Centralising fragmented data for community media using AI

JournalismAI Skills Lab
Written By Lakshmi Sivadas

Sara Cooper is the head of digital product and co-lead of the digital team at Correctiv in Germany. The JournalismAI Skills Lab helped her prototype an AI-powered tool to centralise audience data

When Sara Cooper applied for the JournalismAI Skills Lab, a programme supported by the Google News Initiative, she had a germ of an idea. She wanted to enhance the capabilities of the existing customer relations management tool used by her organisation, Correctiv. Specifically, she wanted to solve the data fragmentation that occurred for the community media organisations that Correctiv worked with. She anticipated that AI technologies might help her develop a solution. 

Cooper wanted to use the Skills Lab to fill a gap at Correctiv. “We have been very slow to develop a good dashboard that helps our media partners collect and make sense of all the data from the customer relations management tool,” she said. “So I wanted to use the programme to prototype this dashboard.”

“I just had an intuition that it would be cool to allow our partner newsrooms to chat with their audience data, ask questions and get answers chatbot-style.”

Correctiv works with 16 partner newsrooms in community journalism, including local newsrooms and national outlets. These community partners’ content is free to access and does not have a strict paywall.

“They do a lot of community initiatives and ask for memberships or ask for donations to support their core work instead of actually tying the revenue to individual pieces of content or products,” she explained.

The result is that most of the data collected is fragmented across platforms such as Mailchimp for newsletters and ticket managers for events, making it difficult for newsrooms to make sense of the information. Cooper’s solution was to centralise all of this information in their existing customer relations management tool.

As she started building it out, she learnt that, despite knowing data flows, they had no idea how to integrate AI, specifically realising that an LLM could not effectively run over a real database due to the massive infrastructure needed and the LLM's poor performance with numbers.

“I knew how to do the API calls and I knew in theory how one interacts with an LLM via Python code, but I had absolutely no idea how I would feed it data and get something in return,” Cooper said. Moreover, she found that her original idea was a little different from what she actually built.

“I had some kind of very uninformed idea that I could actually let the LLM kind of run over my actual database. I realised that the size of the infrastructure that you would need to do that is huge and it's really bad at doing that. For example, it doesn't work with numbers very well,” she explained.

The prototype she ultimately created used Gemini as the LLM and a fake SQL light database set up with a Python script and filled with fake data. She then used Gradio to build the interface, which passes the user's prompt to Gemini, receives the SQL code, and runs it over the fake database to return the response to the user.

She credits the learning experience and Skills Lab instructors Jenny Romano and Pedro Henriques for the final output.

They plan to first implement the new tool within their software using a local newsroom as a test case, and then roll it out to more partners. Cooper hopes the test newsroom will help define actual engagement metrics for community journalism, as the current lack of formally documented metrics prevented the LLM from finding patterns or making predictions. They intend to use these definitions to train a model or create prompts for proactive suggestions from the tool.

Her biggest takeaway from the Skills Lab was in the confidence it instilled in her and her newsroom to keep experimenting and building with AI technologies.

“I think without going through the programme, I wouldn't have the confidence that I have now to make kind of executive-level decisions like this is where we're going to invest our resources or this is a good project for us to focus on. So that was definitely because of the Skills Lab.”

Other JournalismAI Skills Lab Case Studies

Lakshmi Sivadas
Next

JournalismAI’s 2025 impact and 2026 vision