Shelby Grossman trains journalism students to use ChatGPT, but one of her newest projects began with a more ordinary problem: almost no parent has six hours to watch a school-board meeting on a weeknight. A five-minute exchange about a closure or safety decision can disappear inside the recording.
At the Howard Center for Investigative Journalism at Arizona State University, Shelby looks for practical ways AI can help reporters pursue ambitious work. Her newest testing ground is The Beam, an Arizona outlet where professional journalists work alongside student apprentices. Its first beat is accountability in education.
The Beam first considered publishing AI-generated meeting summaries. Its pilot showed that parents wanted to ask their own questions: Did my child’s school come up? Was it discussed in connection with closures? What did the board say about 3D-printed knives infiltrating schools? Grossman’s team built eight custom GPTs, one for each district in the initial launch. Each used transcripts from the district’s latest YouTube recordings. A parent could ask a question in ordinary language, find the relevant exchange, and check the answer against the transcript. People with free ChatGPT accounts could ask about five questions a day at launch. Within 48 hours, parents asked The Beam to add their districts, and Grossman added them within hours. The outlet plans to cover all 74 Arizona districts that post board meetings on YouTube.
Verification is part of the product. Student intern Elizabeth Tomlin asked every chatbot the same four questions, then compared each response with the transcripts. She found no invented quotes, but the bots sometimes answered a question about the “most recent” meeting with material from an older one. The Beam tells users to check the original video, which remains the final source after the chatbot shortens the search.
Grossman uses the same principle in her ASU classes: AI should bring reporters closer to evidence. One prompt asks for ten experts on raising llamas in Arizona and requires a link for each name. If two leads fail, the other eight can still save hours. In another exercise, a student used Codex to collect 200 PDFs linked from a government website in minutes, a job that once required a reporter who knew how to build a web scraper in Python.
She also uses AI to share reporting across the newsroom. When a student drops an interview transcript into a Google Drive folder, an automated summary goes to the team’s Slack channel so everyone can see what was learned and what needs reporting next.
