CalMatters: Using custom-AI to enhance its political accountability tool, Digital Democracy

Project: Digital Democracy AI 

Newsroom size: 51 - 100

Solution: An AI-powered transparency tool that analyses legislative data to identify speakers, entities, and financial flows, helping journalists and citizens hold governments accountable


The Digital Democracy project had its origins a little over 10 years ago in the state of California, USA. A senior leader from San Luis Obispo, who had completed his service in the State Senate, was disillusioned by how much impact money had in state government. To solve this, he partnered with the California Polytechnic University to build Digital Democracy, a tool that would give Californians access to information about their State legislature. CalMatters, a nonprofit newsroom that was launched in 2015 later adopted this framework and built upon previous efforts. 

“We took their tool which made all this information available to the public and started building the capability to turn that information into journalism to generate story ideas,” said Neil Chase, CEO of CalMatters. 

The ultimate goal was to make the information more accessible to journalists and citizens by identifying speakers, entities, and financial flows. 

The problem: Expanding accountability capabilities

Once the team saw the tremendous interest from other States for a similar tool, they decided to expand the number of regions that had access to it. The JournalismAI Innovation Challenge grant was one opportunity to help them do so.

“It was the boost we needed. We knew we wanted to do the multi-state expansion. We didn't have a lot of support for it. It has helped us, I think, to see this as a specifically defined project that we can make major progress on this year and do something we think is really good that we hope is going to set an example for other folks. We're getting interest in this from other levels of jurisdictions and even other countries now,” said Chase.

During the course of JournalismAI Innovation Challenge, the team worked on expanding the tool to Hawaii. They partnered with another nonprofit newsroom, the Honolulu Civil Beat to do so. 

“Hawaii has different types of data that are more readily available. They're kind of comparable to California in the sense that there's a lot of stuff that's publicly available on their websites or databases that we could acquire via the help of our Civil Beat partners,” said Ramsey Isler, Director of Special Projects at CalMatters.

Building the solution: AI to decode public meetings

The tool works both externally for citizens and internally for journalists. Once a citizen navigates to the Digital Democracy platform, they will be able to search the database for things like every word spoken in public hearings, money donated to politicians, bills introduced and even research legislators and their activities. It works differently for CalMatters’ journalists who are using the system.

“The part that we're not exposing to the public is what I equate to the kind of thing that's in a reporter's notebook. They are notes. They are not facts. They're not things we can publish. And so we draw a line between data and making that as accessible as we can to people and the conclusions or the the ideas that the AI is identifying in the tip sheets where it says this is an anomaly,” said Chase. 

Eventually, after discussion with the editors, reporters use the tips as potential leads for new stories. 

The entire process involves heavy transcription work to log words from the meetings, and this is where AI comes in. The team is using LLMs from OpenAI to run through the data and text of bills and transcripts for the tip sheet, said Isler. 

“There's a lot of not just raw data, raw structured and unstructured data, but also just plain old text that we need to go through, and the LLMs are really good at that,” he added.

They are also working with chat systems that use custom-built AI models. As they move on to more robust transcription systems, they also plan to partner with a tech startup called Cast Insights which focuses on video transcription using several models including Gemini, DeepSeek and OpenAI tools.

The team

The team is led by David Lesher, a longtime California political journalist who co-founded CalMatters and served as its CEO and editor in the early years. He long admired the work done by the team that built the original Digital Democracy, led by CalPoly professor Foaad Khosmood, and worked with that team to design this new approach for version 2.0. In addition to Chase and Isler, the team includes engineering staff on board from CalPoly along with graduate or undergraduate student employees who have the skill sets required for the project. 

“We've got a technical team of engineers who just focus on the tool set that allows us to not only get the videos and transcribe them, but also we have an internal tool for managing the workflows and managing who's working on what and assigning tasks to our transcribers,” said Isler. 

Meanwhile, the students are mainly transcribers who also verify AI transcription and assign statements to actual speakers since that was a component that AI was not able to do when they started the project. Identification of speakers is crucial so that they may be matched with the organisations they represent, allowing users to track the influence of donations from those organisations.

The opportunities: Boosting reporters’ nose for news

The impact of the Digital Democracy tool on their newsroom has been significant, according to Chase, with the tool enabling multiple reporters to use its data to write stories about political influence and funding. 

“We've built this in a way that is accomplishing the goal of actually identifying not just who the speaker is, but the entities and the money flow and being able to say confidently that this action that was taken by the State was funded by money from these interest groups,” said Chase. 

“I think we knew it would work, but seeing it actually show up in numerous stories lately has been tremendous. We've now trained 60 plus reporters outside of our staff around California to use this and they're starting to do a few stories,” he added. 

Lessons for newsrooms 

  • Value external expert networks: One of the big takeaways from being a part of the Innovation Challenge and developing the product further was the ability to tap into the network that JournalismAI offers and explore news ideas through it, said Isler.

    “It is really useful to have another kind of outside force that is also technical and more expert in some of these topics that we can kind of just ask random questions and get some feedback,” he said.

  • Prioritise public mission and access: Maintain a core focus on the primary mission (e.g., explaining the state to its citizens) and prioritise public access and informing people over achieving commercial success alone.

Looking towards the future, Chase said CalMatters’ primary mission remains to explain California to Californians and use information to improve the state, prioritising public access and informing people over commercial success alone. They aim to expand to multiple states where feasible, but not all 50, and also plan to adapt the approach to the city and county levels to provide local government information. 

Editorially, their goals include enhancing the tool for California, serving local journalism, and commercially, generating revenue to support the nonprofit.

Explore Previous Grantees Journeys

Find our 2024 Innovation Challenge grantees, their journeys and the outcomes here. This grantmaking programme enabled 35 news organisations around the world to experiment and implement solutions to enhance and improve journalistic systems and processes using AI technologies.

Previous Grantees
Read 2024 Report

The JournalismAI Innovation Challenge, supported by the Google News Initiative, is organised by the JournalismAI team at Polis – the journalism think-tank at the London School of Economics and Political Science, and it is powered by the Google News Initiative.