In 2023, we launched the American Journalism Project’s Product & AI Studio. In addition to building out the Studio team, initiating grantmaking, and identifying partners, we formed our first learning cohort of 13 organizations from our portfolio. The Studio, made possible by OpenAI with support from the Patrick J. McGovern Foundation, was created to thoughtfully explore how emerging technologies like generative AI tools might help newsrooms better serve local news communities.
Recently, the cohort met to share what they learned over the course of the year. Here are some highlights:
Exploring the smart adoption of AI
The rapid rise of generative AI tools and technology took many by surprise. Newsrooms typically wait to adopt new technologies until their business benefits are clear. With free tools from ChatGPT, Claude, Gemini and other services that also promise various advantages, newsrooms had little of this traditional control. Anyone could access it, try it and make their own decisions.
Finding smart ways to use and adopt AI, including through discovering its limitations and the practical constraints of platforms, was valuable intelligence for everyone.
“Our first attempts at building GPTs were hilarious,” Stephen Busemeyer, managing editor at Connecticut Mirror, said during the presentation. “The bots were usually wrong, belligerent, uncooperative and reluctant to respond to corrections.”
Stefanie Monge, editor at Silicon Prairie News, said learning these limitations allowed their organization to see there was “a more targeted problem to solve.” Monge said she also initially found it difficult to find the time and resources to truly prioritize AI development over other ongoing projects in the newsroom. “If we could start again, we’d devote more human resources to the project.”
Making progress: Using AI to maximize newsroom resources
For many, understanding AI’s limitations was the first step toward making meaningful progress in AI development. It didn’t take long before the cohort discovered what AI could do well.
Sahan Journal’s team began with a goal to create an AI tool that would generate bespoke media kits for clients and sectors, freeing up their sales team to focus on building relationships.
“We learned there’s a reason our sales team had tabled this work in the past,” Cynthia Tu, AI specialist at Sahan Journal, said. “It is visually complicated.”
While they didn’t meet their original goal of creating custom sales kits, they did learn that with the right prompting, AI can do “heroic” things in other areas of the organization.
“Well-designed prompts do heroic work at extracting and presenting audience data,” Tu said. “Even from a messy PDF.”
Another area news organizations hoped to utilize AI was in text messaging their audiences. Outlier Media inserted AI technology into the new version of its text message information service in the hopes that it would better serve readers and ease the strain on newsroom resources.
Sarah Alvarez, founder of Outlier Media, believes integrating AI into core newsroom technology now offers some short-term benefits, and she expects it to continue saving time and money in the future.
With limited budgets, many nonprofit newsrooms want to use AI to increase productivity and save on resources. With that in mind, Civic News Company focused on building a GPT to help save Chalkbeat reporters time gathering information.
They built a GPT to transcribe public school board meetings and summarize highlights for their small staff of reporters. This has already led to the transcription of over 60 public meetings and the identification of impactful stories they would have otherwise missed.
Chalkbeat’s managing editor, Eric Gorski, said that in 2025, Civic News plans to refine the tool and its interface and incorporate it into the workflow of all eight of Chalkbeat’s local bureaus.
“As we move forward, our goal is to seamlessly integrate AI tools into our newsroom workflows,” Gorski said.
Top takeaways: News organizations identify similar needs as they consider implementing generative AI
After a year of experimenting, the challenges these news organizations faced were ultimately very similar across the industry. They all had limited resources to devote to AI development. There was also resistance to AI within the newsroom for various reasons and a lack of experience using AI tools. Lastly, many news organizations struggled with understanding how to maintain reader trust while employing AI tools.
“It turns out there won’t be a moment where everyone suddenly gets it,” said Scott Smallwood, CEO of Open Campus. “Getting non-technical staff to adopt these tools is a challenge.”
He built several GPTs for both reporter research and data analysis, all of which took time and training for newsroom staff to implement.
Smallwood said during the meeting, when it comes to AI training, “1-on-1 coaching and office hours made much more of a difference than anything else. Hand-holding doesn’t mean you’re doing it wrong. It probably means you’ll make a bigger change.”
For some newsrooms, developing AI wasn’t the problem, but having limited resources to monitor it was. The City built two chatbots and had planned to make them public, but after further research and testing, they decided against it.
“We needed to have more editorial resources; a larger content library and more people to meet potential user demand,” said Celeste LeCompte, a product and research analyst consulting with The City. “AI couldn’t solve that problem.”
LeCompte thinks internally facing AI applications that improved workflow would have helped build newsroom familiarity with AI and been more useful than focusing on public-facing AI development.
Moving forward, the Studio will try to focus on just that — increasing newsroom productivity through AI solutions. As the Product and AI Studio prepares for the 2025 cohorts, we are grateful for the contributions of each of our early partner newsrooms and look forward to building upon this experimentation this year.
Click here to listen to or watch the meeting and presentations.
