The most interesting thing happening in publishing right now is not AI writing articles. It is what AI can now do with the content a publisher already owns. Ask it a question and it can answer from that material directly. Give it a task and it can carry it out. Put it to work inside the tools a user already relies on every day, and content stops being an archive someone has to visit and starts being something that acts on their behalf.
For a CTO, a chief data officer, a product director or a chief executive, that is the shift to watch. Not another chatbot bolted onto a homepage, but the point where your content becomes something a machine can use reliably, on terms you set. It moves what you are selling from access to outcome, and that is a bigger change than it first sounds. It changes what a subscription buys, what a licence is worth, and what a competitor without your archive can and cannot replicate.
Here is the catch, and it is the argument of this whole piece. What follows is real and available now, but only for the publishers who have already done the structural work underneath. Most have not. The foundation is where the real decisions sit, and it is a great deal less exciting than the layer built on top of it. This is not a call to be cautious about AI. It is a call to be deliberate about what you build first, so that the exciting layer has something solid to stand on when you do build it.
What good data now unlocks
Start with what changes once content is trusted, structured and reachable. It stops being a library a user visits and becomes something that can work for them. That single shift, from stored to usable, sits behind every application worth building.
Three stand out:
1. Answers, not just articles
The first is the move from returning documents to finishing a job. A search box hands back a list of links and leaves the work to the user. An assistant grounded in a publisher’s own archive can go further. It can assemble a cross-referenced briefing on a live question, watch a topic and alert someone the moment something material changes, or draft a sourced first answer that an expert then finishes and signs off. The user arrived with a task rather than a keyword, and the task is what gets done.
What makes this possible is not a cleverer interface. It is content a machine can trust and cite. The instant an answer cannot be traced to a reliable source, the whole thing falls apart, and it falls apart in the worst possible way, by sounding exactly as confident when it is wrong as when it is right. Get the sourcing right and you have a product that competes on the quality of the answer, not the size of the index. Get it wrong and you have handed your name to a machine that will use it to sound sure of things that are not true.
2. Your content where the user already works
The second is a change in where that activity happens. Increasingly, a user does not start at a publisher’s own search box. They start inside an AI assistant and expect it to reach whatever it needs on their behalf. Read one way, that looks like a threat: your content used without your say. Read another way, it is a new channel, and the second reading is the more useful one.
The mechanism has a name. The Model Context Protocol, usually shortened to MCP, is an open standard that lets an AI client reach a publisher’s content directly, carrying whatever rules the publisher attaches, so material can surface inside the tools people already use, credited and on stated terms. Adoption is moving quickly. In a survey of data and analytics providers published by Asymmetrix Intelligence in July 2026, 96 per cent of respondents said they had either launched an MCP or were considering launching one, with 56 per cent already live. What the protocol does matters more than what it is called: it moves the point of access from a website to wherever the user happens to be working. The deeper change is about who decides. The choice of what content gets used, and when, used to sit with a developer wiring up a product months ahead of time. It now sits with an AI at the moment a user asks a question. No publisher can force a model to credit its source, but design choices can make that far more likely, and being reachable on good terms beats being scraped on bad ones. Seen this way, the shift opens new products and new audiences rather than simply threatening old ones.
3. Free your experts for the work that matters
The third points inward rather than outward. The same capability that can answer a user’s question can take on the unglamorous load inside a publisher’s own operation: preparing and tagging content, running first pass checks, surfacing connections across an archive too large for any one person to hold in their head. Internal agents like this do not replace editors and analysts. They clear the ground in front of them, so that the expensive, differentiating work, the judgement and the authority a model cannot supply, is where people actually spend their time.
For most publishers, this is the fastest return of the three, because it improves the product and the margin together, and it happens behind the scenes, where mistakes can be learned from before anything reaches a user. It also builds a discipline the outward facing applications will need later, because the rule is the same in both directions: an agent is only ever as good as the structure of the content it is working from. Start here, and the harder, more visible applications above become considerably less risky when their turn comes.
This is not really a publishing problem. It is a data problem, and the same shift is landing in every B2B information sector, from financial data to legal research to scientific reference. The thinking here travels, and so does the advantage for whoever acts on it first.
The catch: none of it works without the foundation
None of the three use cases above survives a weak foundation, so it is worth saying plainly. AI does not lift a weak foundation. It exposes it. Point a capable model at content that is inconsistent, poorly structured or unreliable and it will not simply return a worse answer. It will produce fluent, confident nonsense, in exactly the same assured tone as a correct one. For a business whose entire value rests on trust, that is worse than returning nothing.
Which is why the useful reframe is to stop picturing applications served by data and start picturing data served by applications. The interface is not the deliverable. Trustworthy, structured, reachable content is the deliverable, and applications are what get built on top of it once it exists. A simple test cuts through most arguments about where to invest: if the content were guaranteed accurate and reachable but no new interface were built at all, would that still be valuable? If the answer is ‘yes’, and it almost always is, the foundation is the real product and everything else is presentation.
The trouble is, that foundation work is invisible. Nobody demonstrates a data model at a board meeting, so it is easy to underfund next to a striking front end. The way to counter that is to make the value visible early: build something small that people can react to, tied to a workflow a user actually has, rather than perfecting a warehouse nobody outside the team will ever see.
Picture a legal information publisher sitting on a century and a half of editorial judgement, locked inside unstructured court documents. The temptation is to fund a search interface or an assistant on top of that archive first, because that is the part a board can see and approve. The work that actually matters is underneath: turning that material into something structured, linked and queryable. Once that exists, every application described above becomes possible at more or less the same time, and often more cheaply than the single interface would have cost on its own.
This is where the strategic conversation across information, media and events appears to be converging: get the data foundations right, and understand how AI is reshaping your particular market, before committing to anything built on top of it. That sounds obvious stated plainly, and yet few organisations sequence their work in that order, choosing instead to fund the visible layer first and hope the foundation catches up behind it.
The takeaway
The gap this opens will not be gentle. Publishers who have done the structural work will deploy this applied layer at scale, launch faster, and absorb each new model and each new tool with far less friction. Publishers whose data is a tangle will find the same clever tools simply amplify the mess, at speed and in full view of their users.
The distance between those two groups is going to widen, and probably faster than most boards expect. So the message is not to panic about AI. It is to take part, and to take part in the right order. The applied layer is real, it is available now, and it is worth pursuing. It is paid for entirely with the unglamorous work underneath, and there is no shortcut that skips paying for it.
For a product or technology leader, the practical implication is one of sequence rather than ambition. Chase the applied layer by all means, but chase it in the knowledge that every use case above collapses back onto the same question. The question worth putting to your own team is not what could we build. It is plainer than that: if a capable AI relied on our content tomorrow, would it find something it could trust?
67 Bricks provides technology consultancy and full-service product development for specialist information publishers to enable them to grow the value of their businesses.
Web: www.67bricks.com
LinkedIn: linkedin.com/company/67-bricks-ltd
