AI knows nothing but it is still an extremely powerful tool for newsrooms

Last December, IMS hosted a meeting in VOX House, Copenhagen, where a group of media came together to discuss how AI is affecting journalistic and editorial work.

The meeting pointed to a clear gap. AI was already entering newsrooms, but often in fragmented ways: as transcription, translation, summaries, analytics, monitoring, social media adaptation or individual experimentation. What was missing was a structured way to test how AI could support editorial judgement inside the newsroom workflow itself.

In response, IMS developed the AI Newsroom Pilot: a model that connects AI-supported analysis to the knowledge already present in a newsroom – its professional standards, archive, audience understanding and experience.

This autumn, 10 independent media outlets from Armenia, Moldova and Ukraine will test the model in live newsroom workflows. In practice, the model is brought in while a piece of journalism is still open to change. It reads the draft through the newsroom’s own standards, archive, audience knowledge and editorial priorities, and returns a few focused reflection points for the journalist and editor before publication.

Before the end of this year, the newsrooms will meet again in Copenhagen to compare what the testing has shown: where the workflow helped, where it created friction, what had to be adapted and whether the approach can be developed further.

The question running through the pilot is simple and urgent in fragmented information environments: how can newsrooms make better use of what they already know to become more relevant to the audiences they serve?

Ten newsrooms, one shared challenge

The 10 participating newsrooms enter the pilot from different positions. That is the point. The shared challenge is the same, but it appears differently depending on workflow, audience, language, format and local context.

Each newsroom already carries a large body of knowledge. It is built through reporting, editing, corrections, audience contact, distribution choices, mistakes, successes and daily editorial judgement. Some of it is written down in policies, routines and style guides. Some of it sits in archives and analytics. Some of it lives with the journalists and editors who know the beat, the audience and the local context.

The problem is that this knowledge is often scattered. Under deadline pressure, it is not always available at the moment where it could make a difference: while a piece of journalism is still being shaped.

That is the moment the pilot focuses on. The question is not whether a text can be improved in general. The question is what this newsroom already knows – about the topic, the audience, the context, the risks, the format and the public value – that should inform the editorial conversation now.

For NikVesti in Mykolaiv, Ukraine, this means testing how years of local reporting and an existing internal knowledge layer can be brought into the daily editorial workflow. For Ampop Media in Armenia, working with explanatory and data-driven journalism, it means connecting verified facts with context, evidence and audience questions. For TV Nord in northern Moldova, a regional bilingual newsroom working across platforms, it means keeping language, format, local relevance and editorial clarity in view at the same time.

The pilot does not look for one universal model for newsroom work. Ten newsrooms will test the same basic idea from 10 different editorial realities. The aim is to learn whether newsroom-owned knowledge can be organised, activated and brought into play when editorial judgement is being exercised.

Live testing across newsrooms

The pilot now moves into live newsroom testing, using real drafts in real workflows.

The starting point is simple. AI does not know anything in itself. It does not contain a complete picture of human knowledge, literature, society or journalism. It has processed enormous amounts of data and become a remarkably powerful statistical machine.

That distinction matters. The pilot does not use AI as a source of knowledge. It does not ask AI to produce journalism. The participating newsrooms already verify sources, facts and data. That responsibility remains with journalists and editors.

What the pilot adds is a structured reflection layer before publication. It reads the draft through different layers of knowledge: the newsroom’s own standards, archive, audience understanding, priorities and professional practice.

The aim is to produce a small number of high-impact reflection points before publication: what may be missing, what may be unclear, what context should be activated and what the audience may need in order to understand why the piece matters.

The value of the pilot will come from live use and adaptation. Real drafts will be tested in real workflows. Some recommendations may help. Some may misunderstand context. Some may be too generic, too cautious or simply irrelevant. That is part of the test. The workflow must be shaped by the newsroom, not the other way around.

The real asset: editorial knowledge

The real asset in the pilot is not the technology itself. It is the editorial knowledge already present in the newsroom.

Some of that knowledge is formalised in editorial standards, internal policies, style guides and workflows. Some of it lives in archives, analytics, distribution experience and examples of stories that worked – or did not. Some of it is less visible, carried by editors and journalists as professional judgement.

The pilot tries to bring that knowledge into play in a more comprehensive and structured way. Each newsroom works with its own layers of knowledge, connecting its track record, audience insights, editorial standards and priorities to the review of new drafts.

This also matters for independence. Models, vendors and tools will change. The newsroom’s knowledge should remain owned, portable and useful across changing technologies – not locked into one system or supplier.

From testing to possible expansion

Before the end of the year, the participating newsrooms will meet again in Copenhagen to discuss what the live testing has shown – and what it may open for further development, adaptation and wider use.

By then, the live testing will give the newsrooms a more practical basis for discussing what added value to daily editorial work, what created friction, where context was difficult to capture and what had to be adapted from one newsroom to another.

It will also show what can travel. Some knowledge will remain local and newsroom-owned. Some methods may be shared across media. Some parts of the model may need to be changed, simplified or expanded. The coming months are therefore less a demonstration than a test. Ten newsrooms will take the same basic idea into 10 different editorial realities. What will be interesting is whether organised editorial knowledge can help them make stronger decisions before publication – and reach more people with journalism that matters.