Google and other search engines and chatbot operators must establish stronger guardrails to prevent the spread and legitimisation of conspiracy theories, according to researchers.
Their call follows new research investigating how Google searches and AI-generated overviews responded to searches on two prominent conspiracy theories, about 15-minute cities, and chemtrails.
The article below was first published at The Conversation.
Kateryna Kasianenko, Ashwin Nagappa and Caroline Gardam write:
Imagine overhearing on a bus that the city council plans to implement new restrictions under an urban planning program called “15-minute cities”. Curious, you search the term on Google.
It’s likely that at the top of the results page, an AI-generated text summarising the search results will appear. This text synthesises content from various webpages that answer specific search queries.
They are increasingly appearing not only in Google, but also in other search engines, and users are now likely to stop reading the search results page beyond this AI-generated text.
But what if the AI-generated answer – the only search result users may see – confidently makes misleading or incorrect claims? Or reinforces conspiratorial ideas suggested by pro-conspiracy search terms?
Our new research, published in the journal Media International Australia, shows this is sometimes true of Google’s AI-generated overviews. It highlights that stronger guardrails are needed to tackle the spread of dangerous conspiracy theories.
Case studies
We focused on two conspiracy theories: chemtrails and 15-minute cities.
“Chemtrails” is a conspiracy theory that has been around since the 1990s. Its believers mainly perceive the condensation trails left behind aircraft as large-scale weather or climate modification programs.
Researchers called out Google for giving visibility to conspiratorial beliefs around chemtrails back in 2015. In response, the company demoted problematic search results.
Conspiracy theories surrounding 15-minute cities are relatively younger. The term was originally coined to represent an efficient urban planning model.
But for the conspiracist, these plans represent a nefarious elite, globalist agenda that seeks to monitor and control residents through technology.
We developed search queries reflecting these conspiratorial beliefs and general ways of searching on the same topics (relying on open fora and Google Trends for query examples). We then collected data from the first page of Google search results for a week in January 2026.
We compared the AI overviews, the sources they link to, and the conventional lists of search results across the two conspiracy theories and two types of queries.
Concerning findings
We found search queries that reflect conspiratorial beliefs fetched notably different AI summaries as compared to queries without pro-conspiracy words.
This was especially true when the search query involved lesser-known narratives or incorporated words that could suggest the searcher believes in the conspiracy – for example, searching for “15 minute city segregation” as opposed to just “15-minute cities”.
The good news is that, for well-established cases, such as chemtrails, the AI overviews mention the conspiracy theory very rarely. Generally, for both topics, AI summaries for “conspiratorial” searches debunked conspiratorial beliefs.
The bad news is that such debunking is cursory. It often does not go beyond statements that conspiracy theories are “misinformation”.
However, to stop these narratives from spreading, people seeking information about them need to be presented with factual and logical arguments debunking them.
Of concern, AI summaries for the less established “15-minute cities” case actively promoted conspiratorial beliefs, even while debunking the conspiracy theory in some instances.
For this topic, 12.5 percent of AI overviews we analysed presented the false claims as one side in a legitimate debate.
For example, in response to our query, “15 minute cities versus smart cities control”, the AI overview contained the following:
Overlap: A 15-minute city can be smart by using smart mobility and [Internet of Things] to manage its local resources efficiently, but a smart city doesn’t automatically become a 15-minute city. In essence, think of the 15-minute city as the “what” (livable design) and the smart city as a potential “how” (technology to enable it), but the control aspect is where critics focus their backlash, blurring the lines between convenience and surveillance.
The ranked lists of search results – a staple of Google search for years – are not foolproof either. Our simulation of “conspiratorial” searches returned links pointing to YouTube videos and Facebook posts promoting such beliefs.
The AI-generated snippets also tended to feature more commercial results. This is understandable – businesses rely on Google to connect them to customers. And if users do not read beyond AI overviews, it is beneficial for companies to be included as sources of such snippets.
However, this can sometimes promote conspiratorial beliefs. For example, one AI overview we analysed contained a source that linked to a book on sale on Amazon by a prominent conspiracy theorist.
Accountability
A spokesperson for Google told The Conversation its AI overviews “operate like traditional search in that they aim to match content from the web to the words in your query”.
They added: “We invest significantly in the quality of AI overviews and the vast majority provide accurate information. When issues arise – like if our features misinterpret web content or miss some context – we use those examples to improve our systems, and we take action as appropriate under our policies.”
As a company connecting users to information in scenarios ranging from mundane to critical, Google has long denied it is a “publisher”.
But as the search engine moves towards providing direct answers in AI-generated snippets or chats, it may become increasingly difficult to continue denying this role.
A court in Munich, Germany, recently found Google liable for false claims contained within its AI overviews – even though, as Google argued, users are warned about possible errors and can check the answers for themselves.
Search literacy is important and users are capable of considering the authority of sources, refining search queries, and comparing results across multiple search engines. Understanding how Google ranks its results and generates AI Overviews is a part of such literacy.
But this does not absolve the company (or any other search engine or chatbot operator) from the responsibility of establishing stronger guardrails to prevent the spread and legitimisation of conspiratorial beliefs.
The example of the more established conspiracy theory, “chemtrails”, demonstrates that Google has the capacity for such guardrails. They should be deployed across a broader range of topics, including more emergent beliefs.
Of course, what gets displayed on a search result page will never be perfect – and this is where stronger user literacy should come into play.
Author details
Dr Kateryna Kasianenko is a Research Fellow at the QUT node of the ARC Centre of Excellence for Automated Decision-Making & Society (ADM+S) and QUT’s Digital Media Research Centre (DMRC). She is a computational social scientist studying social media, search engines, and other digital technologies, and focussing on examination of user practices through computational, ethnographic, and discourse-analytical methods.Kateryna Kasianenko receives funding from the Australian Research Council through the ARC Centre of Excellence for Automated Decision-Making and Society. Her related research has also been partially funded by the 2025 Queensland-Bavaria Collaborative Research Grant.
Dr Ashwin Nagappa is a Postdoctoral Research Fellow at the QUT node of ARC Centre of Excellence for Automated Decision-Making and Society (ADM+S) and the QUT Digital Media Research Centre (DMRC). He is a recipient of the 2024 AXA Post-Doctoral Fellowship (for the theme ‘Navigating misinformation and trust erosion in the digital age’). His research centres on emerging decentralized social media platforms, analyzing how they mitigate online harm and address challenges such as misinformation, hate speech, and polarization. Ashwin Nagappa receives funding from the Australian Research Council through the ARC Centre of Excellence for Automated Decision-Making and Society. His related research has also been partially funded by the 2025 Queensland-Bavaria Collaborative Research Grant. He is also a recipient of the 2024 AXA Post-Doctoral Fellowship.
Dr Caroline Gardam, from the Digital Media Research Centre, Queensland University of Technology, researches climate change communication and information integrity, multimodal and visual communication, organised climate denial, propaganda, and conspiracy theories on social media.
See Croakey’s archive of articles on misinformation and disinformation; and AI and health
Add Comment