User needs in the age of agents
By Ezra Eeman and Dmitry Shishkin
The needs behind the habit (Ezra)
We spend a lot of time talking about what AI will do to journalism. How it changes the way we produce content, how that content reaches people, and what happens to the business model along the way. In those conversations, we seem to spend considerably less time on how people are actually using it. What do they find useful enough to come back to?
Access and convenience are part of the appeal. A conversation is an easy place to start, especially when an assistant is close at hand on the devices we already use. But we should also ask whether something deeper is happening. Is AI meeting needs that we have only partly addressed, or making it easier for people to express what they need in the first place?
Publishers working with user needs models have already moved beyond asking which topics to cover. They also ask what a story should do for someone: update them, give them perspective, help them understand or take action. Conversation brings time and interaction into that relationship more directly. The need can change as the exchange unfolds, and the response can change with it.
Imagine someone following a transport dispute. A few days before a strike, they want to understand why it is happening and whether the negotiations are making progress. The evening before, their question becomes personal: will they get to work tomorrow? By morning, they may need to compare routes and decide when to leave. Once they have a plan, they might want to hear only if something changes.
The subject is the same, but the person’s need has moved. A useful response depends on where they are in that sequence, what they have already understood and what matters next. User needs have always depended on context. An assistant offers the possibility of responding to that context through an ongoing exchange.
OpenAI’s latest Signals data offers a useful clue. In September 2026, roughly two-thirds of messages on individual ChatGPT plans were classified as non-work. Within that use, practical guidance and information-seeking together accounted for over half of messages. These figures do not tell us why people keep returning. They do suggest that we should look closely at the help people seek and at how journalism might become part of those encounters.
In the transport example, journalism could inform several stages of the conversation without the person ever asking for news or visiting a publisher. That makes the next question an editorial one: what happens to our journalism as it moves through that experience?
Preparing journalism for agents (Dmitry)
The four axes of human user needs (Know, Understand, Feel, Do) will be with us for some time. We need to look at what agents do with our content through that lens. There is a difference between making journalism visible to an AI system and making sure it is carried through intact.
Imagine you have spent weeks on an original, insightful “Give me perspective” piece, only to find its most important elements missing from an AI summary. The question is about our own due diligence: Did we do everything we could to protect what we created? What can we do so that it is not ignored, flattened or misrepresented?
A human may need to understand why something matters, know what has changed, or feel connected to a story. The agent acting for that person needs something different: retrievable facts, clear provenance, timestamps, source signals, evidence strength and update status. Agents also serve human needs, but in a very different way. They do not experience content; they process it instrumentally, retrieving, compressing, comparing and generating. Along the way, attribution, uncertainty, chronology and editorial intent can be weakened or lost altogether.
This is my thinking behind the Agent Readiness Layer, which maps each human user need against what an agent requires to serve it faithfully. It sets out eight obligations for journalism when an agent is the reader.
Can an agent find and select you (“Retrieve me”), and check that you are credible (“Verify me”)? Does your meaning survive summary and translation (Transform me), and does attribution travel with the content (“Credit me”)?
Does the piece help the user do something (“Act for me”), and can it be set alongside other sources (“Compare me”)? Can facts and quotes be used on their own without distorting their meaning (“Unbundle me”), and can the agent tell when they apply and what came before (“Situate me”)?
Go back to the “Give me perspective” piece. Its value sits in the argument, the weighing of views and the caveat in the fifth paragraph. Those are precisely the elements a summary can lose. “Compare me,” “Situate me” and “Transform me” give us a way to examine that risk. Generative engine optimisation (GEO) asks how to get cited; the Agent Readiness Layer asks whether what gets cited is still your journalism.
The framework is a prompt for structured thinking, not a validated instrument. It gives editors a way into a discussion that too often centres on technical optimisation. How much of your journalism would pass all eight tests today?
An experience that moves with the reader (Ezra)
Return to the person trying to get to work. If our reporting helps them understand the dispute but leaves them to work out every practical implication, there is a gap between the information we provide and the help they need. That gap is also a place where publishers can build.
On our own sites and apps, an article could become the starting point for an exchange. A reader could ask which parts of a development apply to them, examine the evidence behind an answer and choose what to follow. A local news service could let them decide which changes merit an alert. These are experiences to design and test, drawing on the reporting and editorial judgement that make the service worth trusting.
Some of this interaction may happen through a personal agent instead. That makes the readiness of our journalism and the usefulness of our own products closely connected. In both settings, the challenge is to serve a need as it develops, while keeping the evidence and limits of what we know visible.
The further we move into delegation and anticipation, the more judgement we hand over. Choosing sources is one decision. Remembering someone’s circumstances, watching for a change and deciding when to interrupt are others. Acting on their behalf goes further still. Each step needs a clear understanding of what the person has asked for and where their approval is required.
There will also be moments when people want to spend time with a story, follow a reporter or encounter a perspective they would never have thought to ask for. Designing around user needs means making room for those experiences too. Convenience alone is a narrow brief for journalism.
The opportunity is to look more closely at what happens before and after someone reads our work.
- What brought them here?
- What do they need next?
- What might they want to return to tomorrow?
We can use those questions to build a relationship that continues as their situation changes.
We explore these questions and the Agent Readiness Layer in more depth in our talk in WAN-IFRA’s Knowledge Hub.
Ezra Eeman is Director AI & Innovation at WAN-IFRA. He previously served as the Strategy and Innovation Director at NPO, the Dutch Public Broadcaster. Ezra writes a weekly industry newsletter, Wayfinder, about media innovation and is the author of the annual “Wayfinder Trends Report.”
Dmitry Shishkin is an independent media advisor who works with senior leadership teams building audience-centered organisations without losing editorial ambition or commercial discipline. Since 2020, he has advised global media organisations including Condé Nast, Thomson Reuters Foundation, Yahoo, Globo, Ringier and others.
