For the last thirty years, the newsroom main product has been the same: a website. Readers came to us, we published, they browsed. Every growth strategy (SEO, subscriptions etc), assumed that we have our readers on our website.
That assumption is now the weakest part of the model, I think. Readers (we all, actually) are moving into new interfaces, and a website that waits to be visited is no longer the whole product (we've heard this already when social media appeared). What I believe follows are the three shifts that matter most, and the single idea underneath all three: an archive should work on demand on other platforms as well, not as a finished thing sitting on a shelf waiting to be browsed page by page on the website. Not a popular opinion, but...
1. Chat Is Already a Distribution Channel
Readers are not experimenting with chatbots. They are building a habit around them, and that habit is only going to deepen. Most of the news media's response to this is defensive: optimize content so an AI engine summarizes it correctly, rank higher in AI search, hope for a citation somewhere in an answer (AEO and GEO). For sure those efforts are necessary. But they treat the chat layer as someone else's territory that we are trying to get noticed in.
And there is a second option: why not to build directly into that layer instead of trying to be found in it. If a reader can connect a newsroom's own data source to the assistant they already use, the newsroom stops competing for a mention and becomes part of how the assistant answers. That is a different kind of distribution than anything AEO or GEO can offer, because it does not depend on being indexed, ranked, or cited. It depends on being present actually.
2. MCP
The protocol that makes this possible is MCP (Model Context Protocol). It connects a server holding an organization's data directly into an AI model's context, so the model does not have to guess what a newsroom covers or wait for a search index to catch up. A reader connects it once. After that, their assistant can pull from the archive on demand.
I built this MCP for "AI For Newsroom". It took three days. Watching it work told me more about distribution than any optimization checklist has. When Claude (I've tested it with Claude free plan) had a choice between its own general knowledge and our server, it checked the server first, consistently. Good news, right? Newsroom does not have to wait to be indexed. It can simply be present.
A few consequences follow once we take that seriously. Links served this way can have UTM, so referral traffic from a chat conversation becomes measurable. A simple admin layer shows which topics, sections, or authors people actually ask about, which is a sharper signal than pageviews because it reflects a question rather than a click (we can finally have a clear picture of what our readers need). And because the newsroom controls what context gets served, there is a legitimate path to sponsorship: not an ad slot bolted onto a page, but structured information a sponsor pays to have included in the MCP, as part of what the assistant reads before it answers. But we need to inject this sponsor's information properly (for example as a recommendation, not as 'here is our sponsor and its services'). I'm going to publish some kind of a blueprint based on the testing different ideas with MCP. Hope it can be useful.
3. Customization Means Assembling the Page, Not Just the Feed
Search on a news site has always meant the same thing: a list of articles, ranked by relevance or date, that the reader has to assemble into an answer themselves. This approach now is not the best thing an industry can offer. Now we have a system (LLMs actually) that can retrieve across an entire archive, spanning verticals, formats, older reporting, data, and profiles, and compose the answer directly.
A reader searching a topic does not need ten headlines. He needs the version of the story built for exactly what he asked, with its own URL, something he can bookmark or send to someone. It looks like an ordinary webpage. The difference is that it did not exist as a fixed asset before the query. It was built from the archive the moment someone needed it, and this is not an expensive thing to do, by the way. Retrieval and composition tasks like this do not require the most capable model available, which means the economics work even for a small publisher. Use cheaper open source models, for example.
I did some calculations —creating a 1000 of this custom pages based on "AI For Newsroom" data will cost me up to just 10 euros per month. But engagement is the key! This is personalization built around a moment rather than a profile. The same reader might want the World Cup this week, a local festival next week, and new legislation after that. A system built around long-term consumption habits struggles with that kind of person. A system built around the immediate question does not. I'm going to launch this type of customization later this week as well to test how it works.
One Philosophy, Three Surfaces
Put these three concept side by side and the whole pattern stays the same. Journalism travels to wherever the reader already is. Journalism reassembles itself around whatever the reader is actually asking. The connection itself becomes the channel, rather than a page the reader has to find first. None of the three treats the archive as a finished product, none of them can harm journalism. All three treat the archive as a bricks a system can build from, meeting a specific need, at the moment the need appears.
That is the actual shift worth planning around: not just "add a chatbot," "improve our AI search score," and "personalize the homepage" as three separate tactics. The website stops to be just the only container journalism lives in and becomes one of several surfaces reader can look through. What a newsroom owns is the underlying set of elements and the system that can retrieve and reshape it. A chat connection, a generated page, whatever comes after those, are simply the interfaces available today.
Journalism does not need AI to replace the newsroom. It needs the newsroom to decide, on purpose, how its journalism travels once the reader is no longer the one who has to come looking.