My QUT colleague Kateryna Kasianenko is the next speaker in this session at the AANZCA 2025 conference, focussing on how search engines respond to searches about conspiracy theories. Search engines are a common pathway towards conspiracist information; they have the potential to affect their users’ understanding of such information. What people see when they search for such content also depends directly on how the query itself is formulated, so query variations also need to be studied systematically. Our Australian Search Experience 2.0 project within the ARC Centre of Excellence for Automated Decision-Making and Society explores the impact of such query variations.
It is difficult to do so, however; finding or generating realistic query variations is a non-trivial task. Queries may vary across various dimensions: length, syntax, intent, ideology, beliefs, and stances; they are further changing due to the introduction of AI Overviews and the different forms of engagement that AI-enhanced search offers.
This paper focusses on the chemtrails conspiracy theory. It is now very well established, and often conflated with genuine research and development in geoengineering, for instance with the aim to compact climate change. The assumption here is that search engines may have some guardrails in place to address queries which explore this conspiracy theory.
To do so, the project formulated a range of general and conspiratorial queries related to chemtrails, also making distinctions between shorter and more conversational query formulations. Some such query formulations were informed by Websites promoting the chemtrails conspiracy theory, as well as information gleaned from Google Trends, and conspiracist hashtags groups on Instagram.
We then gathered the search results for these queries from Google, repeatedly, and assessed the similarities between the results sets shown on the first page of result. Results sets tended to cluster thematically rather than on the basis of queries supporting or challenging the conspiracy theory. This means that regardless of the stance embraced by the query itself users will usually receive very similar results; this could be beneficial or problematic depending on the nature of the results.
A closer look at these results shows that the sources recommended are broadly authoritative and reputable, but rarely localised to the user; most search results are recent, but some prominent pages are older. Some conspiracist pages are also included in the results on occasion, however.
AI Overviews accompanying these organic results presented a somewhat different picture; AIOs did not appear for all queries, but those that did usually did not push back explicitly against conspiracist beliefs. Here, AIOs tended to be more specifically localised to Australian users. At times, very specific queries like ‘fake clouds’ returned more general AIOs about geoengineering, which may suggest that the conspiracist concept is indeed linked to real-world geoengineering practices.
More detailed analysis is necessary here, of course; the project will also extend its analysis to other themes and topics to observe the impact of query variation there, especially also with a focus on more recent conspiracy theories that may not have attracted any intervention from search engine providers.
