountability, helping people understand the world. 

The containers can change. The mission doesn’t have to.

The priest on my flight was most likely not just searching for information. He was in a seven-hour conversation with a system that met him exactly where he was, wherever that was. And it was Grok doing the meeting, not a newsroom.