This sentiment perfectly encapsulates the challenge of combating misinformation, particularly in the context of climate change. The rise of social media has made it easier than ever for climate change deniers and sceptics to spread misinformation. This poses a significant barrier to climate action, as misinformation about climate science erodes public trust and hinders progress. While psychological research has identified effective strategies to counter misinformation, implementing corrective interventions rapidly and at scale remains a major challenge.
But what if technology could change that? What if artificial intelligence could detect and debunk climate misinformation as quickly as it spreads? This is precisely the challenge a team of researchers from leading universities in Australia and around the world set out to tackle with the Climate Misinformation Project. Their ambitious goal is to develop an AI-powered model capable of automatically detecting and countering climate misinformation in real-time.
The researchers’ quest for this “holy grail of fact-checking” led them to leverage machine learning to analyse and counteract misinformation. The international team, led by cognitive scientist John Cook at the University of Melbourne, set out to develop large language models (LLMs) that could accept a climate myth as input and automatically generate a complete debunking response.
However, there was a key obstacle: misinformation constantly evolves. New variations of misleading arguments emerge all the time, potentially making it difficult for any system to automatically recognise and debunk false claims.
Through extensive analysis, the researchers discovered a critical insight: while misinformation takes on new forms, the core misleading arguments remain the same. In fact, today’s climate misinformation largely mirrors the same myths that circulated in the early 1990s. The team found that the broad spectrum of contrarian claims about climate change could be categorised into a taxonomy with five overarching themes:
Using supervised machine learning, the team trained their AI model by matching paragraphs of misinformation to contrarian claims and repeating this thousands of times. Once trained, they fed the model 20 years’ worth of climate misinformation, analysing over 250,000 articles from 20 conservative think-tank websites and 33 blogs. The result? A powerful machine learning system capable of classifying climate misinformation, identifying logical fallacies and generating a suitable debunking response.
This groundbreaking work was made possible by a diverse team of interdisciplinary researchers. Along with lead researcher John Cook, contributors from the University of Melbourne School of Computing and Information Systems included Francisco Zanartu, Lea Frermann, and Yulia Otmakhova, as well as Markus Wagner from the Department of Data Science and AI at Monash University, each playing a crucial role in shaping the system’s success.
Once the researchers successfully developed their model, they faced another challenge—how to make it accessible to a broader array of public users. The tool had enormous potential to support informed decision-making and combat climate misinformation, but its utility was limited to the few engineers and analysts with the know-how to work with Python scripts.
The Melbourne Centre for Behaviour Change (MCBC) is a major initiative led by University of Melbourne researchers and clinicians who are recognised internationally for their contributions to understanding human behaviour. MCBC investigates the various processes that direct individual behaviour and studies evidence-based techniques capable of changing behaviour patterns to improve environmental sustainability and human wellbeing.
To bridge this gap, the team partnered with Shine to transform their cutting-edge research into a scalable, user-friendly suite of products, starting with a web application powered by AWS.
Shine took on the challenge of making the researchers’ AI model available as a fully functional product that could be accessed by a broad range of users. Leveraging AWS’s robust infrastructure, Shine built a scalable, serverless architecture that ensured high availability, efficiency, and ease of use while leveraging the Amazon Bedrock Nova Lite Generative AI foundational model.
Key components of the solution included:
Using the fully-managed Amazon Bedrock service has enabled the platform to run entirely within the AWS network and allows for simple switching between the over 150 foundation models available. It also offers enhanced safety controls with Bedrock guardrails to detect and filter harmful content or prohibited topics.
By deploying the AI model as a scalable web-based tool, Shine has empowered users to easily and quickly fact-check and counter climate misinformation. The application enables users to submit text-based statements for assessment of their credibility. If found to be misinformation, the application generates a science-backed debunking response in seconds, which can be shared with others on social media platforms to help curb the spread of falsehoods.
The AI model generates those science-backed debunking responses by applying a psychologically proven structure known as a “truth sandwich,” which consists of a fact, a myth, a fallacy, and a fact. Developed through research led by John Cook, this communication approach—shown here in the screenshot from the application—ensures that rebuttals are just as “sticky” as the misinformation itself.
John Cook, Senior Research Fellow at the Melbourne Centre for Behaviour Change and lead for the misinformation project, said, “For 18 years, I’ve worked with scientists and climate communicators around the world to understand and counter climate misinformation. In the 2000s, I collaborated with Shine to develop an iPhone app that debunked climate misinformation. Now, a decade and a half later, I’ve come full circle, collaborating once again with Shine. By harnessing the power of machine learning and large language models, we’re now closer than ever to the holy grail of countering misinformation in real-time. Having Shine offer their leading-edge AWS engineering expertise is one of the crucial missing keys we needed to make our AI model scalable and accessible to the public.”
The success of the initial application development has laid the groundwork for further expansion. With a modular design and scalable architecture, the platform is poised to integrate additional models and support new use cases in the future.
One of the most notorious platforms for spreading climate misinformation is X (formerly Twitter). Recognising the opportunity to further enhance the tool’s impact, the engineering team developed an API integration to ingest a direct feed of posts from X to continuously grow its dataset and refine its ability to detect and respond to evolving myths.
Dr John Cook, Senior Research Fellow at the Melbourne Centre for Behaviour Change
This enhancement enables the AI model to automatically generate debunking responses and send them to researchers for verification and consideration for dissemination, bringing them closer to realising their holy grail of automated, real-time fact-checking.
To further expand the application’s capability and reach, the team has commenced building gamified experiences around the model, making it more engaging for the wider public.
The collaboration between Shine and the research team has made the powerful AI model accessible for combating climate misinformation at scale. By leveraging AWS’s advanced cloud computing capabilities, the team has developed a scalable and reliable product that informs and empowers a global audience. As misinformation continues to evolve, so too will this AI-powered solution, helping to ensure that the truth can finally get its boots on considerably faster.
Image attribution: Melbourne Centre for Behaviour Change
Copyright Shine Solutions Group 2025. All rights reserved.