A year ago, Director of Product Hilary Becker and her team at The Beacon set out with a big idea to use AI and large language models (LLMs) to help reporters scour public meeting transcripts and other records for compelling story ideas, with a focus on the Kansas and Missouri statehouses.
The Beacon is a nonprofit newsroom based in Kansas City, Missouri, that connects people to the health, government, education, and economic policies that affect their lives in Kansas and Missouri. Becker joined the American Journalism Project’s Product and AI Studio to test tools that could help reporters track the Kansas and Missouri legislatures.
Their approach to AI was cautious, but experimental. They tested ChatGPT Enterprise, Everlaw for Good, Hamlet, and LocalLens. They quickly realized some of AI’s limitations in achieving their original goal. The use of different livestream platforms across states and committees made it challenging for AI to parse the information that The Beacon needed.
These challenges forced The Beacon to pivot to a “testing mindset” to explore how AI could help their newsroom.
How the Beacon used AI in its newsroom
The team’s experience with AI going into the cohort was in its beginning stages, but they wanted to hone in on how it could help them be more efficient. They were already working on an AI policy, and team members were using AI individually to streamline tasks.
Over the past year, the team used AI to:
- Sort through zip codes affected by USPS service cuts and find sources and event attendees
- Double-check information from school board meetings
- Parse over 1,000 feedback form submissions by readers to support a funder pitch
- Generate story-specific CTAs to grow newsletter subscribers
- Extract quotes and themes from large data sets
- Produce monthly community engagement reports more efficiently, find editorial themes and questions
- Use custom GPTs to test writing news quiz questions
- Design a health “brief” for their health reporters to find the latest health headlines across Missouri (example below)
The Beacon’s education reporter used LocalLens to capture and transcribe school board meetings, and the rural health reporter used ChatGPT to identify zip codes most affected by postal service cuts, aiding in targeted Facebook ads for events and sourcing for stories.
Becker and her team designed a custom GPT to generate newsletter CTAs for high-traffic stories, particularly those performing well on Google Discover. The process involves adding a story to the custom GPT, generating CTA options, editing, and implementing them via Wisepops templates. These custom CTAs have consistently shown a 1% to 2% conversion rate, significantly higher than the 0.2-.09% conversion rate of their CTAs before they started creating them with their custom GPT.
They also built custom GPTs to test an idea for news quizzes designed to deepen reader engagement. Unfortunately, their readers didn’t latch on. But using AI to build the quiz has them testing other options for interactive elements they can use on their website.
Challenges and advice for adopting AI
The Beacon’s team is relatively new to using AI, so there’s still some hesitancy among staff members. “You should always approach AI with a grain of skepticism because it can still hallucinate. It’s not perfect,” said Becker.
For example, the newsroom is not willing to use AI to compose full stories due to integrity concerns, but sees value in tools that can find themes in archival data over extended periods.
Its “scrappy” and testing-oriented culture, supported by leadership, fosters openness to new tools. Becker invites her co-workers to approach her with questions about how they can use AI to improve their workflows. She encourages them to be curious.
Their AI policy prohibits using AI for writing editorial content but permits its use for writing things like CTA and development messages, and coming up with ideas to optimize for SEO.
“Hilary is the best steward of the integrity of our product because she’s built it from the ground up. No one cares about it more than her. So I just trust her and our editor-in-chief so much in terms of the decision-making and holding that line when it comes to AI,” said The Beacon’s CEO, Stephanie Campbell.
CalMatters’ Digital Democracy website is a treasure trove of useful information, but it was overwhelming for readers to navigate. That’s why they used AI to deliver the information in one weekly report.
