'What they're not telling you about ChatGPT': exploring the discourse of AI in UK news media headlines

The paper investigates how AI, particularly ChatGPT and other LLMs, are discursively represented in UK news media headlines. Motivated by rapid advances in AI capabilities, heightened public attention, and policy calls for AI guardrails, the study positions news media as central to shaping public understanding. Guided by agenda-setting and framing theories, the authors ask how AI/ChatGPT are framed in headlines and whether there are discernible patterns over time and across outlets. The aim is to provide an initial assessment of portrayals and their implications, and to identify temporal dispersion in headline volume from January to May 2023.

The authors outline the technological context of LLMs, tracing developments from BERT (2018) to GPT-3 and GPT-4, and noting ChatGPT’s rapid uptake alongside acknowledged limitations such as hallucinations and truthfulness. They highlight difficulties detecting AI-generated text and an immature but growing social research literature around AI in academia and policy. On headlines as objects of study, prior work shows headlines’ distinctive linguistic features, clickbait tendencies, and their ability to provoke negative sentiment, as well as methods for analyzing headlines as autonomous texts. Research on AI in media is comparatively sparse; Brennen (2018) found UK right-leaning outlets emphasize economics, security, and politics, while left-leaning outlets emphasize ethics and privacy, with sensationalist coverage of existential risks. These strands motivate examining headlines’ frames and agenda-setting effects for AI/LLMs.

Data were collected using Lexis Library News Search across UK in-print and online outlets for headlines containing: ChatGPT, OpenAI, AI, Artificial Intelligence, LLM (noted as L.I.M in text), and Bard. Inclusion required the keyword in the headline and clear relevance to ChatGPT/AI/Bard/LLMs; embedded link mentions were excluded. Between 1 January and 30 May 2023, 671 qualifying headlines were collected. Outlets with fewer than 20 articles were grouped as 'Other.' Seven categories were used: The Times, The Independent, The Guardian, The Daily Mail, The Daily Star, The Telegraph, and Other. Sunday and online editions were aggregated under each brand. Political leanings were assigned using YouGov’s National Readership Survey (2017). Headlines were stored and manually tagged by month and brand. The authors conducted descriptive analyses of dispersion and frequency by outlet and month, followed by inductive thematic analysis per Braun and Clarke’s six-step approach. Codes were iteratively refined to produce six cross-outlet themes: Impending Danger, Explanation/Informative, Capabilities - Negative, Capabilities - Positive, Humorous/Comedic, and Experimental Reporting, with a small 'Unclassified' remainder. Results were compared with Brennen (2018).

• Corpus and temporal trend: 671 headlines (Jan–May 2023). Across all outlets, there was a sustained increase over time, culminating in a five-fold increase in May 2023 relative to January. Six of seven news brands reached their highest monthly counts in May, indicating continued momentum. • Thematic distribution: The majority of headlines fell into two themes. Impending Danger was most prevalent with 248 headlines (37%), emphasizing imminent societal damage or severe consequences. Explanation/Informative was second with 173 headlines (26%), focusing on defining AI/ChatGPT or explaining product updates/features. Positive Capabilities constituted 11% of headlines, highlighting beneficial applications and societal gains. Negative Capabilities included 77 articles, documenting concrete errors, failures, or harms (e.g., incorrect answers, defamation, exam failures). Humorous/Comedic accounted for about 4% of headlines, showcasing playful or entertaining uses (e.g., poems, comedy, Eurovision-style songs). Experimental Reporting appeared less frequently, featuring journalists or individuals testing AI for tasks with forward-reference/clickbait elements (e.g., 'here’s how it went'). Only 32 headlines were Unclassified. • Cross-outlet patterns: Themes appeared across all political leanings, with limited ideological divergence. Some tendencies (e.g., right-leaning outlets featuring more Impending Danger and Explanation; left-leaning outlets showing relatively more Negative Capabilities) were noted but overall differences were modest compared to prior work.

The findings suggest UK media headlines commonly frame AI via two dominant lenses: explanation and impending danger. In an online attention economy, headlines' clickbait and negativity biases likely reinforce sensational frames, echoing Brennen (2018). Through an agenda-setting lens, frequent danger- and regulation-oriented headlines may elevate AI risk as a salient public concern. Framing theory clarifies how outlets selectively emphasize threats versus utility, shaping perceptions of AI as complex, uncertain, and potentially hazardous. Unlike earlier findings of strong ideological polarization, this dataset shows relatively similar framing across outlets, implying a shared, multifaceted discourse rather than a simple left–right divide. These representations may skew public understanding toward extremes of hype and fear, underscoring the need for balanced, expert-informed reporting that contextualizes risks and capabilities.

Media coverage of AI and ChatGPT surged in early 2023, with headlines increasingly framing AI in terms of both explanation and impending danger. This bipolar portrayal risks misrepresenting current capabilities and fostering anxiety. The study contributes an initial, cross-outlet snapshot of UK headline discourse, highlighting largely apolitical but sensational tendencies and underscoring media’s role in shaping public agendas and frames. The authors call for future research to link media frames to public opinion (e.g., via polling) and to extend analyses longitudinally and across geographies to capture evolving portrayals and cultural-political influences. Policymakers, AI developers, educators, and newsrooms can use these insights to promote responsible communication that supports public understanding and engagement.

The temporal window was limited to the first five months of 2023 due to rapid growth in AI-related coverage, providing only a snapshot rather than a longitudinal view. Inductive thematic analysis involves human judgment and may introduce bias despite reflexive coding. The focus on UK outlets constrains generalizability to other national contexts. Broader temporal and geographical sampling would enable a more comprehensive account of AI media portrayals.