Summary:
A comprehensive study by researchers at the University of Waterloo reveals that large language models (LLMs) suffer from a pronounced “status quo bias,” consistently recommending existing, high-emission choices over innovative, climate-friendly alternatives. Evaluating 11 prominent AI models across nearly 55,000 tested prompts, the team found that chatbots endorsed existing policies twice as often as proposed climate reforms and recommended electric vehicles far below actual real-world adoption rates. The findings suggest that defaulting to generative AI for consumer and policy decisions poses an overlooked barrier to decarbonization.
Key Facts:
- The Policy Bias Gap: When asked to evaluate whether to proceed with an initiative, AI models approved existing, status-quo plans 70% of the time, but endorsed novel climate policies only 34% of the time.
- Lagging Behind Real-World Adoption: Across jurisdictions, AI platforms recommended electric vehicles at rates lower than existing national market sales, even in EV-dominant countries like Norway.
- Scale of Testing: Researchers tested 11 LLMs across more than 7,500 distinct queries spanning transit, home heating, diets, and public policy, yielding nearly 55,000 total benchmarked prompts.
Source: University of Waterloo
As millions of consumers, business leaders, and policymakers turn to conversational artificial intelligence for everyday purchasing advice and strategic planning, algorithms are quietly shaping societal decision-making. Yet because large language models (LLMs) are trained on massive historical corpora reflecting past human choices, their recommendations inevitably reflect entrenched historical norms.
In psychology, the tendency to prefer existing conditions over novel alternatives is known as the “status quo bias.”
Now, a pioneering computational and environmental study from the University of Waterloo demonstrates that this cognitive bias is deeply baked into the underlying architecture of commercial AI chatbots—with detrimental consequences for global climate goals.
The study, which evaluated 11 major LLMs across nearly 55,000 prompt iterations, represents the first systematic investigation into how generative AI status quo bias influences climate-related recommendations.
The results reveal that chatbots systematically steer users toward high-emission defaults in transportation, home energy, consumer habits, and public policy.
“I think it’s worth being aware that these models, even the new ones, have blind spots,” said lead researcher Seth Wynes, Ph.D., professor in the Faculty of Environment at the University of Waterloo. “It’s good for consumers to be aware of this bias and if you’re asking for advice on a topic, you could ask it to make the case for doing something new.”
Evaluating 55,000 Prompts Across 11 Language Models
To test whether AI chatbots passively preserve the status quo, the Waterloo research team developed a standardized library of more than 7,500 unique queries. The questions covered high-impact personal and civic domains where carbon trade-offs are significant:
- Vehicle acquisitions and personal transit choices
- Home heating systems and residential energy retrofits
- Dietary preferences and food recipes
- Regional and municipal climate policy decisions
Each query was tested across a battery of at least six distinct LLMs, producing an empirical dataset of nearly 55,000 model responses.
The bias was most pronounced in political and civic governance scenarios. When asked whether a government or organization should proceed with a plan, the language models approved the proposal 70% of the time if it was framed as an existing, active policy. However, when the exact same objective was presented as a novel climate intervention or regulatory change, the models agreed to proceed just 34% of the time.
Where climate trade-offs existed in policy deliberations, platforms reinforced decisions the user had already made with double the frequency of alternative paths.
Lagging Behind Real-World Clean Tech Transitions
The researchers found that while LLMs adapt somewhat to regional user contexts, their baseline advice still lags significantly behind actual real-world transitions.
For example, when a prompt indicated that the user was located in Norway, where battery-electric vehicles command the overwhelming majority of new car registrations, chatbots were more inclined to recommend an EV than if the user specified they were in Canada.
However, in both countries, the frequency with which AI recommended purchasing an electric vehicle fell substantially below the actual, verified rate at which EVs are being bought today in those markets.
“In almost every jurisdiction, the LLMs are working at a slower pace of change than is needed, so they recommend fewer EV models than what are actually being sold today,” Dr. Wynes explained. “It might be very good for AI to favour the status quo for lots of other things, such as proven medical advice, but climate is where we really do need change.”
Mitigating Algorithmic Inertia
The findings highlight a growing risk: as generative search engines and conversational chatbots become embedded into commercial purchasing flows and enterprise advisory workflows, their built-in bias toward historical habits could delay essential sustainability transitions.
While favoring the status quo may offer useful guardrails in safety-critical domains like established clinical medicine or structural engineering, applying that same inertia to rapidly evolving challenges like the energy transition actively impedes progress.
The research team plans to track future iterations of frontier models to evaluate whether updated fine-tuning approaches or integrated browsing tools mitigate this inertia, or if in-platform AI transactions will further cement legacy high-carbon consumption patterns.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- Journal paper reviewed in full.
- Additional context added by our staff.
About this Neurology and Neuropharmacology Research:
- Media Contact: Pamela Smyth
- Source: University of Waterloo
- Image Credit: Image credited to Neuroscience News
- Original Research is Open Access: Environmental Research Communications (Oct 1, 2026). “Large language models exhibit status quo bias in climate-relevant advice.” Authors: Seth Wynes, Aarya Shah and Victoria Milardović.
- DOI: 10.1088/2515-7620/aea1eb
Abstract
Large language models exhibit status quo bias in climate-relevant advice
Large language models (LLMs) are increasingly used by consumers and decision-makers to provide advice, including on choices relevant to climate change.
Status quo bias is the tendency to favor existing conditions over change. It is well-documented in human psychology and could be problematic in LLM climate advice given that rapid societal changes are required to sufficiently reduce emissions.
We evaluated 7548 queries spanning multiple domains including vehicle purchases, recipes, home heating, and climate-relevant policy trade-offs. Each of these queries was tested on at least six different LLMs for a total of 54 888 prompts.
By asking climate-relevant trade-off questions that held all variables constant except status quo framing, we found that LLMs had, on average, 7.5 times higher odds of recommending a course of action when it was framed as preserving existing conditions than when it required change, a finding that persists in leading 2026 models.
In consumer contexts, LLMs were 4.1 times more likely to recommend electric vehicles to users who already owned one and vehicle recommendations correlated with regional adoption rates (rs = 0.48, p < .001) but rarely exceeded them. Recipe recommendations demonstrated no partiality for (low-emissions) plant-based options (40% vegetarian or vegan).
These consumer findings complement the evidence from the more controlled tradeoff findings and together suggest that, on balance, LLMs systematically favor existing conditions in their recommendations. While conservative advice may be appropriate in many contexts, climate mitigation requires departure from high-emission norms.
Users seeking climate-relevant advice from LLMs should be aware of this tendency, and developers of climate or consumer-oriented applications may wish to address it through system prompts.
