Journalists build AI tools to track news trends & audience analytics
The JournalismAI Skills Lab brings together six newsroom leaders to present hands-on AI tools built during the program, highlighting how large language models (LLMs) can solve newsroom problems when paired with editorial judgment.
The 2025 Skills Lab program focuses on building AI literacy for newsroom leaders. It emphasises understanding how AI works, where it fails, and how it can be applied responsibly in journalism.
The first presentation — Afore Hsieh (CBC, Taiwan), introduced an AI tool designed to help foreign journalists track trending topics in China.
Foreign reporters face physical restrictions and language barriers when reporting on China, making it harder to understand what people inside the country are talking about in real time, she explained.
To address this, Hsieh built a system that scrapes trending data from Baidu Search (China’s equivalent of Google Trends) three times a day, translates it into English using an LLM, categorises it, and stores it in a searchable database.
The tool currently holds more than three months of data — potentially over 10,000 data points — and allows journalists to filter topics by keyword, date, or category.
“AI did most of the coding, but humans still need to design the workflow,” Hsieh noted, sharing lessons learned around performance issues, scaling, and the need to redesign systems as data grows.
Despite having no technical background, Hsieh used AI for over 90 per cent of the coding process. It is publicly accessible at https://whatstrendinginchina.com/
Automating the boring stuff (so journalists can do journalism)
The second presentation — Nalin Diri (ANKA News Agency, Türkiye) shared how she built a tool to automate the conversion of dozens of daily municipal press releases into publishable news stories.
While these press releases are important, they consume entire workdays for reporters and editors despite offering low editorial value, Diri explained.
Through the Skills Lab, Diri learned to avoid costly and inflexible approaches like fine-tuning and instead built a prompt-based Application Programming Interface (API) workflow that extracts the core story, removes promotional language, applies editorial style rules, and produces short drafts for editor review.
The result: less time spent on routine tasks and more time for original reporting.
“AI literacy was the real gain,” Diri said. “Once you understand the limits, you stop being afraid of it.”
From gut feeling to data-informed editing
The third presentation — Graciela Rock (La Cadera de Eva, Mexico) introduced AURA — a tool to support editorial decision-making with real insights rather than intuition.
Rock explained that editors often spend hours analysing trends manually, while monthly performance reports frequently go unused.
AURA connects data from sources such as Google Analytics and Smartocto to generate daily leaderboards, competitor comparisons, and automated email recommendations suggesting topics, angles, and even potential authors.
The tool evolved significantly during the program — from a simple concept into a more complex system using APIs, visualisation tools, and scoring agents — and is now being rolled out across more newsroom sections.
Despite challenges such as inconsistent data and API limitations, Rock outlined future plans including dashboards, social signals, and potential partnerships with small and medium-sized newsrooms in Latin America.
Fixing visuals, fast
The fourth presentation — Mads Ommundsen (Fædrelandsvennen, Norway) presented DataCanvas — a tool to help journalists create better data visualisations — without forgetting basic storytelling rules.
The tool reads articles to understand the journalistic angle and recommends appropriate charts using specialised AI agents to ensure accuracy, clarity, and correct sorting.
DataCanvas can also clean messy spreadsheets and suggest the best visual format.
“Automation works best when journalists stay in control,” Ommundsen said.
One story, many platforms
The fifth presentation — Rina Torchinsky (The Washington Post, USA) introduced Nexus — an AI-powered analytics tool that connects articles, social posts, podcasts, and other formats to a single story using text embeddings and similarity scoring.
By linking content across platforms, Nexus helps newsrooms better understand reach, engagement, and impact — supporting smarter editorial and business decisions.
While still in development, Torchinsky shared plans to expand platform integration and continue testing models through real user feedback.
Cleaning the metadata mess
Closing the showcase, Briana Smith (NPR, USA) presented Clio — a metadata-focused tool designed to tackle data quality problems caused by disconnected systems and inconsistent tagging.
Rather than building a single tagging script, Smith developed Clio as a strategic sandbox that allows newsrooms to clean historical data, test new taxonomies, and use semantic search powered by vector embeddings.
Demonstrating its impact, Smith showed how semantic search dramatically outperformed keyword search — surfacing more relevant stories by understanding concepts, not just words.
What this showcase made clear
From tracking China’s trending topics to fixing messy data and automating low-value newsroom work, journalists from around the world shared how they’re using AI — practically, responsibly, and creatively.
Across all projects, one theme stood out: AI works best when journalists understand it.
The JournalismAI Skills Lab emphasised experimentation, iteration, and AI literacy. Showing that meaningful newsroom innovation doesn’t require everyone to be an engineer, but it does require curiosity, critical thinking, and editorial responsibility.
For more information on Skills Lab: https://www.journalismai.info/programmes/skillslab
Written by: Nerina Rosli
