Beyond Words: Exploring User Engagement with AI-Generated Visual News Summaries

In an increasingly visual and fast-paced media landscape, traditional text-heavy news formats struggle to maintain user engagement, especially among mobile and social media users. This study explores the potential of generative artificial intelligence (AI) tools to create image-based news summaries that are semantically accurate and emotionally resonant. Specifically, we investigate how current text-to-image models perform in summarizing real-world news content, and how users perceive the resulting visuals.

We conducted a quantitative survey with 63 participants who were shown AI-generated images corresponding to short news articles. The participants rated the images in four dimensions: content accuracy, emotional tone, visual quality, and independent comprehensibility. ChatGPT 4o consistently outperformed other models, receiving the highest scores in all categories. However, no model succeeded in conveying the full complexity of the article without limitations. In addition, we collected data on news consumption habits and AI usage frequency. The results indicate strong correlations between positive image ratings and demographic factors such as age, gender, and education level. Younger university-educated users, particularly those who frequently use AI tools, responded most positively to the generated visuals.

In general, our findings highlight both the promise and current limitations of AI-generated news images. Although editorial oversight remains necessary, the tools under review offer a valuable opportunity to improve accessibility, inclusion, and engagement, especially in the context of digital transformation in journalism.