Households drive up to 72% of global greenhouse gas emissions through everyday choices in food, transport, housing and energy (Newell et al., 2021). Most people say they care. Far fewer change what they do. This article looks at why that gap exists, where AI can help close it, and how to tell whether behaviour is actually changing.
The attitude–behaviour gap
Green consumption is widely accepted as an idea, but price, convenience and brand loyalty still decide most purchases (Yin et al., 2025). Awareness is also shallower than it looks. Finnish consumers recognised the term “product carbon footprint”, yet few could explain it or link it to climate change (Rondoni & Grasso, 2021). In tourism, people’s environmental concern has almost no bearing on how they holiday (Holden et al., 2022).
Psychology plays a part. Price sensitivity and status quo bias make the low-carbon option harder for some groups to choose, and age, income, education and region all shape how people respond to environmental information (Steiner et al., 2017; Rondoni & Grasso, 2021).
But the gap is not only about willpower. Whether sustainable options are available and affordable, how choices are presented, and what the people around us do all matter as much as personal intent. Newell et al. (2021) argue that relying on conscientious individuals to “do their bit” will never be enough without changes in high-consuming groups and in the systems around them.
So the question is not how to lecture people harder. It is how to make the lower-carbon choice visible, relevant and easy at the moment it is made.
What AI and data can see
Smart meters, phones, connected devices and online purchases now produce a detailed record of how people live. Big data analytics can use that record to study energy use, low-carbon transport and recycling at both individual and population level, and machine learning improves predictions of who will choose eco-friendly products (Chandra & Verma, 2021).
In homes, smart meter data and non-intrusive load monitoring can break total electricity use down by appliance (Trakadas et al., 2022), and neural networks can learn a household’s consumption patterns and forecast its load (Wang et al., 2024). Across domains, combining carbon data with AI models makes it possible to estimate a household’s footprint more accurately and to predict how it will change (Du et al., 2024).
The practical result is personalisation. Traditional carbon calculators ask people to type in their bills and mileage, which few do more than once (Raghavasamy.M et al., 2024). AI-based tools can collect and analyse the data automatically and return feedback and suggestions that fit one person’s life rather than an average (Parhamfar et al., 2024).
Three ways AI can change behaviour
1. Nudging at the right moment
A nudge changes how a choice is presented without removing options or changing prices. Digital nudges can steer people towards lower-carbon options while they shop online, especially people who already care about climate change (Yin et al., 2025). In food, apps that track expiry dates and suggest meals reinforce less wasteful habits (Clark et al., 2025). At home, AI assistants can learn preferences and build energy-saving actions into them, though only for people who choose to adopt such devices in the first place (Zamponi & Barbierato, 2022).
2. Conversation and explanation with generative AI
Generative AI can do more than predict: it can explain, answer questions and adapt to the person asking. In a study of 467 participants, Foroughi et al. (2025) found that interactivity, responsiveness, knowledge gained, environmental concern and a sense of personal responsibility predicted whether people used generative AI for environmental information — and that use translated into greener purchasing.
Yin et al. (2025) organise this into the “Dual G.R.E.E.N Circle”: mechanisms such as guiding awareness, rewarding behaviour and nudging, paired with applications such as result visualisation, entertaining education and storytelling. The point is to make environmental information personal and engaging enough to act on.
There is a caveat. Poorly designed AI advice can reduce people’s intention to buy green compared with a human recommendation (Yin et al., 2025). Trust and transparency are part of whether the advice works at all.
3. Continuous personal feedback
Emission trackers and mobile apps let people see the footprint of what they already do (Rayburn et al., 2025). Timing matters: for climate-conscious consumers, a first eco-friendly choice often becomes a habit, so feedback that arrives early and fits the person can start a positive cycle.
Measuring whether behaviour really changes
An intervention that cannot show its effect cannot be improved or defended. Measurement is what separates behaviour change from good intentions.
Pick concrete metrics per domain. In transport: kilometres driven, share of trips by public transport, cycling or walking, and fuel per trip — all trackable from GPS and mobility data (Wu et al., 2024). In energy: kWh per household and appliance-level profiles (Wang et al., 2024). In food: waste per meal and the mix of what is eaten (Clark et al., 2025). In shopping: the share of lower-impact products in the basket (Chandra & Verma, 2021).
Compare people with themselves. Machine learning can build a baseline of each person’s usual behaviour and detect meaningful departures from it after an intervention (Chandra & Verma, 2021; Du et al., 2024). A personal baseline is fairer than a population average: someone who already cycles has less room to improve than someone who drives everywhere, and both deserve to see their own progress.
Track over months, not days. A single good week is not a habit. What matters is whether a change lasts. AI coaching systems work this way: they analyse behaviour over consecutive periods, give comparative feedback and justify their suggestions, and long-term use can shift real behaviour patterns (Cranefield et al., 2022).
Combine data with what people say. Surveys capture attitudes and motivations that sensors cannot, but they suffer from recall errors and the wish to look good (Rondoni & Grasso, 2021). Pairing self-reports with measured behaviour gives a more reliable picture.
Look across domains. Emissions come from food, transport, energy and goods together. Du et al. (2024) propose assessing household footprints across these domains and at individual, household and city level, so a cut in one area is not quietly offset by a rise in another.
Newell et al. (2021) add a broader point: lasting change spreads through feedback between individuals and the systems around them. Good measurement captures not just kilograms of CO₂ saved but whether new choices stick and spread.
The risks
The same tools that nudge people towards sustainability can push them towards overconsumption. Companies already use AI to study and steer buyers in ways that increase consumption — sometimes called “hyper-nudging” (Dubey & Alam, 2024). Three principles follow:
- Respect autonomy. A nudge should make a better choice easier, never manipulate. Nudging alone also misses the values and beliefs that motivate lasting change, so it should support deeper engagement rather than replace it (Newell et al., 2021).
- Protect privacy. Many people are wary of sharing data about what they buy and eat with AI systems (Clark et al., 2025). Collect only what is needed and be clear about where it goes.
- Share the benefits fairly. Interventions should not widen existing inequalities (Dubey & Alam, 2024).
Conclusion
Consumer choices are a major source of emissions and a major lever for cutting them. AI can help close the gap between caring and acting by estimating personal footprints, giving feedback that fits, nudging at the moment of choice and explaining trade-offs in plain language. What turns these from promising ideas into evidence is measurement: clear metrics, personal baselines and tracking over time. And all of it has to be designed around trust, privacy and fairness, as part of wider change by governments, companies and communities.
This is the ground SAVE is built on. Instead of asking people to audit themselves or face an abstract carbon total, SAVE learns each person’s patterns in travel and food and suggests realistic lower-carbon alternatives. Progress is measured against each person’s own starting point, so it is possible to see whether everyday choices are shifting, and by how much.
Sources
- Big Data and Sustainable Consumption: A Review and Research Agenda — Chandra & Verma (2021)
- AI-driven food waste management strategies in the hospitality industry — Clark et al. (2025)
- Partnering with AI: digital productivity assistants — Cranefield et al. (2022)
- A review of household carbon footprint: measurement methods, evolution and emissions assessment — Du et al. (2024)
- Corporations greening strategies: Overlooking the negative implications of AI’s contributions? — Dubey & Alam (2024)
- Determinants of Generative AI in Promoting Green Purchasing Behavior — Foroughi et al. (2025)
- The Future of Tourism in the Anthropocene — Holden et al. (2022)
- Scaling behaviour change for a 1.5 degree world — Newell et al. (2021)
- Towards the net zero carbon future: blockchain peer-to-peer carbon trading — Parhamfar et al. (2024)
- Mitigating Carbon Footprint through Green Software — Raghavasamy.M et al. (2024)
- Future Thinking Climate Change: Consumers Sustainability Perspectives — Rayburn et al. (2025)
- Consumers behaviour towards carbon footprint labels on food — Rondoni & Grasso (2021)
- Multi-Product Category Choices Labeled for Ecological Footprints — Steiner et al. (2017)
- A Reference Architecture for Cloud–Edge Meta-Operating Systems — Trakadas et al. (2022)
- Research on energy consumption in household sector — Wang et al. (2024)
- Green behavior and perceived benefits in shaping green car buying behavior — Wu et al. (2024)
- Dual G.R.E.E.N Circle: generative AI for green consumption — Yin et al. (2025)
- The Dual Role of AI in Developing Smart Cities — Zamponi & Barbierato (2022)