Breaking News or Breaking Trust? Exploring Challenges and a Design Space for Trustworthy LLM Integration in Journalism

Key findings from this study

This research indicates that:

• Five distinct challenges structure the design space for trustworthy LLM integration: factuality verification, editorial neutrality, appropriate autonomy levels, efficiency gains, and journalist AI literacy.
• Interactive interface design can operationalize abstract principles for responsible AI use by embedding transparency, verification controls, and human oversight mechanisms into tools.
• Trade-offs exist between providing sufficient explainability for informed editorial judgment and maintaining practical tool usability under deadline pressure.

Overview

This research examines the integration of large language models (LLMs) into journalistic workflows, focusing on how interactive interface design can promote trustworthy journalism. Eight interviews with news industry stakeholders identified five key challenges: factuality, neutrality, autonomy, efficiency, and AI literacy. Four prototype interfaces were iteratively developed and evaluated to address these challenges.

Methods and approach

The study employed semi-structured interviews with eight news industry stakeholders to identify challenges in trustworthy LLM integration. Based on interview findings, researchers constructed a design space encompassing the five identified challenges. Four interactive interface prototypes were developed and iteratively evaluated through stakeholder feedback to explore how design choices address trustworthiness concerns.

Results

The research identified factuality, neutrality, autonomy, efficiency, and AI literacy as fundamental tensions in LLM-infused journalism. Factuality challenges emerged from LLM hallucination risks and verification workflows. Neutrality concerns reflected tensions between automation gains and editorial voice preservation. Autonomy involved balancing human oversight with efficiency improvements. Efficiency gains risked undermining the verification and deliberation practices essential to trustworthy journalism. AI literacy gaps affected journalists' ability to understand and appropriately deploy LLM-infused tools.

The four prototypes demonstrated how interactive interfaces can operationalize AI governance principles within editorial contexts. Design interventions included transparency mechanisms for LLM reasoning, controls enabling human verification, and decision support that preserves journalistic judgment. Stakeholder evaluations revealed trade-offs between providing sufficient detail for informed oversight and maintaining tool usability. Opportunities emerged in embedding explainable AI into everyday editorial interfaces, though challenges persisted in communicating model limitations without overwhelming users.

Implications

The design space offers a framework for developing LLM tools that support rather than circumvent journalistic verification norms. Interactive interfaces can bridge the gap between AI capabilities and editorial trustworthiness requirements by making model behavior transparent and controllable. Guidelines for responsible AI use in newsrooms require concrete instantiation in tool design rather than remaining abstract policy documents.

Newsroom adoption of LLM-infused tools necessitates sustained attention to AI literacy among journalists and editors. Explainable AI implementation must balance technical completeness with practical usability in fast-paced editorial environments. The research indicates that trustworthy integration requires iterative design processes involving newsroom stakeholders rather than tool-first approaches imposed on existing workflows.

Scope and limitations

This summary is based on the study abstract and available metadata. It does not include a full analysis of the complete paper, supplementary materials, or underlying datasets unless explicitly stated. Findings should be interpreted in the context of the original publication.