Study finds Copilot AI summaries sideline Australian local and independent media

A University of Sydney study released today finds that Microsoft Copilot’s AI-generated news summaries disproportionately cite major national and international outlets while under-representing Australian local and independent newsrooms. The pattern, uncovered after researchers analyzed hundreds of Copilot responses to contemporary news queries, highlights a new form of algorithmic gatekeeping with implications for journalism, civic life and the economics of local reporting.

The study’s authors examined responses produced by Copilot when asked to summarize recent events and tracked which publications the AI referenced. By categorizing cited sources by scale and ownership, the researchers identified a consistent tilt toward established national and international publishers, and a notable scarcity of local and independent Australian outlets in the returned summaries. That bias persisted across a range of topics, the report says, suggesting the pattern is structural rather than incidental.

Local newsrooms in Australia have faced years of economic pressure and closures, and researchers warn that exclusion from AI summaries could deepen that decline. Many readers now turn to generative AI tools for quick briefings; being omitted from those briefings risks reducing audience reach and advertising income for smaller outlets, while further concentrating attention and authority in a shrinking set of large publishers.

The study places the problem in the context of how generative systems are constructed. AI models that power tools such as Copilot rely on training data and retrieval algorithms that privilege sources with high visibility, stable online archives and clear metadata. Those attributes often correlate with larger, well-resourced organizations, meaning local reporting that is paywalled, distributed across community sites, or published by small independent organizations can be harder for retrieval systems to surface. The University of Sydney team argues that those technical design choices have social consequences when the tools are used as primary gateways to news.

The findings also raise questions about attribution, transparency and licensing. When AI summaries aggregate reporting, the absence of diverse source representation can mute local perspectives on issues ranging from planning decisions to public health. The study recommends greater transparency about source selection criteria, routine audits of source diversity, and technical adjustments that can elevate geographically relevant outlets in retrieval and ranking steps of AI-generated summarization.

Microsoft, which develops Copilot, has previously emphasized investments in attribution and partnerships with publishers, but the study’s results suggest more targeted steps may be needed to ensure equitable representation of local journalism. For policymakers, the research underscores a need to consider how platform design intersects with media policy and public interest obligations, including potential rules around algorithmic transparency and support for news diversity.

Academics and industry technologists say this moment offers an opportunity to redesign news-facing AI so it does not simply mirror existing power asymmetries. Without intentional interventions from platform operators, regulators and funders, the growing role of generative AI in shaping news consumption risks accelerating the decline of local reporting while concentrating influence in a few large outlets. The University of Sydney study calls for immediate monitoring and corrective measures so AI can inform citizens without eroding the plurality of local voices that underpin democratic life.