SPOKANE, Wash. — The Columbia Basin Ledger has five reporters to cover 14 towns, three school districts, two counties and a regional water authority. Until this year, that meant most public meetings went uncovered.
Now an artificial intelligence transcription tool listens to livestreams and recordings of more than 30 meetings a week, producing searchable transcripts within an hour. Reporters scan them for news, and an editor checks every quote against the original recording before anything is published. “We used to pick two meetings a night and hope nothing big happened at the others,” said managing editor Diane Holcomb. “Now we at least know what was said everywhere. That has changed what we can cover.”
A tool, not a reporter
The Ledger is one of dozens of small and mid-size news organizations experimenting with AI to stretch thin staffs. Most are using it for tasks such as transcription, translating community announcements and summarizing long public documents, rather than writing stories.
Holcomb said the paper adopted strict rules before turning the tool on. No AI-generated text appears in published stories. Every quote must be confirmed by a reporter listening to the recording. Transcripts are treated as notes, not sources, and the paper publishes a short explanation of its policy on its website.
- 30-plus public meetings transcribed each week
- Under 1 hour from meeting end to searchable transcript
- Every quote checked against the recording by a human
- No AI-written text in published stories
The approach has already produced stories the paper might have missed. In August, a reporter searching transcripts noticed that a rural water board had quietly discussed a 40 percent rate increase. The Ledger’s coverage led to a packed public hearing the next month, where the board scaled back the increase.
Errors and trust
The tools are far from perfect. Holcomb said transcripts frequently misspell local names, confuse speakers when several people talk at once, and occasionally invent words when audio quality is poor. In one early test, the software attributed a council member’s comment to the mayor.
The transcript is a map that shows us where to look. It is never the destination. If we ever publish a quote we didn’t verify ourselves, we have failed our readers.
Diane Holcomb, managing editor, Columbia Basin Ledger
Media researchers say that kind of discipline will determine whether AI helps or hurts local news. A recent study by the Center for Community Journalism at Cascade State University found that readers were generally comfortable with newsrooms using AI for behind-the-scenes tasks but were much less comfortable with AI-written articles, especially when the use was not disclosed.
“Trust is the only asset a local newsroom really has,” said Professor Alan Whitmore, who led the study. “Being open about what the tools do, and keeping humans responsible for every word, is how you protect it.”
Some journalists worry that publishers will see AI as a reason to cut staff further. A representative of a regional journalists’ guild said members support tools that reduce drudgery but will push back against any effort to replace reporters who attend meetings in person and build relationships with sources.
Holcomb said reporters has no plans to reduce its staff and is instead hoping to hire a sixth reporter next year with savings from a smaller freelance budget. “The software can tell us what was said,” she said. “Only a reporter can tell you why it matters.”
