Generative AI has swiftly moved from an intriguing concept to a pivotal tool across industries, and the media sector is no exception. 

During the recent Generative AI Town Hall, presented by the INMA Generative AI Initiative, Liz Lohn, director of product at The Financial Times in the United Kingdom, shared insightful details on the implementation and impact of generative AI in their operations.

Early AI engagement

Lohn reflected on the initial skepticism and excitement surrounding AI at the beginning of the year: “It would be a very different answer in January this year than it is now,” she said, highlighting the journey from hype to practical application and understanding.  

The Financial Times’ early engagement with AI allowed the news company to navigate the transition from seeing AI as a potential threat or bubble to recognising its transformative potential in media.

A critical aspect of the Financial Times’ approach to Generative AI is maintaining the trust and integrity of their content. Lohn stressed AI-generated content would not replace human judgment or compromise editorial standards.

“We would never risk the trust of our readers and use AI for creating any reader-facing content until and unless we are convinced of quality after sufficient testing,” she asserted.

This careful, measured approach reflects a broader media industry concern about how best to integrate AI without diluting journalistic quality or ethics.

Lohn shared the iterative and experimental nature of integrating new technologies into a complex organisation like Financial Times. Early in the year, her team embarked on a project to use Generative AI to automate the creation of newsletters — starting with summarising articles for just one newsletter type. The team thought this would be a simple and manageable use case to explore the potential of AI in journalism.

However, what seemed like a straightforward task turned into a significant challenge, Lohn said: “It took us about a month and a half to work through the initial prompt.”

Fine-tuning AI content

The difficulty lay in fine-tuning the AI’s output to ensure it captured the right balance between being informative and engaging without losing the publication’s voice or editorial integrity. Additionally, during this period, major language models were being updated, which meant the models the team had assessed quickly became outdated.

Despite the considerable effort, the initial experiment did not yield the desired results. 

“We didn’t get the effect we wanted,” Lohn said. Rather than abandoning the project, the team viewed this as a learning opportunity. “It was the perfect kind of failure for us,” she said, emphasising how the experience provided real data and insights that helped reshape their approach. 

Instead of sticking to the original plan, they pivoted toward a more flexible approach — one that encouraged more experimentation and direct involvement from the newsroom staff.

Rolling out the AI Playground

From the lessons learned, Financial Times launched an innovative new tool called the AI Playground — a dynamic, interactive platform that allows newsroom staff to experiment with AI-generated content in a way that is flexible, iterative, and highly customisable. 

Instead of being a rigid tool that dictates what journalists can and cannot do, the AI Playground empowers them to drive the experimentation themselves, making it a far more effective solution than the initial newsletter project. 

The team realised that by getting closer to the actual content creators — the journalists themselves — they could accelerate the learning curve and continuously improve how AI is applied.

The core idea behind the AI Playground is to enable users to create and refine their own AI-generated content through a simple, intuitive interface. 

“Users can search the FT content using natural language and pre-select specific articles … then they build prompts, generate responses, and rate them,” Lohn explained.

This system allows journalists to develop their own use cases by experimenting with different ways to generate content, whether it’s for internal research, content summaries, or even inspiration for new story ideas.

AI Playground + the newsroom

The AI playground offers a range of features that support different levels of interaction with AI. Journalists can start by using pre-made templates for generating content, and as they become more familiar with how the AI responds, they can refine the prompts to produce more tailored outputs.

“Now dozens of users are using it and experimenting with various prompts,” said Lohn, reflecting on how the tool quickly became integrated into the newsroom’s daily workflow.

One of the most compelling aspects of the AI Playground is its ability to foster a culture of experimentation within the newsroom. Lohn pointed out that unlike the original approach — which focused on a small team of editors and a lengthy, trial-and-error process — the AI Playground allows for quicker iterations, empowering journalists to test their own hypotheses in real time. 

This shift in approach — from a top-down model to one that is more collaborative and user-driven — has proven highly effective.

The process is also designed to be highly iterative, where ideas are quickly tested, evaluated, and refined based on direct journalist feedback. 

“We needed to accelerate the feedback loop between ideas from outward testing and then maybe switching to a different idea,” Lohn explained. This iterative model is crucial for refining AI models to fit the unique needs of Financial Times, where the integrity of the content and its alignment with editorial standards is paramount.

Moreover, the platform encourages collaboration among journalists with the ability to share prompts and responses across the newsroom. Lohn emphasised the AI Playground fosters a communal approach to problem-solving, with staff members exchanging ideas and strategies for getting the best results from AI. This collaborative spirit makes the tool more than just a product — it becomes a platform for innovation within the organisation.

While the AI Playground started as a tool to help journalists experiment with content generation, it has since become a more powerful resource for data-driven journalism as well.

For example, one of the first use cases involved analyzing complex data sets such as the UK Parliament’s register of MPs’ interests. By applying machine learning and AI to this raw, often unstructured data, the Financial Times’ team was able to surface stories that would have been nearly impossible to identify manually.

These AI-driven insights not only help journalists discover compelling stories but also democratise the ability to work with data. Lohn noted some journalists at Financial Times were already adept at data-driven reporting, but many others were not. The AI Playground allows specialists in non-technical roles to engage with complex data sets, thus broadening the scope of data-driven journalism within the newsroom.

Additionally, the AI Playground has proven highly scalable, as the same algorithmic approach used for the Parliament dataset can be applied to other data sources, such as the Federal Election Commission in the United States. 

Lohn is optimistic about the future applications of the AI Playground, particularly as it evolves to incorporate more sophisticated AI models, ensuring the newsroom remains at the cutting edge of innovation.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .