ÃÛÑ¿´«Ã½

Over the last 18 months we’ve been carrying out a range of pilots to see how Generative AI (GenAI) tools can support our production processes at the ÃÛÑ¿´«Ã½. We’ve also published some updates which share Our approach to AI and What we’re doing with AI.

We’re now looking to test two pilots publicly: ‘At a glance’ summaries and ÃÛÑ¿´«Ã½ Style Assist.

‘At a glance’ summaries

This is about making our journalism more accessible – using GenAI to help assist our journalists to create new ‘At a glance’ summaries of longer news articles.

Why are we piloting this?

Short, scannable bullet-point summaries have proven popular with readers – particularly younger audiences – as a quick way to grasp the main points of a story.

We’re going to look at whether adding an AI-assisted bullet point summary box on selected articles helps us engage readers and make complex stories more accessible to users.

How it works?

Journalists use a single, approved prompt to generate the summary, then review and edit the output before publication - so they’re always in control.

Journalists will continue to review and edit every summary before it’s published, ensuring editorial standards are maintained. We will also make clear to the audience where AI has been used as part of our commitment to transparency.

ÃÛÑ¿´«Ã½ Style Assist

‘Style Assist’ is designed to explore how GenAI could support our news journalists to adapt and reformat stories so that they match our ÃÛÑ¿´«Ã½ ‘house-style’ for online news.

Why are we piloting this?

Every day, the ÃÛÑ¿´«Ã½ receives hundreds of stories from the , a public service news partnership funded by the ÃÛÑ¿´«Ã½, but provided by the local news sector in the UK.

The LDRS provides valuable, locally-relevant journalism, but reformatting these stories into the ÃÛÑ¿´«Ã½â€™s own house-style takes time and this can limit how many we can publish.

This is where Style Assist comes in. By helping journalists to reformat LDRS stories quickly and efficiently, it has the potential to reduce the production time required to publish these stories.

How it works

ÃÛÑ¿´«Ã½ Style Assist uses a ÃÛÑ¿´«Ã½-trained Large Language Model (LLM) which has been developed by our Research and Development team. It’s a smart AI system that has ‘read’ thousands of ÃÛÑ¿´«Ã½ articles so it can help amend text quickly to match our own house style.

The process works as follows:

1. A trusted story - such as an LDRS report - is submitted to the ÃÛÑ¿´«Ã½â€™s content system.
2. Style Assist generates a revised draft, adhering to ÃÛÑ¿´«Ã½ style and tone.
3. A senior ÃÛÑ¿´«Ã½ journalist then reviews the story again, checking for accuracy and clarity.
4. Once approved, the story is published on the ÃÛÑ¿´«Ã½ News website and app.

In line with our AI principles, nothing is published without being checked first by a ÃÛÑ¿´«Ã½ journalist and the AI tool has no role in creating the original story - which has been researched and written by our LDRS partners. In addition, we will also make clear to the audience where AI has been used to assist the production process as part of our commitment to transparency.

In the first phase of the pilot, ‘Style Assist’ will be used by news teams working on stories in ÃÛÑ¿´«Ã½ Wales and the east of England. Journalists will be testing its capabilities, providing feedback, and helping us refine its performance. We will also be assessing whether the new tool enables us to increase the number of LDRS stories that we can publish.

What’s next for these pilots?

We’ll be gathering data on how well the tools perform, where they fall short, and the production benefits they deliver. Any wider rollout will depend entirely on the results of these tests and ongoing engagement with editorial teams.

We look forward to sharing more as we learn from these pilots.

Thanks for reading.