How RTL NEWS Is Shaping the Future of Newsrooms With AI-Powered Workflows

In a modern newsroom, thousands of stories flood in daily that require sorting, verification and publication. And even for the most experienced editorial teams, balancing speed and quality can be a challenge. 

As they sought a solution to speed up processes and make manual consistency-checking more effective, RTL NEWS, a pioneer in digital news media, asked a bold question: How can we empower our editors with AI-powered tools so they can focus on what truly matters? 

With the support of our partner Google Cloud, we helped RTL NEWS further develop their existing prototype – which has additional AI-specific application options – into a solution that not only accelerates existing processes but also provides editors with full visibility and control.  

With the right AI tools in place, RTL NEWS can now automate routine tasks such as topic selection and quality control, accelerating time-to-market, relieving resource constraints and paving the path for higher-quality content. 

AI module 1: Hot Topic Discoverer

Navigating a sea of information can slow down editors when trying to pinpoint relevant topics. Using AI, the Hot Topic Discoverer analyses news articles and evaluates them based on several factors and criteria that determine the news’ value. 

While the criteria remained consistent across media outlets, the weightings for these factors varied depending on the editorial priorities of each platform. For instance, some platforms prioritised emotional resonance, while others focused more on geographical proximity or unique topics. These weightings were carefully calibrated using an algorithm and validated by editorial teams to ensure accuracy and usability. 

Thanks to Google Cloud’s Vertex AI, the solution analyses large datasets efficiently and provides full transparency by explaining the basis for its evaluations, ensuring editors remain in control of their decisions.  

AI module 2: Consistency Checker

The sentence-based approach, selected after testing three methodologies, ensures high accuracy by splitting text into sentences using Python’s SpaCy library and validating them with Gemini Flash. Compared to ‘claim-based approaches’, which risk hallucinations or misinterpreted claims, the sentence-based method maintains the original message and reduces errors. 

Additional tests explored embedding techniques and full-text processing, but these were less accurate than the final method. By focusing on individual sentences, the tool delivers a recall rate of over 99% with an average latency of just four seconds, making it ideal for long-form articles. 

Technical architecture of the Consistency Checker from the “RTL Consistency Checker Case Study.”

Example evaluation

The close collaboration between the partners was highly appreciated, not just by RTL NEWS but also by Google: 

“Working with Endava was excellent,” said Thomas Heigl, Key Account Director Media and Entertainment, Google Cloud. “Their technical expertise and ability to seamlessly integrate solutions into existing workflows were instrumental in successfully delivering this innovative project. The solution provided will automate routine tasks and accelerate time-to-market, which will save valuable time and costs in a fast-paced newsroom environment.”