Skip to content
Article Columbia Journalism Review Jul 2026

Columbia Journalism Review: how to develop AI guidelines for newsrooms

Emily Goligoski wrote this feature for the Columbia Journalism Review in July 2026, drawing on interviews with newsroom leaders and policy experts. It is a practical guide for news organizations that need to build AI policies from scratch or strengthen ones already in place. The article does not evaluate specific tools or weigh in on whether AI belongs in journalism; it focuses entirely on how to build durable governance.

Establish principles before drafting rules

Goligoski’s first recommendation is to begin with values, not rules. Rigid rules go stale as tools and use cases change; principles based on journalism’s existing commitments — human oversight, editorial accountability, transparency with audiences — tend to hold up longer. The article suggests drawing on existing industry resources and being explicit about the newsroom’s current and intended AI use before writing anything down. Knowing who the audience for the policy is (staff, readers, regulators?) shapes what it needs to say.

Internal programming makes policy real

A policy document alone changes nothing if staff don’t internalize it. The article recommends ongoing education rather than a single onboarding session: periodic refreshers, town halls, demos of tools the newsroom is considering, and show-and-tell sessions where staff share what they are experimenting with. Goligoski emphasizes discussing the reasoning behind the rules, including data privacy considerations and what journalistic ethics mean in the context of AI-generated or AI-assisted work. Policy that explains its own logic is more likely to be followed and to generate useful feedback when it falls short.

Governance needs skeptics and a kill switch

The governance section addresses the structural side. Goligoski recommends forming a cross-functional oversight group — not just editorial leadership, but technology, legal, and audience roles — that meets regularly and rotates membership so no one accumulates too much influence over AI decisions. The group should treat the policy as a living document with explicit review dates. She also argues for a kill switch: a clear process for halting a tool that is not working, without it becoming a political decision each time.

Including skeptics in the governance group is presented not as a concession but as a design requirement. People who push back on AI use tend to surface edge cases and failure modes that enthusiasts miss.

Publish what you know

The final area is transparency. Goligoski recommends two versions of AI policy: a public-facing statement that explains the newsroom’s approach to AI for audiences, and a more detailed internal document for staff. Beyond publishing, the article encourages newsrooms to share what they learn — including failures — in industry forums and peer exchanges. The more openly organizations discuss where AI governance is working and where it isn’t, the faster the field can develop shared standards.

The piece is most useful for editors-in-chief, managing editors, and heads of product or technology at news organizations that are building or revising AI governance. It is less a prescription than a scaffold: the four areas it covers provide a structure to work through, with the organization’s own values and use cases filling in the specifics.