With generative AI as an interface to web knowledge, searching the web is shifting from navigating visible results to receiving AI-mediated answers whose processes of retrieval, source selection and citation are less inspectable.

This paper introduces AI Search Methods, a methodological framework for studying how AI systems search the web through two prompting methods. The methods facilitate AI observability by asking how search behaviour varies over time and which web environments systems reach. The Longitudinal Prompt Method repeats prompts over time, tracing shifts and contradictions in responses, source selection, citations and interfaces. The Comparative Prompt Method varies prompt framing across AI platforms, revealing platform-specific source vernaculars and epistemic behaviours. Previous applications expose two tensions: stability versus plasticity, and convergence versus divergence across systems. Between December 2024 and July 2025, ChatGPT recycled 54.8% of cited web sources, while meta-cognitive prompts elicited greater source diversity. Comparative findings show that ChatGPT, Claude and Gemini converge on canonical authors while diverging sharply on the web environments through which those canons are assembled, suggesting platform-specific citation practices and distinct versions of the web. Methods are operationalised through the AI Source Scraper, demonstrating the potential of reproducible digital methods and method-driven tool-making for understanding AI search.

References

• https://blogs.lse.ac.uk/impactofsocialsciences/2026/04/15/before-ai-agents-act-for-us-we-need-to-know-how-ai-searches-for-us/DMI26_Longitudinal Prompt Method
• https://www.digitalmethods.net/Dmi/SummerSchool2026AgenticAIontheWeb
• https://www.researchgate.net/publication/411742113_Agentic_AI_on_the_Web_Search_Action_and_the_Mechanics_of_Reassembly

This is a hybrid event; details will be shared upon registration.

Speaker:

Janna Joceli Omena is a methodologist and interdisciplinary scholar advancing the practice and theory of digital methods. Her work repurposes web technologies and AI to facilitate reproducible research methods and tool development. She authored Métodos Digitais (2019) and founded the SMART Data Sprint (2016-2023), a pioneering initiative to advance digital methods in Portugal.