Field Note 02: Why Humans Must Own the Narrative
Summary: The information hierarchy that has governed knowledge work since the 1980s — Data, Information, Knowledge, Wisdom — has a blind spot. It has no place for narrative. This issue adds the missing layer, draws the line AI must not cross, and explains why the boundary is about accountability, not capability.
In every conversation about how AI will affect a newsroom and journalism, the same questions come up sooner or later: “Will AI now create journalism? Will AI take over the jobs of journalists? Will audiences even be able to tell the difference?” What follows is a debate about the exact nature of slop that AI tools produce, from good to bad to downright ugly. The better question — the urgent and necessary question — is: “What are the uniquely human responsibilities in an AI-enabled newsroom?”
The hierarchy with a blind spot
For the answer, we need to turn to information science, which has used a four-layer hierarchy since the 1980s, codified by Russell Ackoff, and which fits most knowledge work. There is Data, Information, Knowledge, Wisdom.
- Data — raw facts, numbers, transcripts.
- Information — data organised into context.
- Knowledge — information integrated into explanatory structures.
- Wisdom — judgment good enough to act on.
The model fits most knowledge work, but it does not fit journalism. Narrative is missing.
Narrative is more than facts or numbers. It is more sophisticated than contextualised information. But it is not yet wisdom. Narrative is deliberately structured knowledge — verified facts shaped through editorial judgment into a coherent account of a subject that helps a reader understand what happened, why it matters, and how the pieces and players connect. Sequencing. Causality, framing, and intentional emphasis.
Narrative is not a by-product of reporting. It is reporting. Narrative is the foundation of storytelling, and storytelling is one of the oldest human ways to bring people together, to capture attention, make people think, laugh, cry, or wonder. Insert that layer where it belongs and the hierarchy reads: Data, Information, Knowledge, Narrative, Wisdom. DIKNW.
The uniquely human responsibility sits at the narrative layer of this sequence. The reason it sits there has to do with accountability — which is the part of this conversation AI most threatens to dissolve.
Frequency is not reasoning
Journalists who frame a story can be asked why and be held to account. Why this angle, but not that? Why is the headline cast as a provocation rather than a question when the circumstances of the story are unclear? Why does one voice carry the third paragraph and another the seventh? Journalists can answer. Their reasoning can be examined, values debated, and judgment overturned. If they are wrong, they can be challenged on the grounds they used. There is a chain of accountability journalism runs on.
An AI system can be asked questions. But the chain of reasoning is less clear. An AI tool does not have ethical grounds, it has patterns. Asked why it framed a story the way it did, the most accurate answer it can give is: because text shaped this way appeared in the material I was trained on. That is not a reason. It is a frequency. And the system that produced it is, by acknowledged design, a black box. Even its makers cannot fully explain why a particular output emerged from a particular input. The opacity is not a bug to be patched. It is the technology.
Confine the black box
AI’s black box is not going to open; at least, for the time being. The response must be to confine it to the layers of work where its opacity is manageable, and to keep it out of the layer where opacity destroys what journalism is. AI can sit at data, information, and the assistive end of knowledge — places where its outputs can be checked against verifiable sources and corrected if wrong. It must not sit at narrative, where decisions must be accountable. They must be human.
Everything in the architecture I am developing rests on this distinction. The structured knowledge base. A two-layer separation between knowledge creation and AI-assisted distribution. An editorial firewall between them. Defences against the slow drift toward content that extracts attention but is not additive to audience value (otherwise known as enshittification). The next issues will work through the architecture piece by piece. The hierarchy is the map. What follows is the territory.
FROM THE FIELDS – THREE THINGS WORTH READING
Google has announced AI Overviews and AI Mode will include “Further Exploration” links
The advent of Google’s AI Overviews is estimated to have cost 58% of publisher clicks. But now Google has announced AI Overviews and AI Mode will be updated to include a “Further Exploration” links section, subscription labels, and inline link context. This comes amid an antitrust lawsuit from Penske Media, a formal European Publishers Council complaint to the European Commission, and a third of publishers saying they will block AI Overviews once the tools exist, according to one survey. Ironically, Google’s “traditional” search advertising business generated more than $50 billion in Q1 2026 alone — and that revenue depends on the continued existence of the web content now being overtaken by its AI summaries.
Nine per cent of US newspaper articles show signs of AI generation. But almost none of them say so.
A University of Maryland study audited 186,000 articles from 1,500 American newspapers published between June and September 2025 using Pangram, a high-precision AI detector. Roughly 9% were flagged as either fully or partially AI-generated. The rate was far higher at smaller local papers (9.3% for outlets under 100K circulation) than at national titles (1.7%). Of 200 AI-flagged articles manually reviewed, 96.5% had no disclosure of AI involvement. The factuality problem compounds the transparency one: 41% of articles labelled as AI-generated contained hallucinated claims — fabricated quotes, wrong statistics, misdated events — compared with 5% of human-written pieces. At the New York Times, Washington Post, and Wall Street Journal, opinion articles were 6.4 times more likely than news articles to be flagged for AI use, with guest contributors including Nobel laureates and US senators among those whose submissions showed AI markers.
Arya: a chatbot that delivers news, and the accuracy trade-off it exposed.
Researchers at Universitas Multimedia Nusantara in Indonesia built Arya, a chatbot that uses retrieval-augmented generation to deliver verified local news from outlets including Jawa Pos and Suara Surabaya Radio in conversational form. Users responded positively: 67% rated Arya highly for helping them understand stories. But media professionals involved in the study insisted on source disclosure, verifiable links, and honest answers when data was unavailable — not vague or silent responses. The core tension: humanising AI increases engagement, but must not come at the cost of accuracy. What makes this worth reading is not the chatbot itself — it is a prototype — but what the journalists asked for when they saw it. Source transparency, disclosure and the ability to update and correct.
At the time of writing, the university’s website was down. I am happy to send the academic paper on request.
I was at the INMA World Congress in Berlin last week. Trust came up in many of the sessions I attended — as something speakers were trying to measure, design for, and defend in specific operational terms. In the Field Note, I’ll share what I took away from Berlin and lay out the framework I use to think about trust, authenticity, and personalisation as three forces in tension — and what that means for AI deployment in a newsroom.
Until next time,
Dietmar
The Field Note: Twice a month, for editors, publishers, and heads of product working out what AI means for their newsrooms. One piece of original analysis. Two things worth a senior reader’s time.