AI Assistant for Investigative Journalists

Local journalism has been facing a significant crisis, with an average of 2.5 local newspapers shutting down every week in 2023. This alarming trend has resulted in over half of U.S. counties now lacking reliable news coverage, which poses a unique threat to democracy. For centuries, local journalism has played a critical role in engaging voters and holding politicians accountable. However, as exemplified by the Patriot-News uncovering the Jerry Sandusky scandal, small newspapers have also been pivotal in breaking nationally significant stories.

In response to the crisis, experts and technologists propose leveraging artificial intelligence (AI) as a potential solution. Monica Lam, a computer science professor at Stanford University, acknowledges the increasing interest in deploying AI tools in journalism, but she warns that many existing tools are not sufficiently reliable. A troubling study conducted by the BBC in 2025 revealed that over half of the responses generated by major AI models contained significant issues, including factual inaccuracies and fabricated quotes. Lam’s sentiment reflects the challenges faced in creating trustworthy AI applications.

Developing DataTalk

As a direct response to the shortcomings of existing AI, Lam is collaborating with technologists and journalists to build DataTalk, a chatbot aimed specifically at supporting investigative journalists and struggling newsrooms. Funded by the Stanford Institute for Human-Centered AI and the Brown Institute for Media Innovation, DataTalk harnesses a large language model capable of extracting and interpreting information from vast public databases. This initiative has been undertaken in partnership with Cheryl Phillips, the founder of Big Local News.

Lam asserts that the decline in journalism makes it crucial for tools like DataTalk to assist journalists in efficiently carrying out their work without compromising on factual integrity. The goal is to sustain essential investigative journalism, which is increasingly under pressure.

What DataTalk Offers

Traditionally, investigative journalists need expertise in database languages and data science to uncover valuable information. DataTalk simplifies this process, allowing journalists to pose questions in a chat interface and receive rapid responses. Currently focused on campaign finance data, the tool is intended for broader application, enabling journalists to pose inquiries such as the fundraising amounts of Congressional candidates from outside their states.

The tool’s functionality has already been leveraged by the Baltimore Banner to uncover news from non-emergency call log data. Looking to the future, collaboration with Big Local News aims to expand DataTalk’s capabilities to integrate state-level campaign finance records and other vital datasets into the tool.

To ensure accuracy, Lam and Phillips consulted Derek Willis, one of the leading campaign finance journalists, to refine DataTalk’s query processes, ensuring that the AI interprets questions as intended by the journalist. By providing code used for analyses alongside explanations in simple language, DataTalk aims to bridge the gap between technical query language and Journalists’ inquiries.

Classroom Pilot and Future Directions

In the fall of 2024, DataTalk was piloted in Phillips’ “Big Local Journalism” class, focusing on campaign finance stories. Students were able to produce various articles in collaboration with local newsrooms, highlighting notable contrasts between donor pools and campaign spending. The successful narrative outcomes arose from students who not only leveraged DataTalk but also implemented their fact-checking and coding skills. Phillips emphasized that local newsrooms found great value in these stories that may not have otherwise been reported.

Further reaching out, the Maine Monitor took the initiative to conduct its analysis on campaign contributions, inspired by the results achieved in the classroom. This ensures that DataTalk isn’t just an academic experiment but a practical tool for real-world journalism.

The Broader Vision for Investigative Journalism

DataTalk represents one component of a broader initiative by Lam and Phillips to empower investigative journalism through advanced technology. They envision a comprehensive toolkit to aid newsrooms with varying capacities, from small local outlets to larger national publications. This strategy aims to produce impactful stories while offering tutorials for using the tools effectively.

Future expansions include enhancing DataTalk with Agenda Watch features that utilize AI and computational methods to collect relevant documents from local governance, enabling journalists to identify newsworthy information from community decisions.

Ultimately, Phillips highlights that their efforts aim to lower the costs associated with accountability journalism, enabling reporters to engage more deeply in investigations that matter. The integration of AI into journalism could potentially reshape how stories are generated and how accountability is upheld in local governance.