American Journalism Project: what worked in 28 local newsroom AI experiments
The American Journalism Project’s Product & AI Studio published this report on November 6, 2026, drawing on a survey of 28 local newsrooms that ran AI experiments and observations from the 2026 News Product Alliance Summit in Chicago. Authors Maggie Cogar and Caroline Porter identify what distinguished projects that delivered real value from those that didn’t.
The projects with the greatest impact shared three traits. First, they freed up human capacity — taking over repetitive tasks like first drafts, transcription, translation, and data processing — so small reporting teams could spend more time on reporting and community relationships. Second, they fit into existing workflows rather than asking staff to redesign work around a new tool. Third, they included organizational context: style guides, audience research, past campaigns, and similar knowledge supplied to the model so that outputs reflected the newsroom’s actual voice and priorities.
Specific examples are concrete. Signal Ohio used an AI campaign assistant and saw 23% year-over-year revenue growth across the three publications it supports, though the report acknowledges other factors contributed. Enlace Latino NC built an AI-assisted translation workflow that saved an estimated 150 hours and $6,500 across its first five newsletter editions, enabling a new English-language newsletter that attracted new donors. CalMatters created a “Giving Editor” tool that provides feedback on fundraising appeals while preserving human editorial voice. The Salt Lake Tribune built an archive assistant that draws on 150 years of content.
The survey counts show how tasks were distributed: 17 of the 28 newsrooms used AI for campaign and fundraising copy, 14 for audience and data analysis, 13 for transcription and summarization. Self-reported time savings on reader revenue drafting ranged from 30 to 50 percent.
The report is candid about limits. The Beacon found that a broader AI goal exceeded current technology and scaled back to smaller support tasks. CT Mirror found that AI tools didn’t know its editorial voice, its audience, or its impact — a problem addressed by explicitly supplying that context. This context dependency — the need to inject organizational knowledge into AI tools for outputs to be relevant — emerges as the most consistent design challenge across the 28 projects.
For newsroom leaders, the practical guidance is to start with a specific bottleneck rather than a general workflow redesign, and to treat building organizational context into AI tools as a core part of implementation, not an afterthought.