Beyond the Headline: Methodological Limits in the Analysis of Artificial Intelligence News

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

A digital news article does not necessarily maintain a stable configuration of agency across all its structural levels. In the headline, artificial intelligence may be represented as a subject that decides, refuses, creates, or attacks, whereas the lead and main text specify the roles of developers, companies, users, and institutions, as well as the technical conditions shaping the course of the event. The article aims to construct a multilevel model for analysing agency in the headline, explanatory text, image, caption, and elements of the digital platform. The model was applied to a corpus of 30 consecutively selected publications posted in Suspilne Media’s publicly accessible “Artificial Intelligence” thematic archive between 23 July and 28 August 2026. The methodological design combines quantitative content analysis, frame analysis, syntactic and pragmatic analysis, discourse analysis, multimodal analysis, and digital platform analysis. The regulatory and security macroframe dominates 19 items, accounting for 63.3% of the corpus; the instrumental and developmental macroframe occurs in 8 items, or 26.7%; and the cultural and transformational macroframe occurs in 3 items, or 10.0%. A detailed cross layer analysis of the selected publications demonstrates that AI may be grammatically represented as the agent of an action in the headline, whereas the lead and main text specify the roles of human, institutional, and technical actors in the occurrence and development of the event, as well as the distribution of responsibility among them. This configuration is defined as managed agency. The cases examined illustrate the analytical potential of the proposed model but do not provide a basis for determining the frequency of managed agency across the entire corpus. Images and captions may further recontextualise technology within its social setting, whereas tags, links, and recommendation modules reflect the observable organisation of content on the digital platform but do not provide evidence of actual audience reception. The proposed model enables a distinction between grammatical agency and the documented attribution of roles and responsibility, between digital visibility and audience reception, and between visual contextualisation and interpretation.