What if the story doesn’t end when the credits rolled?
Stories usually end when the episode does. The audience watches the final scene, the credits roll and the mystery is solved. But what if the investigation had only just begun?
That was the question behind Zapp Detective, an interactive whodunit by Dutch public broadcaster KRO-NCRV. Rather than ending with the credits, the investigation continued in conversations with the suspects, allowing children to uncover new clues between episodes.
Extending the story beyond the screen
The AI chat wasn't designed to replace the series. It was designed to continue it.
Each week, a new suspect became available for interrogation. Children could ask their own questions, challenge alibis and investigate different theories. The conversations revealed information that had been deliberately left out of the episodes, rewarding curious players who asked the right questions.
This approach was reinforced by the way the season was filmed. Because the story followed one continuous camera perspective, viewers only saw what happened in front of the camera. The chat allowed children to explore everything that happened beyond the frame by questioning the suspects directly.
The investigation no longer ended when the credits rolled. It continued in the chat, where children gradually uncovered clues that weren't part of the episodes.
AI as a storytelling tool
The AI chat was the latest step in KRO-NCRV's broader exploration of interactive storytelling. Previous editions of Zapp Detective experimented with a livestream in which viewers helped YouTubers escape a maze by solving puzzles, a deepfake quiz that challenged children to distinguish AI-generated images from real ones and an Among Us-inspired Roblox game. Each project explored a different way of bringing audiences closer to the story.
Building the AI experience started with a simple conversation flow sketched in a notebook. From there, KRO-NCRV developed the dialogue system in collaboration with Respeak, a European AI platform that they discovered through the Future Media Hubs network.
Every suspect received their own prompt describing their personality, knowledge and behaviour. Additional prompts and trigger words allowed conversations to evolve naturally while staying true to the story. The team also found that conversations became much richer when text was combined with audio, images and video instead of relying on text alone.
More than entertainment
Although children experienced the project as a murder mystery, KRO-NCRV designed it with a broader purpose in mind.
Every conversation encourages children to question first impressions, gather information from different sources and draw their own conclusions before identifying the murderer. As Rosanne van de Leemkolk described it, the project became media literacy disguised as a murder mystery. Rather than teaching media literacy through lessons, children learn by playing.
Keeping AI safe for younger audiences
Because the experience targets children, privacy and safety were built into the project from the very beginning.
The conversations ran on European infrastructure, personal information is monitored and children are gently redirected whenever they move too far away from the investigation. The team also explored ways to encourage healthy screen time by introducing conversation rules that naturally bring chats to an end, while features such as timers are being considered for future versions.
Instead of maximising engagement at all costs, the goal was to create short, meaningful interactions within responsible boundaries.
Results
What started as an eight-week experiment exceeded expectations.
Over the course of the eight week series, the project generated around 23,000 chats and more than 580,000 messages, with children spending an average of just over seven minutes per conversation. Without any paid marketing, audiences returned week after week to question new suspects and continue the investigation, demonstrating sustained engagement throughout the series.
Key takeaways for other media organisations
Zapp Detective shows that successful AI experiences are built around the audience, not the technology. Some of the lessons from the project include:
Use AI to extend the story, not replace it. The chatbot offered new clues that encouraged audiences to return after each episode.
Keep editorial control. Carefully designed prompts and clear boundaries created more believable and engaging conversations than giving the AI complete freedom.
Design responsibly from the start. Privacy, safety and healthy screen-time considerations should be part of the design process from day one.
Hide the learning inside the entertainment. Children developed critical thinking and media literacy while simply trying to solve the mystery.
Looking ahead
As generative AI becomes more accessible, creating conversations with fictional characters will become increasingly straightforward. The real opportunity lies in designing experiences that genuinely add something to the story.
Zapp Detective shows that AI creates the most value when it supports storytelling rather than becoming the story itself. Instead of using AI for its own sake, KRO-NCRV used it to deepen audience engagement, encourage curiosity and turn viewers into active investigators.
