Lake Region Bulletin: Turning published stories into video explainers
A link goes in, a one-to-three minute explainer comes out — and the unsolved part is making the voice sound Kenyan.
- How can a text-first newsroom produce video without field production?
- What do AI voices get wrong locally?
- What does it cost?
Problem
LRB began as a text-led publication built on long-form stories, but its online audiences increasingly prefer video. With a staff of eight to 10 and high turnover, producing video conventionally is time-consuming, and building real video capacity would have meant a major change to how the newsroom works. The objective was to produce short video explainers of already published stories without having to go back out to the field.
Solution
The tool turns a published long-form story into a video explainer of one to three minutes. Once a story is published, an editor uploads the link to the tool, which summarises the story and generates the explainer. The explainer is posted on social media with a link back to the full story, acting as a teaser that leads audiences to the website.
After trials with several AI tools, LRB settled on NotebookLM as the closest fit to what it was looking for. The same two editors who run the fact-checking agent run this tool, and production is kept at a central point rather than letting everyone generate video independently, a rule written into the newsroom's AI guide.
Results
A few explainers have been produced and some published. Viewership has not changed much yet, which the newsroom attributes to having only just begun publishing. The refinement list is specific to tone, accent and logos. They are training it toward a more Kenyan voice because the default voice "feels too American or European." They plan to scale up to seven videos a week, compared to one or two previously.
Insights
The explainers are teasers, and not replacements of original stories. Short video reaches audiences where they are and pulls them to the long-form content on the website. The wider lesson is economic, according to LRB. AI can help when newsrooms cannot afford to support full multimedia production. However, it cannot do everything on its own, and people still need to be involved in the process.
Cost
The clearest cost signal is the plan to hire two video journalists to support it. The subscription cost USD 858.40 for two 12-month licences (USD 452.40 for NotebookLM, USD 406 for ChatGPT).
Replicability
This is a low-cost route into video for text-first newsrooms. A text story link is inputted producing the explainer with no field production, and it's run by two editors. It needs a steady flow of long-form stories worth repurposing, social channels where short videos reach audiences the website does not, and a human checking each explainer before it goes out. The real adaptation work is voice. Default AI voices did not fit LRB's Kenyan audience, and localising tone and accent remains unsolved. The expectation is to trial several tools before one fits.