Smart Content Assistant

The problem and solution 

Content creators at VRT face a common challenge: reworking content for multiple platforms and channels is time-consuming, leaving less time for creative work and research. The Smart Content Assistant addresses this by providing smart, tailored help that maintains each brand’s unique tone of voice while speeding up repetitive tasks.  

The tool operates as a shared prompt library where users can select pre-defined prompts, upload text, audio or documents and receive AI-generated content suggestions. Currently, 112 prompts have been created, with 18 different (sub)brands used daily. Everyone at VRT can create prompts, adapt them, and create variations on existing prompts. This enables knowledge sharing across the organization.The system is built on AWS Bedrock and Azure, utilizing both Claude and OpenAI models in a secure VRT environment.  

Real-world applications 

Previously, the production team spent around two hours writing descriptions after each three-hour live broadcast of De Ochtend, a current affairs talk radio show. These descriptions were needed to repackage the programme for VRT Max, VRT’s streaming platform. Now they upload transcriptions to the Smart Content Assistant, which generates first drafts of descriptions and titles, allowing them to focus more on preparing tomorrow’s content rather than processing today’s.  

For lower-priority football matches during a soccer league, VRT combines match data with broadcast subtitles to generate first-draft articles. A human editor always reviews content before publication, and clear AI disclaimers are included. This approach enables coverage of matches that would otherwise receive minimal attention.  

A part of the tool is integrated directly into the CMS, offering features like bullet point generation, title suggestions and final checks. It also helps repackage news articles for WhatsApp channels and generates on-screen titles and text for live news broadcasts, which is a context where there is high time pressure. 

Key learnings

Success depends on understanding real user needs, that’s why you need to start with discovering what the user wants. VRT invested significant time talking with content creators to identify where AI could genuinely help, ensuring users of the Smart Content Assistant to recognize the value and adopt the right mindset.  

Despite its success, VRT identified several challenges around sustainability. Users need help with prompting skills, so they added a “prompt maker” feature (where it suggests a prompt for you). Finding the right model for each task requires matching prompts and models. Being transparent about who manages and maintains the prompts is important to gain and keep the trust of the users of the Smart Content Assistant.  

While the tool works well as a standalone application, VRT recognizes that adding another tool to the existing workflow creates friction. They have developed an API to integrate popular prompts directly into existing systems like their CMS. 

Users familiar with ChatGPT or Copilot sometimes expect similar conversational capabilities, but the Smart Content Assistant isn’t designed that way. VRT is working on better communication what the tool can and cannot do while exploring how to expand its capabilities.  

VRT is now entering a redesign phase, conducting UX/UI exercises to improve prompt discovery and usability; identified as key factors for increased adoption. They’re also addressing challenges around prompts versioning, team-based workflows and verifiedprompts for specific brands. 

Conclusion  

VRT’s approach demonstrates that successful AI implementation in media organizations doesn’t require complex systems. By focusing on real user needs, maintaining simplicity and iteration based on feedback, they’ve created a tool that saves time on repetitive tasks while preserving the unique voice and quality that audiences expect. For media organizations looking to implement similar solutions, VRT’s experience offers a practical roadmap: start small, involve users from day one and be prepared to continuously adapt based on real-world usage.  

Article written by Emma Gysen with the imput of Catho Van Den Bosch (VRT).