The Conversation: AI in journalism — weighing the risks and benefits

Timothy Koskie, a postdoctoral researcher at the University of Sydney’s School of Media and Communications, published this analysis in July 2026 as a response to Australia’s public broadcaster ABC beginning a Claude AI trial. The piece is useful beyond that immediate context because it surveys the evidence on AI in journalism more broadly — what has worked, what has failed, and what conditions determine the difference.

Koskie’s framework separates AI applications in journalism into two categories. The first is AI as an investigative amplifier: tools that process large volumes of text, video, or data faster than a human team can, surfacing patterns that reporters then evaluate and act on. The BBC’s analysis of Ukraine coverage is the example he cites — AI enabled a scale of content review that would have taken months to do manually, and the output informed journalism rather than replacing it. This category of use has a reasonably strong track record and a clear human-in-the-loop structure.

The second category is AI as a content generator: drafting articles, converting formats, summarizing documents for publication. This is where most of the trust risk sits. Koskie draws on examples from across the industry — from time-pressured AI that produced plausible but inaccurate material that went unchecked, to outlets that published AI-generated content without adequate editorial review — to argue that the failure mode is not the technology itself but the absence of rigorous verification before publication. An AI system converting a radio bulletin into a text article is automating a format-conversion task; whether the resulting text is accurate and editorially sound still requires a human to confirm.

The piece also addresses a less-discussed structural risk: AI summarization in search results redirects traffic away from the original sources, which disproportionately disadvantages smaller and regional outlets that produce the journalism being summarized. This does not directly affect newsroom workflow but matters for the sustainability of the outlets doing the work.

For editorial managers and journalists evaluating AI tools, the operational takeaway is that human oversight needs to be designed into the workflow as a required step, not treated as an optional quality check. The cases where AI in journalism has caused damage are largely cases where verification was skipped because the output looked credible. The cases where it has worked are those where reporters used AI to do work that humans cannot do efficiently, and humans retained decision-making authority over what got published.

The article was written in the context of a public broadcaster’s trial but reads as a broadly applicable analysis for any news organization considering where and how to introduce AI into its production process.