Turn your AI experiments into a strategy

The call to move AI adoption from experiment to strategy is getting louder. To start with prioritisation we can start with the question of what we actually want to achieve. We can ask what creates and adds to the value that subscribers are willing to pay for. Things like quality, consistency, an editorial identity you want to get familiar with, relevance and an engaging presentation come to mind.

Which of these matter most depends on your brand, your audience and your strategy. That half of the equation only your newsroom can answer. What we can map out is the other half: where AI is genuinely good. Because not all tasks lend themselves equally to AI automation.

Which assistants do we build first?

A useful lens to start with is the set of questions every journalist learns: Who, What, When, Where, How, Why. And then the ones that make a story matter: So what? What does it mean?

Read from left to right, the questions show a rising gradient of judgement. Take proofreading: a surprising share of newsroom house rules concern names and places. The Who and Where of a story. An assistant can check those against a stylebook or attached sources to be genuinely helpful. The same goes for much of When and What: dates, figures, consistency with what the sources say. Not easy to solve by any means. But for these, we can define what good looks like, because we can define and probe correct results. As a tool, AI is only as good as our ability to define what we actually want.

Move right and the ground shifts. An AI will happily produce a plausible-sounding Why. And that is precisely the problem. The most plausible-sounding justification is often not the right one. The essence of news is to recognise when the patterns break. When the reader's expectations should change because a simple continuation from the past would lead us astray. Glossing over a messy, unresolved situation with polished language is easy to read and completely beside the point.

This is why boldly automating whole stories is a bad trade. Readers are not short of content; they are short of attention and trust. Producing more, faster and cheaper buys a currency that is losing value. And to produce whole stories, AI has to produce the Why. But getting the Why wrong in your own voice is how a brand loses trust it cannot buy back.

Focusing on the Who, What, When and Where therefore is the right place to start. AI might not have good judgement but it can search, apply simple rules, cross-reference, identify people and places and enforce newsroom standards. These activities increase quality, accelerate production and can elevate the daily work of journalists by supporting them in predictable ways.

Compounding efforts

Even more important than choosing the right tasks is figuring out how they complement and strengthen each other. A simple example is the consistent naming of a person in a proofreading step. A newsroom can define the preferred spelling and AI can be used to check consistency.

This might sound trivial but it already provides a foundation other skills can build upon. A fact-checking mechanism can only search all claims made about someone after we can reliably identify said person. And once we are able to identify claims, other skills can check for consistency, figure out which claims are verifiable with public data or ask the author about their sources to check against those. Every complex system is built out of many small pieces that add up to a whole that is more than the sum of its parts.

AI is strong when we know exactly what we want. Along the editorial workflow the possibilities are almost endless. Journalism is an especially diverse field where every department and format and sometimes even individual stories have different needs. AI is genuinely good as a research assistant, to quickly learn about a topic or to ensure consistency. It can prepare and clean data, dig through vast amounts of obscure information, repackage your formats, build charts and provide background information, to name just a few. This breadth is exactly why you need a strategy. No central AI team can build the all-knowing coworker that covers it all. Small assistants with clearly defined outcomes can be built in parallel, by the people closest to the work. Together they can cover the whole playing field.

By empowering your newsroom with clearly defined capabilities where high value and AI capabilities meet, you get a compass for what should be approached next and why. And assistants with a clear outcome where you can specify what good looks like are dependable enough that they can build on top of each other.

Smart workflow orchestration

Assistants can only build on each other if something connects them: carries one skill's verified output to the next, provides the right context and keeps track of what happened along the way. And when many people across the newsroom build assistants in parallel, someone has to govern what runs where.

This is the final piece of such a strategy: orchestrate all assistants along the editorial workflow in a governed platform. To achieve this an editorial system must allow integrations at every step. An editor needs to see what ran, what changed, what was merely proposed, who triggered it, and what still needs human review. This record is worth more than just internal safety. It is verification captured as data. Records of which claims were checked, against which sources, who signed off. This can provide a new layer of transparency that helps to earn trust. And the orchestration has to know which assistants are appropriate for which story and at which point in time.

This strategy deliberately leaves a gap: the Why, the judgement, the moment when the patterns break. This has to be filled by your journalists. The people behind journalism are becoming more essential; the product is a person. A workflow that takes over the checkable work without ever touching judgement is exactly what the most valuable writers will be looking for. Good people are essential. They are the answer to your hardest questions.

This is an AI automation strategy that keeps the reader's trust in the short term and one that will transform what is considered good journalism in the long run by raising the bar for transparency and verification. It keeps judgement in people's hands by automating where we can clearly define expectations. Each workflow step defined this way turns individual craft into an organisational standard that is teachable, inspectable and improvable. That may be the most lasting gain of all.