Framing the future of search: A discourse analysis of Google’s AI Overviews

Summary

This paper analyses how Google’s AI Overviews (AIOs) — generative AI summaries inserted atop search results since May 2024 — are discursively constructed by three industry actors: Google itself, technology journalists, and SEO marketers. Combining Leximancer concept mapping with Fairclough-style critical discourse analysis on a corpus of 613 English-language texts spanning the first year of AIO rollout, the authors argue that AIOs are not a neutral technical upgrade but a contested redefinition of search, trust, and platform authority. They read AIOs as a hybrid platform-infrastructure project that deepens the “Googlization of knowledge,” shifting Google from indexical gatekeeper to generative epistemic authority while strategically displacing accountability and silencing user, ethical, and environmental concerns.

Key Contributions

• One of the first systematic, multi-stakeholder discourse analyses of Google’s AI Overviews during its rollout year.
• Methodological demonstration of pairing Leximancer concept mapping with critical discourse analysis for scaled platform discourse study.
• Theoretical extension of platformisation and the “Googlization of knowledge” to generative AI-powered search.
• Identification of discursive absences (users, ethics, environment, democracy) as a structural feature of industry framing of AI search.
• Contribution to platform studies, AI communication, and sociotechnical imaginaries scholarship.

Methods

Mixed-methods discourse analysis on 613 English-language texts (May 2024–May 2025) collected via custom web scraping for variants of “Google AI Overview,” classified by domain into Google corporate communications (30), tech journalism (253), and marketing/SEO content (330). Leximancer was used to extract themes, concept co-occurrences, and sentiment likelihoods over 2–3 sentence context blocks. Representative texts were then subjected to qualitative critical discourse analysis using Fairclough’s three-level model and van Leeuwen’s framework.

Findings

• Four dominant frames organise the discourse: (1) Generative AI Technologies and AI-Platform Wars; (2) Reconfiguring Search — Let Google do the searching for you; (3) Commercial Implications of AIOs; (4) Utopia versus Dystopia.
• Sentiment diverges sharply by actor: Google is neutral-positive (~5% favourable / 4% unfavourable); SEO marketers are highly engaged but conflicted (66% / 52%); journalists are cautiously critical (22% / 40%).
• Concept emphases differ: Google foregrounds Gemini, model, people, AI; SEO actors foreground search, engine, content, AIOs; journalists foreground Google, search, AI, company.
• Google deploys metaphors of seamlessness and helpfulness (e.g., “supersmart, superfast librarian”) while hedging via “experimental” labels; journalists mobilise hallucination anecdotes (glue on pizza, eating rocks) and framings of disaster and mistake; SEO actors document clickthrough collapse and pivot toward Generative Engine Optimization (GEO).
• Users, environmental costs of LLMs, and democratic/ethical implications are largely absent across all three actor groups.

Connections

This paper sits among other studies tracing how generative AI is rhetorically packaged, legitimised, and contested by industry and adjacent actors — closely adjacent to work on tech press and corporate AI imaginaries such as Waight2026-ts and Waight2025-al, and to discourse-analytic scrutiny of AI search and information intermediaries like Dodds2026-df. Its concern with AIO accuracy, hallucination, and epistemic risk also resonates with empirical work probing the reliability of LLM-mediated information, including Stanusch2026-ec and Schroeder2026-im.

Podcast

A research-radio episode discusses this paper: Listen