Generative AI in Climate Change Communication and Education

Dionísia Laranjeiro

, Ana I. Lillebø

and Helena Vieira

CIDTFF – Research Centre on Didactics and Technology in the Education of Trainers,

Department of Education and Psychology, University of Aveiro, Aveiro, Portugal

ECOMARE, CESAM - Centre for Environmental and Marine Studies, Department of Biology,

University of Aveiro, Aveiro, Portugal

CESAM - Centre for Environmental and Marine Studies, Department of Environment and Planning,

University of Aveiro, Aveiro, Portugal

Keywords: ChatGPT, Climate Change, Climate Education, Climate Communication.

Abstract: Climate communication faces challenges such as scientific complexity, misinformation or lack of personal

connection that make it difficult for the public to understand and act on climate change in an informed way.

This study was conducted to ascertain whether Generative AI may facilitate public understanding and reduce

barriers to climate communication. Questions were asked to ChatGPT, that provided clear and informative

answers, synthesising key concepts, clarifying doubts and excluding misinformation. Some answers were too

brief or general, requiring more information. As Generative AI depends upon open access information,

academia has a key role in ensuring availability of accurate science-based and policy-relevant knowledge.

1 INTRODUCTION

Climate change is one of the greatest challenges for

humanity, with severe consequences, widespread

impacts and risks predicted. The global temperature

of the earth is rising, mainly due to human activities

related with the emission of greenhouse gases. This

global warming causes changes in the atmosphere,

land and oceans and affects weather and climate

extremes, with considerable damage to nature and

people (IPCC, 2023b). The Paris Agreement (2015),

signed by 195 countries, set a global target of limiting

temperature rise to between 1.5°C and 2.0°C, which

is not being met. A study based on 40.000 interviews

with citizens of twenty countries (representing 72%

of CO

emissions) showed that citizens had little

willingness to reduce CO

emissions, namely driving

less, reducing the heating and cooling of their homes,

or limiting beef consumption. The lack of support for

climate measures was related to a perception of

economic regression, energy taxation and carbon

pricing. An important finding was that informing the

public about the mechanisms, individual costs and

gains associated with climate measures significantly

https://orcid.org/0000-0003-3347-7967

https://orcid.org/0000-0002-5228-0329

https://orcid.org/0000-0002-3663-289X

increases their overall support (Dechezleprêtre et al.,

2022). Hence, climate communication becomes

essential to meet the goals of the Paris Agreement.

There are barriers and challenges to climate

change communication, resulting in the topic being

perceived as ambiguous, uncertain and complex,

making the public understanding incomplete, distant

in space and time, and disconnected from their

personal experiences (Wibeck, 2014). These affect

people’s willingness to adopt climate-friendly

practices. In traditional media, such as television and

newspapers, the frequency of communication on

climate change is uneven and unsystematic, marked

by cyclical moments such as extreme weather events,

conferences and political meetings, and the

publication of scientific reports (Horta & Carvalho,

2017). The media's tendency towards alarmist and

sensationalist reporting, combined with the

journalistic pursuit of balance, often portrays climate

sceptics and scientists as equal in number and

influence (Wibeck, 2014), which doesn’t accurately

reflect the scientific consensus, given that over 95%

of climate scientists attribute global warming to

human activity (Cook et al., 2013). Interest in the

Laranjeiro, D., Lillebø, A. I. and Vieira, H.

Generative AI in Climate Change Communication and Education.

DOI: 10.5220/0013478400003932

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 17th International Conference on Computer Supported Education (CSEDU 2025) - Volume 1, pages 123-134

ISBN: 978-989-758-746-7; ISSN: 2184-5026

123

topic varies across countries: it is higher in Portugal,

Greece, and Chile, and lower in the United States, the

United Kingdom, and Germany. The difference in

interest is attributed to the effects of climate change

felt in the first countries, such as major fires, but also

to political polarisation. In the United States and

Australia, where the left-right divide is more

pronounced, people are generally less interested in

climate change news (Newman et al., 2022). Two

national surveys in China, in 2009 and 2016, show

that Chinese people recognise the anthropogenic

causes of climate change, strongly support

government measures and are willing to take

individual action. However, respondents consider it

less urgent than air pollution (Liu, 2023).

News consumption habits are changing, with

television losing ground to the internet. Young

people, in particular, are using social networks as

their main source of news, with X being the social

platform most widely used for this purpose. Climate

scientists are intensifying efforts in public

engagement and communication, but they compete

with a multitude of subjective opinions on social

media, making it difficult to establish themselves as

an expert authority (Alinejad & Van Dijck, 2023).

Newman (2017) conducted a study consistent with

these findings. When analysing the posts and most

active actors on X during the release of the IPCC Fifth

Assessment Report

, he concluded that the majority

came from bloggers, activists and concerned citizens,

suggesting that large audiences were more exposed to

non-traditional voices than scientists. Also, Meyer et

al. (2023) analysed the public discourse on X

choosing five climate events from 2017 to 2021.

Results showed that discussions were heavily

politicised, frequently called for action while

criticizing administrations, and highlighted potential

negative future scenarios. The platform was centred

around controversial debates and polarizing

personalities such as G. Thunberg and D. Trump.

The role of online and social media in climate

communication divides scholars: optimists highlight

the potential of interactivity and audiovisual to

enhance science communication and empower

unprivileged groups, while pessimists warn of

fragmentation and susceptibility to misinformation.

The amount of online climate content is significant

and increasing, but the quality of communication is

poor, because climate scientists play a limited role in

the social media debate. On other perspective, climate

NGOs communicate extensively online to inform,

IPCC Report https://www.ipcc.ch/assessment-report/ar5/

ChatGPT https://chat.openai.com/

https://explodingtopics.com/blog/chatgpt-users -----------

Exploding Topics identifies and tracks emerging trends and

build support, change behaviour and mobilise action

(Schäfer, 2012). For Bushell et al. (2017) climate

communication is improving, but there is still a gap

between awareness of scientific knowledge and action

by governments, industry and people. Causes cited

relate to the nature of the problem (e. g., long-term

challenge requiring action now and lack of immediate

evidence) and the narratives used to communicate,

which are not effective in changing behaviour (e.g.,

doomsday and alarmism). A more recent paper

(Brown et al., 2023) reported on the adoption of

mitigation and adaptation actions by individuals, such

as travelling by public transport and cycling, installing

solar panels, switching to high efficiency vehicles or

changing to a plant-based diet. However, the primary

motivation was economic rather than environmental.

Interpersonal communication may also play a

significant role in beliefs and feelings about climate

change. Goldberg et al. (2019), in a survey of a

nationally representative U.S. sample, found that

discussing global warming with friends and family

enhances knowledge on the topic, increases awareness

of the scientific consensus on human-driven causes,

and fosters further discussion and deeper engagement.

Regarding the family, it is worth highlighting the role

that children can play in transferring knowledge and

changing their parents' attitudes. Lawson et al. (2019)

conducted a study with 238 families over two years, in

which middle school children received a climate

change curriculum designed to promote

intergenerational learning. The discussion between

parents and children increased parents' concerns about

climate change, especially among more conservative

fathers. In fact, engaging in deliberative discussion is

considered one of the most effective climate change

education strategies (Monroe et al., 2019).

Generative AI (GenAI) brings new tools for

climate change communication. GenAI refers to a

class of artificial intelligence (AI) systems that can

generate new data based on existing data, often using

deep learning models. This means that the new content

is not copied from the training data but is based on

patterns and structures learned during the training

process. GenAI can be applied to text generation. A

prominent example is ChatGPT

, a large language

model (LLM) that has been trained to produce text

optimised for dialogue, using terabytes of data written

by humans, and obtained from different sources such

as websites, books and research articles. ChatGPT was

chosen for this study by its prevalent use

. In January

2025, ChatGPT had around 3.66 billion visits per

issues. Its trustworthiness is based on transparency

(explanations of data collection and analysis), reputation

and credibility. It is often cited by professionals and

companies looking for market trends.

CSEDU 2025 - 17th International Conference on Computer Supported Education

124

month. It is important to highlight that these LLM are

trained solely on predictive tasks that do not require an

understanding of meaning. This would involve linking

linguistic forms to communicative intents, that these

models currently are not equipped to do (Bender &

Koller, 2020). ChatGPT answers use natural language

processing techniques that makes the interaction

sound like human conversation, even though it is

artificial communication (Esposito, 2022). This

versatile tool is being used for many applications,

including the creation of stories, poetry, programming

code, marketing campaigns, in scientific writing,

translations and business predictions. It is also used as

an advanced search tool that presents the results in a

narrative way (Dwivedi et al., 2023).

Regarding climate, ChatGPT has the advantage of

accessing large amounts of information, from various

scientific disciplines such as atmospheric science,

oceanography and ecology, to quickly relate

seemingly independent information, in order to reach

conclusions (Biswas, 2023) and present the most

likely sequence of words based on its training data. In

this way, it can present relevant and contextualised

answers, synthesizing and explaining complex

concepts (Haluza & Jungwirth, 2023). One good

advantage is the capability of summarizing

information from long texts, identifying key points

and main themes to facilitate comprehension (Zhu et

al., 2023). Because ChatGPT can store input/output

responses, it enables continuous conversation

(Dwivedi et al., 2023), making the experience more

dynamic and interesting for the user. Immediate

feedback makes it possible to clarify and deepen the

topic in real time, as new questions arise. Due to its

interactive nature, ChatGPT has great potential for

active learning, as the user actively investigates the

problem. ChatGPT can also be part of a scalable,

tailored and automated climate communication,

combined with other communication tools in a

concerted intervention. It can be used to create articles

with reasons for climate action, providing informative

content on impacts and encourage behaviour change

(Nisbett & Spaiser, 2023).

A growing body of studies has highlighted the

relevance of using GenAI in the educational context.

In a Harvard Business Review article (Acar, 2023), the

author proposed the PAIR framework for using GenAI

in education, emphasising the need to use AI in

education, rather than prohibiting it. PAIR (Problem,

AI, Interaction, Reflection) is designed to proactively

integrate AI into students' curriculum, developing

skills such as the ability to formulate problems,

explore different AI tools, think critically and reflect

on AI results. For students, chatbots have been

suggested to improve learning and motivation, as they

can provide personalised learning experiences with

content tailored to their needs and learning style

(Kuhail et al., 2023). For teachers, GenAI can help to

address science education topics such as climate

change and be used as a tool for creating educational

resources. A relevant example is given by Cooper

(2023), who asked ChatGPT to create a teaching unit

on renewable energy for seventh graders based on the

5E instruction model (Bybee et al., 2006). He obtained

the design of a unit, divided into rubrics, with a

sequence of activities for each phase (engage, explore,

explain, elaborate, evaluate) and quizzes for a final

(self-)evaluation. Content like this can benefit teachers

with mixed views or limited understanding of climate

change. In the USA, where only about half the

population attributes climate change mainly to human

activity, a national survey of science teachers revealed

widespread gaps in training, knowledge, and

awareness of the scientific consensus on

anthropogenic global warming. This resulted in varied

classroom approaches: 54% taught that climate

change is mainly human caused, 31% presented mixed

messages, 10% denied the issue, and 5% avoided it

altogether (Plutzer et al., 2016).

It is important to recognize some limitations and

ethical concerns associated with ChatGPT. It can

generate inaccurate or false information, requiring

fact-checking with reliable sources (Zhu et al., 2023).

The quality of its output depends on the training data

and using content without scientific verification risks

perpetuating biases and inaccuracies. Therefore, it is

recommended to use the generated content as a first

draft, refined by specialists to ensure accuracy and

relevance (Salvagno et al., 2023). A major concern is

the risk of political bias. In several political orientation

tests, ChatGPT was classified as having left-leaning

political views. Political bias in a widely used tool can

be harmful for society, as it can impact human

perception and increase the spread of misinformation.

Ethical AI should present balanced, neutral arguments

based on available scientific evidence and not favour

political viewpoints (Rozado, 2023). There is also the

issue of plagiarism. When using ChatGPT we may not

recognise and give due credit to the authors of the

content (Cooper, 2023). Sometimes the tool is used to

aid decision making, but it can be inconsistent and

contradictory, yet people tend to trust and let their

judgement be influenced by ChatGPT advice (Krügel

et al., 2023). Trust in ChatGPT is a critical factor, as

highlighted by Choudhury and Shamszare (2023) in a

survey of 607 U.S. adults. Their findings suggest that

trust significantly impacts adoption, with blind trust

posing risks in decision-making and lack of trust

Generative AI in Climate Change Communication and Education

125

leading to underuse and missed opportunities. In

education, the trend towards acceptance by students is

noteworthy. A survey of university students in the

USA found that 89% used ChatGPT for schoolwork,

raising ethical concerns. While it can be used as a

pedagogical complement that increases students'

motivation and self-efficacy, it can also be used to

produce homework, projects and reports, which can

lead to more superficial learning and weaken critical

thinking, challenging teachers’ assessments (Yu,

2024). Another concern related to the origin of the

training data is ChatGPT’s ability to exclude

manipulative information (Zhu et al., 2023). To

mitigate this limitation and to ensure accuracy,

relevance and coherence, ChatGPT is trained in a

multi-stage process, using advanced machine learning

techniques. It is pre-trained on a large internet corpus,

fine-tuned with human supervision, to enhance the

conversation. It also uses a Reward Model

Construction, where the generated responses are

reviewed and ranked by humans. Finally, the model

uses a reinforcement learning technique called

Proximal Policy Optimisation, which uses the reward

model to maximise predicted quality and generate

knowledgeable and contextually appropriate

conversations (Zhu et al., 2023; OpenAI, 2023b). A

final ethical issue relates to the carbon impact of

ChatGPT due to its high energy consumption. GPT-3

training consumed approximately 190,000 kWh of

energy and produced 85,000 kg of CO2 (Quach,

2020). A study of the carbon impact of various GPT

models, based on GPT3 and GPT2, aimed to

determine the energy efficiency of each. The power

(watts) used for each query varied between 48.9 and

61.7 (Everman et al., 2023). As a technology that is

growing rapidly on the market, the environmental cost

cannot be ignored.

2 OBJECTIVES AND METHOD

This exploratory study examines whether ChatGPT

can generate information and explanations to support

climate communication, to help people understand

and act on climate change in an informed way,

targeting the general public and the youth.

The research team is part of 'BESIDE', an ERA

Chair in Research and Development in

Environmental Economics, which aims to promote

multidisciplinary research on topics related to climate

change, sustainability and socio-economic sciences.

One of the research areas is focused on science

communication and literacy of related topics. With

the rise of GenAI, the team aimed to explore its

impact on climate change communication and

education.

Using its web interface, we created a set of

generic prompts, designed to understand how

ChatGPT generates answers about climate change,

assuming that the general public and youth may rely

on it to obtain information:

 What is climate change?

 What are the real problems associated with

climate change?

 What can we do to reduce climate change?

 Is climate change a hoax?

We also used ChatGPT as a potential tool to

support teachers by asking for suggestions on how to

approach the topic with students. We selected the

elementary and high school levels as these grades are

identified as having most climate change education

interventions (Monroe et al., 2019). Climate Change

education should inform and prepare these

generations to understand, make lifestyle changes and

adaptations to reduce GhG emissions and other

ongoing and upcoming impacts (UNESCO, 2010).

The prompt was:

 How can I explain the concepts of climate

change, climate adaptation and climate change

mitigation with concrete examples to high

school students versus elementary school

students?

We conducted an initial session with ChatGPT-

3.5 in August 2023, as it was the most accessible

version but limited to pre-2021 data. In June 2024, we

repeated the questions with ChatGPT-4.0, a more

advanced version with improved data coverage,

though still constrained by session length and

response complexity. We then analysed differences

between versions and compared with scientific

publications and recent grey literature including

reports from IPCC (2023b, 2023a), OECD

(Dechezleprêtre et al., 2022) and UNESCO (2010).

The complete references used for interpretation of

results are available in the bibliography.

In the presentation of the results, ChatGPT

responses are presented in figures, to better

distinguish the generated text from the authors’ text.

The generated text was edited, cut to reduce the

length. We present ChatGPT-4 responses (July 2024),

as they are more recent, but for interpretation,

answers from ChatGPT-3.5 (August 2023) were also

consulted to verify consistency and evolution

(completed answers available by emailing the

authors).

CSEDU 2025 - 17th International Conference on Computer Supported Education

126

Responses were evaluated based on relevance

(whether the answers addressed the questions,

matched the required knowledge level, and were clear

for different audiences), completeness (whether they

covered key scientific facts), and scientific accuracy

(whether they aligned with peer-reviewed literature

and official reports). Overall, our findings indicate

that ChatGPT provides sufficiently relevant

information for the intended audiences. Due to the

characteristics of a LLM, repeating the questions to

ChatGPT will not result in the same answers.

3 RESULTS

Figure 1 to Figure 5 present the answers from

ChatGPT to the set of prepared prompts.

3.1 [Prompt 1] What Is Climate

Change?

Figure 1: Answer to prompt “What is Climate change?

The first question was asked to understand how

ChatGPT introduces climate change. The answer

defines the term, presents causes and a few

consequences, and briefly mentions the concepts of

mitigation and adaptation. The information is relevant

but not complete. Since the question was general, the

answer remained broad, making it suitable for an

initial approach. It also introduces terms like GHG

and explains the greenhouse effect, which may be

important for a better understanding and opens up

possibilities to be explored with new prompts. There

are some differences between answers from

ChatGPT-3.5 (2023) and ChatGPT-4 (2024). The

language in version 3.5 was simpler and attributed the

causes of climate change mainly to human activity,

whereas the new version begins by mentioning that

Climate Change can result from natural and human

activities. This may lead to believe that these have

equal importance, which is not accurate. However,

throughout the explanation, the focus is indeed on

anthropogenic causes. In 2023, the response included

a call to action, adaptation and mitigation measures

and a reference to the Paris Agreement, offering a

broader perspective. The 2024 version presents a

more direct answer to the question.

3.2 [Prompt 2] What Are the Real

Problems Associated with Climate

Change?

The second question intended to increase

understanding of the problem of climate change and

expand the first response, which ChatGPT did.

Figure 2: Answer to prompt “What are the real problems

associated with Climate Change?”

While the first answer focused on environmental

and health problems, the second repeated information

from the first answer, but went into more detail on the

economic, social and cultural aspects, all connected

with anthropogenic actions. The answer from

ChatGPT-3.5 had minimal differences (e.g., the

cultural impact was not mentioned, but all the other

points were presented in both answers, although

organised differently).

Generative AI in Climate Change Communication and Education

127

Figure 3: Answer to prompt “What can we do to reduce

Climate Change?”

3.3 [Prompt 3] What Can We Do to

Reduce Climate Change?

There were several differences between the 2023 and

2024 responses. In 2023, ChatGPT-3.5 provided a

shorter response with mostly global actions and only

one individual measure. It listed 13 actions without

distinguishing mitigation from adaptation, and some

points lacked detailed explanations.

In 2024, the ChatGPT-4 generated a longer list of

measures but focused on giving instructions without

explaining why they should be implemented.

Additionally, some points such as ‘Choose products

with lower carbon footprints’ were treated in a

superficial manner, which may make them difficult

for the public to understand due to a lack of context

or suitable explanation and may compromise

behaviour change and action.

3.4 [Prompt 4] Is Climate Change a

Hoax?

Figure 4: Answer to prompt “Is climate change a hoax?”

This question was asked with the intention of

understanding ChatGPT's responses to fake news and

climate change conspiracy theories. The answer

indicates that ChatGPT training attributes denial and

scepticism about climate change to misinformation,

vested interests and misunderstandings. It responded

based on scientific consensus, pointed to the evidence

to justify the veracity of the facts, and briefly

explained why there are groups that deny climate

change. Comparing responses from ChatGP-3.5 and

4, the latest version offers more developed and

accurate content, mentioning different scientific

CSEDU 2025 - 17th International Conference on Computer Supported Education

128

organisations and relevant scientific support and

provides ways to refute sceptical claims.

3.5 [Prompt 5] How Can I Explain the

Concepts of Climate Change,

Climate Change Adaptation and

Mitigation with Examples to High

School Students versus Elementary

School Students?

The answer supports initial learning of the topics

proposed by the teacher by presenting essential

content adapted to the students' schooling levels and

age groups, using appropriate language and

examples, aligned with contemporary theories of

cognitive development and learning. The 2024

version is again more developed, introducing

explanations, examples and discussion points on

climate change, adaptation and mitigation. It

concludes with pedagogical tips to tailor the

complexity and examples to the appropriate school

level. Limitations are addressed in the discussion.

4 DISCUSSION

In this study, we used ChatGPT to understand its

potential contribution to climate communication,

literacy and critical thinking, targeting the general

public and also focusing on young students and

teachers in the educational community. Some prompts

were created to emulate what the common citizen

could ask the free version of ChatGPT to get

information about climate change. Simple questions

with little context were deliberately asked to simulate

this type of usage. The results allowed a qualitative

evaluation of the relevance, accuracy and scientific

acumen of ChatGPT replies to climate related

questions and showed some aspects worth

highlighting.

The user can obtain clear and general information

about what climate change is and its main causes, but

also more specific information about the

consequences, adaptation and mitigation measures, as

well

as potential individual actions to take, related to

Figure 5: Answer to prompt 5.

this topic. Accessible language, selection of essential

information and clarity of content presentation are

benefits commonly associated with GenAI (Biswas,

2023; Zhu et al., 2023) and also mentioned as good

practices in various existing models of science

communication (Longnecker, 2016).

Generative AI in Climate Change Communication and Education

129

The information presented is correct and relevant,

with facts widely recognised in the scientific

community (e.g., IPCC, NOAA). One aspect that

caught our attention in Prompt 1 of ChatGPT-4 was

the way it included natural and human causes for this

phenomenon, but without developing the theme

equally for both, providing only additional content for

the anthropogenic related causes.

While the answer is not wrong, this may lead the

public to consider natural and human causes with

equal importance, which is not accurate, as the

acceleration of climate change is mainly due to human

activity (IPCC, 2023a). Nevertheless, in Prompts 4

and 5, the emphasis is again, and appropriately, only

focused on human action.

Analysing the set of information resulting from

Prompts 1 to 4 in terms of completeness, the

information is sufficient for the objectives and the

target, as a very detailed approach could overwhelm

and demotivate this public. Regarding climate

communication, Stoknes (2014) advises not to

overwhelm people with the catastrophic scenarios but

instead focus on a positive message of a greener

future, in a smart society, using less resources and

having a better quality of life. To understand

perceptions of climate change and the factors

influencing behaviour change, Dechezleprêtre et al.

(2022) conducted a survey of 40000 respondents from

20 countries. They concluded that support for climate

actions depends on beliefs about the effectiveness of

the policy (in reducing emissions), concerns about

inequality (impact on low-income households) and

personal interests (impact on household). Explaining

climate measures and their benefits to the citizens

increases the willingness to support them, while

simply informing about the impacts of climate change

seems to be ineffective. But, as concluded by

Robertson (2022, p. 57) ‘Climate change is a difficult

news topic to cover, and it is not clear that there is a

one-size-fits-all approach to it’.

When addressing measures towards climate

mitigation (i.e., enhance decarbonisation by reducing

the GhG emissions) and/or adaptation (i.e., take

actions to minimise present and future impacts)

both

collective and individual measures were mostly

presented in an instructional way, indicating what

needs to be done without justifying why the measure

is needed or how it impacts the overall goal, which

may not be motivating enough to change behaviour.

Measures are presented without distinguishing

degrees of importance, but they clearly have different

European Environmental Agency - https://www.eea.

europa.eu/en/about/contact-us/faqs/what-is-the-

difference-between-adaptation-and-mitigation

impacts, which would be interesting for the public to

understand. Giving the same relevance to all the

measures may lead to distrust or confusion among

individuals (Nerlich et al., 2010).

Another issue to underline relates to the content

of the reply to Prompt 4. We questioned the veracity

of climate change, and the answer reinforces the fact

that ChatGPT has been trained to not promote fake

news or respond/apply to conspiracy theories, related

to the theme. Additional prompts were made to test

the robustness of ChatGPT regarding climate change

conspiracies (1- Regarding climate change, it is too

late to do something. 2 - Why is the Antarctic Sea ice

increasing? 3 - Plants requires CO

, thus fossil fuel

emissions will be beneficial for plants? – complete

answers available by email), and it seems to be

robustly trained in this theme. This may be related

with the algorithm's automatic filters, the training of

humans to improve the quality of the answers and the

pre-training text corpus (OpenAI, 2023b), as the field

of climate science has a substantial number of state-

of-the-art summaries and assessment reports publicly

available (Schäfer, 2012). As a tool that evolves with

the corpus of text available on the internet, it is

necessary to ensure that the programming and

training techniques exclude misinformation and offer

scientifically correct information (OpenAI, 2023a;

Raman et al., 2024; Montoro-Montarroso et al.,

2023).

This represents an advantage in the current digital

and social media era, where citizens use them as

quick access to information and most trust the content

provided, as contradicting fake news circulating on

social networks is of uttermost importance to

education and awareness on climate change. In

contrast, it's important to recognise that ChatGPT can

be exploited by malicious actors to spread false

climate change information and manipulate public

opinion. Like other GenAI tools, its ability to rapidly

generate large volumes of text makes it easier to flood

online platforms with misleading, persuasive, and

deceptive content (Goldstein et al., 2023).

Regarding the use of ChatGPT for educational

purposes, a more elaborate prompt [Prompt 5] was

created to address the topic of climate change, as well

as adaptation and mitigation measures, at different

levels of education. The content of the answer is

relevant and respects the students’ ages, presenting

appropriate language and examples depending on

whether it was elementary or high school education.

CSEDU 2025 - 17th International Conference on Computer Supported Education

130

In terms of development and learning, elementary

school children (7-11 years) are beginning to think

logically, but they are still very attached to the

concrete and have difficulty understanding abstract

concepts. Therefore, the content of teaching should be

simple, using examples from everyday life and

practical activities (Piaget, 1952). In secondary

school, students are developing the capacity for

abstract, hypothetical-deductive and systematic

thinking. An appropriate lesson can include critical

analysis of concepts, problem solving, debates and

abstract and ethical questions (Piaget & Inhelder,

1969). At both levels of education, students learn best

through interaction with peers and teachers

(Vygotsky, 1978), so discussion is essential. The

ChatGPT response considers the developmental

levels of the age groups concerned. For elementary

education, it includes simple and concrete

explanations (e.g., ‘Adaptation means changing how

we live so we can handle the changes in the weather’),

with real examples (e.g., ‘In cities near the ocean,

people are building big walls to keep the water out’)

and discussion. For high school, it introduces more

complex terms and examples (e.g., ‘gases like carbon

dioxide (CO2) and methane (CH4) trap heat in the

Earth’s atmosphere’), also encouraging discussion

(e.g., ‘Discussion Point: different countries

adaptation strategies on their specific vulnerabilities

and needs’). One point that could be improved in the

high school response would be to encourage

collaborative learning by suggesting group projects

that promote the exchange of ideas, as well as the use

of research tools, such as books and technology, that

help to develop critical thinking and autonomy

(Johnson & Johnson, 1987).

ChatGPT's response also considers the evolution

of learning over time. For Bruner, all concepts can be

taught to younger children as long as the content is

appropriate to their level of development, and it can

be deepened in the future (Wood et al., 1976). The

role of the adult is to provide scaffolding for the

child's construction of knowledge, interacting with

them and helping them to progress to higher levels.

Ausubel (2003) reiterates the importance of prior

knowing in the construction of knowledge. For

meaningful learning to occur, new information must

relate to the knowledge the child has already

acquired. Content gains meaning when it is learned in

interaction with relevant concepts that already exist in

the learner's cognitive structure. There is then a

process of knowledge modification. Meaningful

learning facilitates the retention and use of

information in other learning. Even if it is forgotten,

it is easier to relearn. In ChatGPT's response, there is

an evolution of content, with increasing complexity

from elementary to secondary education, respecting

these highly accepted theories of learning.

However, as the answer is too concise, it only

serves as a starting point for the teacher to organise

ideas and structure an educational plan or activity,

reinforcing the need to complement it with other

resources. As Haluza and Jungwirth (2023) suggest,

it is important to formulate the questions accurately,

to provide sufficient context, and to indicate the

specifics intended in the answer, such as length,

preferred structure, and key content. Writing effective

prompts to get the desired results requires some

thought (Lin, 2023). It is important to use an

elaborate, specific prompt to achieve the defined goal.

It is also possible to regenerate answers and give

feedback on the answer (thumbs up and down) to

improve results. A way to get a more structured

response could be to ask ChatGPT to generate a

hierarchical table of contents and then elaborate on

each point in subsequent iterations. In our case, one

way to overcome the limited answer obtained would

be to continue the dialogue in subsequent prompts,

asking ChatGPT to elaborate on each point and to

supplement it with teaching and learning activities.

However, this could still result in a limited

pedagogical activity, as ChatGPT being a LLM lacks

human skills such as creativity, empathy and critical

thinking. Therefore, it is advisable to use it mostly as

a complementary tool. Once used, it is essential that

teachers critically evaluate and adjust the content

generated to their objectives (Kooli, 2023). Cooper

reached a similar conclusion. After generating a

teaching unit with ChatGPT, available in Cooper

(2023), he found it to be a valuable tool for educators

in designing science units, rubrics and quizzes, as

long as they review AI-generated content to ensure

alignment with their educational context. When

entering the level of detail, it is important to verify

ChatGPT responses, either with expert support or by

consulting other sources, to ensure accuracy and

consistency (Salvagno et al., 2023). Furthermore, for

educational purposes ChatGPT can be complemented

with creative GenAI apps to create images, like

DALL-E, LENSA or similar, in addition to standard

teaching support material.

5 CONCLUSION

In addition to being a global environmental, economic

and social challenge, climate change is also a major

communication challenge. Traditional mass media do

not seem to have a consistent communication

Generative AI in Climate Change Communication and Education

131

approach on the issue, contradictory messages are

circulating on social media, and scientists are

struggling to convince people with environmental

evidence.

To answer que question ‘Can Generative AI

support people’s understanding and decrease barriers

to climate-related communication?’ we created a set

of prompts, with different degrees of depth and

complexity. The responses, taken together, provide

clear, accurate and comprehensive information that

can effectively serve as a starting point for relevant

climate change communication for general and young

education audiences. ChatGPT answers applied good

scientific background information and

communication practices and used them within its

limitations to provide useful content. It also seems to

address some of the barriers identified in climate

communication, such as alarmist and sensationalist

reporting or the proliferation of subjective opinions

and misinformation on social media.

We have noticed some differences between the

ChatGPT-3.5 and ChatGPT-4 free versions, affecting

the answers from 2023 to 2024. These changes may be

because ChatGPT-3.5 is not as updated as ChatGPT-

4, but also to differences in the algorithm of each

version. ChatGPT-4's answers were more literal to the

question but also provided more scientific support,

while ChatGPT-3.5 provided more contextual

feedback. This may indicate that the newer version

requires more elaborate prompts to provide further

contextualised information or is more oriented

towards developing an ongoing dialogue to refine

replies. Nevertheless, by analysing the responses

received in the space of a year, we consider that the

free versions of ChatGPT can be a valuable tool to

support people's understanding and decrease barriers

to climate-related communication. ChatGPT may be

useful as a first approach to the topic, to synthesise

concepts, provide a backbone structural framework

for the topic and clarify initial doubts, mostly at a

general level.

For the general public, ChatGPT answers provide

a first understanding of the issue, as it briefly and

clearly explains what climate change is, its main

causes, consequences, adaptation and mitigation

measures, and possible individual actions. The

information is relevant and presented in accessible

language.

For teachers, the generated response is also

relevant and appropriate to the defined school levels,

albeit too brief and insufficient at a first level of

iteration. It is recommended that teachers use

ChatGPT as a first approach to organise ideas and

structure educational plans or activities,

complementing it with other resources and further

iterations to ensure accuracy, completeness of detailed

information and adaptation to specific contexts.

Some prompts led to answers that were too brief

or general. In this study, we defined a set of prompts

and analysed the answers without editing them.

Editing the prompts could have made it difficult to

determine when to stop, making the interpretation

inaccurate. Nevertheless, improving the prompts to

obtain better responses is a new skill associated with

using GenAI text tools that needs also to be acquainted

and integrated into learning experiences at all

educational levels. For more complete information,

questions must provide context by specifying the

purpose, intended audience, desired length and format

of the answer.

A major advantage of this tool over social media

is that its learning database and programming

algorithm seem to exclude fake news on climate

change. However, given the characteristics of

ChatGPT, it is still advisable not to rely entirely on the

information generated and to fact-check the responses

and request for scientific literature citations to increase

robustness of the content. When creating content for

climate communication, we recommend using GenAI

as an assistant, without excluding human expertise and

judgement, and other sources consultation, as

ChatGPT, being an LLM, lacks empathy and critical

thinking, and its learning data may be biased.

The results of our study are aligned with the PAIR

(Problem, AI, Interaction, Reflection) methodology.

In fact, these same principles shall be applied to all

these types of interactions with GenAI, where the user

intends to obtain relevant, validated and consistent

information or feedback from an AI bot, to leverage

its time and capacities in dealing with such complex

issues as climate change.

Our study showcases that ChatGPT can facilitate

communication and provide information on climate

change to support people's understanding and

willingness to adopt more climate-friendly practices.

The academic community can play a significant role

in the continuous improvement of the model's

knowledge base, ensuring the provision of scientific

accurate information, in line with current scientific

consensus. The availability of peer-reviewed scientific

research through open access, following the principles

of FAIR principles (findability, accessibility,

interoperability, reusability) can provide a diverse and

comprehensive set of reliable information,

contributing to increase the quality of the answers

generated and positioning ChatGPT as a valuable tool

for climate communication. New avenues of research

can be opened now, to quantitatively test and validate

relevance, perceptions and usages of these types of

information generated by AI in the different

CSEDU 2025 - 17th International Conference on Computer Supported Education

132

stakeholders’ groups to evaluate its adoption for

innovative climate change mitigation solutions.

FUNDING

This work was funded by the ERA Chair BESIDE

project financed by the European Union’s Horizon

2020 under Grant Agreement ID: 951389, DOI:

10.3030/951389.

AKNOWLEDGMENTS

The authors acknowledge financial support to

CESAM by FCT/MCTES(UIDP/50017/2020 +

UIDB/50017/2020 + LA/P/0094/2020) and CIDTFF

by FCT/MCES(UIDB/00194/2020), through national

funds.

REFERENCES

Acar, O. (2023). Are Your Students Ready for AI?: A Four-

Step Framework to Prepare Learners for a ChatGPT

World? Harvard Business Publishing Education.

https://tinyurl.com/y4n9tj3c

Alinejad, D., & Van Dijck, J. (2023). Climate

communication: How researchers navigate between

scientific truth and media publics. Communication and

the Public, 8(1), 29–44.

https://doi.org/10.1177/20570473221138612

Ausubel, D. P. (2003). Aquisição e Retenção de

Conhecimentos: Uma Perspectiva Cognitiva. Plátano

Editora.

Bender, E. M., & Koller, A. (2020). Climbing towards

NLU: On meaning, form, and understanding in the age

of data. Proceedings of the 58th Annual Meeting of the

Association for Computational Linguistics, 5185–5198.

Biswas, S. S. (2023). Potential Use of Chat GPT in Global

Warming. Annals of Biomedical Engineering, 51(6),

1126–1127. https://tinyurl.com/w4w5cra8

Brown, H., Austhof, E., Luz, P., & Ferguson, D. (2023).

Economics, health, or environment: What motivates

individual climate action. PLOS Clim, 2(8).

Bushell, S.; Buisson, G. S.; Workman, M.; Colley, T.

(2017). Strategic narratives in climate change: Towards

a unifying narrative to address the action gap on climate

change. Energy Research & Social Science, 28, 39–49.

Bybee, R. W., Taylor, J. A., Gardner, A., Van, P., Powell,

J. C., Westbrook, A., Landes, N., Spiegel, S., Stuhlsatz,

M. M., Ellis, A., Thomas, H., Bloom, M., Moran, R.,

Getty, S., & Knapp, N. (2006). The BSCS 5E

Instructional Model : Origins and Effectiveness.

Choudhury, A., & Shamszare, H. (2023). Investigating the

impact of user trust on the adoption and use of

ChatGPT: Survey analysis. Journal of Medical Internet

Research, 25.

Cook, J., Nuccitelli, D., Green, S. A., Richardson, M.,

Winkler, B., Painting, R., Way, R., Jacobs, P., & Skuce,

A. (2013). Quantifying the consensus on anthropogenic

global warming in the scientific literature.

Environmental Research Letters, 8(2).

https://doi.org/10.1088/1748-9326/8/2/024024

Cooper, G. (2023). Examining Science Education in

ChatGPT: An Exploratory Study of Generative

Artificial Intelligence. Journal of Science Education

and Technology, 32(3), 444–452.

https://doi.org/10.1007/s10956-023-10039-y

Dechezleprêtre, A., Fabre, A., Kruse, T., Planterose, B.,

Sanchez Chico, A., & Stantcheva, S. (2022). Fighting

Climate Change: International Attitudes Toward

Climate Policies. In OECD Economics Department

Working Papers (1714) https://tinyurl.com/3pnm3thr

Dwivedi, Y. K., Kshetri, N., Hughes, L., Slade, E. L.,

Jeyaraj, A., Kar, A. K., Baabdullah, A. M., Koohang,

A., Raghavan, V., Ahuja, M., Albanna, H., Albashrawi,

M. A., Al-Busaidi, A. S., Balakrishnan, J., Barlette, Y.,

Basu, S., Bose, I., Brooks, L., Buhalis, D., … Wright,

R. (2023). “So what if ChatGPT wrote it?”

Multidisciplinary perspectives on opportunities,

challenges and implications of generative

conversational AI for research, practice and policy.

International Journal of Information Management,

71(March). https://tinyurl.com/49j36hzp

Esposito, E. (2022). Artificial communication (M. Press,

Ed.).

Everman, B., Villwock, T., Chen, D., Soto, N., Zhang, O.,

& Zong, Z. (2023). Evaluating the carbon impact of

large language models at the inference stage. IEEE

International Performance, Computing, and

Communications Conference (IPCCC), 150–157.

Goldberg, M., van der Linden, S., Maibach, E., &

Leiserowitz, A. (2019). Discussing global warming

leads to greater acceptance of climate science.

Proceedings of the National Academy of Sciences,

116(30), 14804–14805.

Goldstein, J. A., Sastry, G., Musser, M., DiResta, R.,

Gentzel, M., & Sedova, K. (2023). Generative

Language Models and Automated Influence

Operations: Emerging Threats and Potential

Mitigations. http://arxiv.org/abs/2301.04246

Haluza, D., & Jungwirth, D. (2023). Artificial Intelligence

and Ten Societal Megatrends: An Exploratory Study

Using GPT-3. Systems, 11(3).

https://doi.org/10.3390/systems11030120

Horta, A., & Carvalho, A. (2017). Climate change

communication in Portugal. Oxford Research

Encyclopedia of Climate Science, 1–32.

IPCC. (2023a). AR6 Synthesis Report: Climate Change

2023. https://www.ipcc.ch/report/ar6/syr/

IPCC. (2023b). Summary for policymakers. In Climate

Change 2023: Synthesis Report. A report of the

Intergovernmental Plane on Climate Change.

Johnson, D. W., & Johnson, R. T. (1987). Learning

Together and Alone: Cooperative, Competitive, and

Individualistic Learning. Prentice-Hall, Inc.

Generative AI in Climate Change Communication and Education

133

Kooli, C. (2023). Chatbots in Education and Research: A

Critical Examination of Ethical Implications and

Solutions. Sustainability (Switzerland), 15(7).

https://doi.org/10.3390/su15075614

Krügel, S., Ostermaier, A., & Uhl, M. (2023). ChatGPT’s

inconsistent moral advice influences users’ judgment.

Scientific Reports, 13(1), 1–5. https://doi.org/10.1038/

s41598-023-31341-0

Kuhail, M. A., Alturki, N., Alramlawi, S., & Alhejori, K.

(2023). Interacting with educational chatbots: A

systematic review. In Education and Information

Technologies (Vol. 28, Issue 1). Springer US.

https://doi.org/10.1007/s10639-022-11177-3

Lawson, D., Stevenson, K., Peterson, M., Carrier, S.,

Strnad, R., & Seekamp, E. (2019). Children can foster

climate change concern among their parents. Nature

Climate Change, 9(6), 458–462.

Lin, Z. (2023). Why and how to embrace AIsuch as

ChatGPT in your academic life. Royal Society Open

Science, 10(23658). https://tinyurl.com/377vkf7s

Liu, J. C.-E. (2023). Public opinion on climate change in

China—Evidence from two national surveys. PLOS

Climate, 2(2). https://tinyurl.com/hw7w345c

Longnecker, N. (2016). An integrated model of science

communication-More than providing evidence. Journal

of Science Communication, 15(5), 1–13.

https://doi.org/10.22323/2.15050401

Meyer, H., Peach, A. K., Guenther, L., Kedar, H. E., &

Brüggemann, M. (2023). Between Calls for Action and

Narratives of Denial: Climate Change Attention

Structures on Twitter. Media and Communication,

11(1), 278–292. https://tinyurl.com/p8h9pmky

Monroe, M., Plate, R., Oxarart, A., Bowers, A., & Chaves,

W. (2019). Identifying effective climate change

education strategies: a systematic review of the

research. Environmental Education Research, 25(6),

791–812.

Montoro-Montarroso, A., Cantón-Correa, J., Rosso, P.,

Chulvi, B., Panizo-Lledot, Á., Huertas-Tato, J., Calvo-

Figueras, B., Rementeria, M. J., & Gómez-Romero, J.

(2023). Fighting disinformation with artificial

intelligence: fundamentals, advances and challenges.

Profesional de La Información, 32(3).

https://tinyurl.com/yw5b2v8y

Nerlich, B., Koteyko, N., & Brown, B. (2010). Theory and

language of climate change communication. Wiley

Interdisciplinary Reviews: Climate Change, 1(1), 97–

110. https://doi.org/10.1002/wcc.2

Newman, N., Fletcher, R., Robertson, C. T., Eddy, K., &

Nielsen, R. K. (2022). Digital News Report 2022.

https://tinyurl.com/fkpjctd2

Newman, T. P. (2017). Tracking the release of IPCC AR5

on Twitter: Users, comments, and sources following the

release of the Working Group I Summary for

Policymakers. Public Understanding of Science, 26(7),

815–825. https://tinyurl.com/45pk4yzt

Nisbett, N., & Spaiser, V. (2023). How convincing are AI-

generated moral arguments for climate action?

Frontiers in Climate, 5. https://tinyurl.com/yemt4e34

OpenAI. (2023a). ChatGPT-4. Online. https://openai.com/

index/gpt-4-research/

OpenAI. (2023b). ChatGPT-4 System Card. Online.

https://cdn.openai.com/papers/gpt-4-system-card.pdf

Paris Agreement. (2015). Paris Agreement. In Report of the

conference of the parties to the U. Nations framework

convention on climate change (21st session).

Piaget, J. (1952). The Origins of Intelligence in Children.

International Universities Press.

Piaget, J., & Inhelder, B. (1969). The Psychology of the

Child. Basic Books.

Plutzer, E., McCaffrey, M., Hannah, A. L., Rosenau, J.,

Berbeco, M., & Reid, A. H. (2016). Climate confusion

among U.S. teachers. Science, 351(6274), 664–665.

https://doi.org/10.1126/science.aab3907

Quach, K. (2020). AI me to the moon… Carbon footprint fr

‘training GPT-3’ sames as driving to our natural

satellite and back. Online. https://tinyurl.com/

59392npk

Raman, R., Nair, V. K., Nedungadi, P., Sahu, A. K.,

Kowalski, R., Ramanathan, S., & Achuthan, K. (2024).

Fake news research trends, linkages to generative

artificial intelligence and sustainable development

goals. Heliyon, 10(3). https://tinyurl.com/2298kn9z

Robertson, C. T. (2022). How People access and Think

about climate change. In Digital News Report 2022 (pp.

54–57). https://tinyurl.com/fkpjctd2

Rozado, D. (2023). The political biases of chatgpt. Social

Sciences, 12(3), 148.

Salvagno, M., Taccone, F. S., & Gerli, A. G. (2023). Can

artificial intelligence help scientific writing? Critical

Care, 27(1), 1–5. https://tinyurl.com/yatn9s32

Schäfer, M. S. (2012). Online communication on climate

change and climate politics: A literature review. Wiley

Interdisciplinary Reviews: Climate Change, 3(6), 527–

543. https://doi.org/10.1002/wcc.191

Stoknes, P. E. (2014). Rethinking climate communications

and the “psychological climate paradox.” Energy

Research and Social Science, 1, 161–170.

https://doi.org/10.1016/j.erss.2014.03.007

UNESCO. (2010). Climate change education for

sustainable development. The UNESCO Climate

Change Initiative. With Assistance of Julia Heiss.

https://tinyurl.com/3rrwdmx3

Vygotsky, L. S. (1978). Mind in Society: The Development

of Higher Psychological Processes. Harvard University

Press.

Wibeck, V. (2014). Enhancing learning, communication

and public engagement about climate change - some

lessons from recent literature. Environmental

Education Research, 20(3), 387–411. https://doi.org/

10.1080/13504622.2013.812720

Wood, D., Bruner, J. S., & Ross, G. (1976). The Role of

Tutoring in Problem Solving. Journal of Child

Psychology and Psychiatry, 17(2), 89–100.

Yu, H. (2024). The application and challenges of ChatGPT

in educational transformation: New demands for

teachers’ roles. Heliyon, 10. https://tinyurl.com/

4p34ehka

Zhu, J. J., Jiang, J., Yang, M., & Ren, Z. J. (2023). ChatGPT

and Environmental Research. Environmental Science

and Technology, 1–4. https://tinyurl.com/5n6d49yw

CSEDU 2025 - 17th International Conference on Computer Supported Education

134