AI and Sustainable Tourism – What's on the Scales?

Sunday, 26 July 2026 07:21

Nowadays, it's rare to attend a tourism conference without hearing that artificial intelligence is fundamentally transforming the industry's operations. A few presentations later, the discussion often turns to how AI consumes a lot of energy, requires water for cooling data centres, and continuously increases the demand for IT infrastructure. Both statements are true. Artificial intelligence does have an environmental impact, but it also provides access to knowledge and tools that were previously unaffordable for many tourism businesses. The real question, therefore, is not whether AI is good or bad. It's much more about how we use it, what data it works with, and the goals we set for it.

During the development of I-DEST, we also work within this duality. We create digital solutions that utilise artificial intelligence while aiming to support more sustainable operations for destinations and tourism providers. This is why we cannot avoid the question of what price we pay for using technology and what we receive in return.

What does all this cost the planet?

According to calculations by the International Energy Agency (IEA), the world's data centres consumed approximately 415 terawatt-hours of electricity in 2024. This accounted for about 1.5% of global electricity consumption. Under the baseline scenario, consumption could approach 945 terawatt-hours by 2030, more than doubling and representing nearly 3% of the world's total electricity use. (IEA, 2025)

One of the main drivers of this growth is AI, although data centres, of course, do not exclusively serve artificial intelligence systems. Cloud services, online videos, business IT systems, and many other digital services also require significant capacity.

The environmental impact of a single AI query, however, may be much smaller than earlier general estimates suggested.

According to operational measurements published by Google in August 2025, serving an average, or more precisely, median Gemini text query:

used 0.24 watt-hours of energy,
resulted in 0.03 grams of CO2 equivalent emissions,
and required approximately 0.26 millilitres of water.

The energy consumption was roughly equivalent to nine seconds of television viewing, and the water usage was about five drops. (Elsworth et al., 2025)

However, these data should be treated with caution. The measurement pertained to a specific service, Gemini text queries, and cannot be automatically applied to all AI systems, complex research tasks, or image and video generation. Furthermore, the calculation does not fully reflect the environmental burden of model development, training, hardware manufacturing, and data centre construction. Google's results have also been criticised by independent experts, as the method of calculating indirect water use and emissions can significantly influence the final outcome.

Thus, the two statements do not contradict each other. The impact of a single short query may be negligible, but billions of queries are not. Systems are becoming increasingly efficient, but their usage is also spreading rapidly. As a result, total energy and water demand may increase even if less resource is required for individual tasks.

Moreover, with water usage, it is not just the global quantity that matters. The location of the data centre is also important. The same water demand has very different consequences in a rainy region compared to an area regularly struggling with water shortages.

The lifecycle of hardware cannot be ignored either. Manufacturing high-performance chips requires raw materials, water, and energy, and the rapid replacement of devices increases the amount of electronic waste.

What do we gain in return?

The benefits of artificial intelligence in tourism are not necessarily most interesting at large international companies. It can bring much more noticeable changes for smaller providers. A small guesthouse, family restaurant, or local event organiser often did not previously communicate in multiple languages, regularly analyse guest reviews, or create detailed sustainability reports because they lacked the necessary staff or external expertise.

Today, some of these tasks can be performed with just a laptop. AI does not replace professional expertise but can significantly reduce the cost and time required for the initial steps.

One important area of social benefit is accessibility. With the help of artificial intelligence, it is easier to create:

real-time or quick translations;
information written in simple language;
readable content;
image descriptions for visually impaired people;
information tailored to various needs;
and overviews compiled from accessibility data.

These tasks were often neglected in the past because providers could not afford them.

In tourism, which struggles with labour shortages, automating administrative tasks can also be beneficial. It does not necessarily replace staff but can free up time. Less time is spent filling out spreadsheets, drafting repetitive emails, or organising data, leaving more time for guests.

However, this is only true if the business leadership genuinely uses the technology for this purpose. AI is equally capable of supporting employees, increasing surveillance, or reducing staff numbers. The technology itself does not decide which direction we choose.

There is also great potential in presenting cultural heritage. Local history collections, old photographs, ethnographic recordings, dialect materials, and lengthy monographs can become searchable. A town's history can thus be incorporated into a thematic walking route, an audio guide, or a school programme.

However, the literature is much more cautious about the sustainability benefits of AI than the marketing materials of technology companies. Gössling and Mei's 2025 review highlights that while much is said about the possibilities, there is still relatively little reliable evidence of actual, long-term measurable sustainability outcomes. The authors also warn that AI can exacerbate data concentration, consumer surveillance, and the intensification of tourism (Gössling–Mei, 2025).

A systematic review of 213 scientific publications found that tourism research on intelligent automation primarily focuses on visitor experiences, heritage preservation, quality of life, measuring visitor experiences, and environmental protection. However, practical measurement of environmental impacts has received less attention compared to economic and social issues (Majid et al., 2023).

Marketing as the first point of contact

Most tourism businesses first encounter artificial intelligence in marketing. They use it for writing texts, translations, creating social media posts, newsletters, generating images, and responding to guest reviews. The reason for this is understandable. The results are immediately visible, usage is relatively simple, and usually, no special IT development is required.

From a sustainability perspective, however, marketing is much more significant than just speeding up the creation of a Facebook post. Communication can have a major impact on who visits a place, when they travel, how long they stay, which attractions they visit, and where they spend their money.

A well-structured spring or autumn campaign, for example, can reduce the pressure of the summer peak season. Introducing a lesser-known part of a town or a local business can help distribute visitors more evenly. AI enables smaller providers or tourism organisations to create tailored messages for various target groups. Different content can be aimed at those who prefer quieter periods, families with young children, cyclists, or those seeking local foods and producers.

Multilingual communication can also become cheaper and faster. This is particularly useful for smaller source markets that previously lacked a dedicated marketing budget.

AI can also assist in showcasing lesser-known attractions, local stories, and smaller providers. Many destination websites feature detailed descriptions of the most famous attractions, while smaller assets barely appear. Technology can speed up the processing of information, but local knowledge and authentic stories still need to be added by people.

The key risks of marketing

Everything starts to sound the same

If every provider uses the same AI tool with similar instructions, soon every accommodation will be "cosy," every town a "hidden gem," and every programme will "offer an unforgettable experience." However, in tourism, the character of the place is one of the greatest values.

This is why it is worth creating a unique linguistic guide. This can include expressions characteristic of the place, the desired tone, clichés to avoid, and the stories that truly distinguish the service.

Generated images do not replace reality

An image showing a non-existent room, pool, view, or landscape is not just a creative solution. It can mislead guests, cause consumer protection issues, and damage trust. Generated images can be used as illustrations or mood elements, provided this is clearly indicated. However, for showcasing a service that is actually available, real photographs must be used.

Greenwashing can also become faster

Today, it takes only a few minutes to create a text about sustainability. However, measuring operations still requires data.

If an environmental claim is not backed by measurements, documentation, or verifiable results, AI does not make the business more sustainable but merely produces unsubstantiated green messages faster. This poses both business and legal risks. Environmental claims are increasingly scrutinised, so every communicated result must be verifiable and precisely formulated.

The system reinforces the goals we set

If a campaign focuses solely on increasing bookings or immediate revenue, it is likely to further strengthen the most popular periods and products. These are the easiest to achieve good results with.

Sustainability will only appear in decisions if it is included among the metrics.

Such metrics could include:

the proportion of off-peak bookings;
average length of stay;
spending at local businesses;
traffic at less-visited locations;
the proportion of guests arriving by public transport;
or energy and water consumption per guest night.

The same AI system can provide entirely different suggestions if it has to consider these aspects alongside the number of bookings.

Where AI is already working well today

Demand forecasting and capacity planning

Based on past booking data, event calendars, weather, transport information, and other factors, more accurate forecasts can be made about expected demand.

This can help with:

staff scheduling;
inventory planning;
adjusting transport capacities;
creating visiting time slots;
and identifying peak times early.

Forecasting does not eliminate overcrowding but can make problems visible earlier, allowing more time for intervention.

Reducing food waste

AI-based image recognition and measurement systems can identify which foods generate the most waste and during which periods. This data helps kitchens adjust procurement, portion sizes, and menus. In Hilton's 2025 Green Ramadan programme, 45 hotels across 14 countries participated. During the programme, the amount of food left on plates decreased by 26%, avoiding over 2.6 tonnes of food waste. The initiative also utilised Winnow's AI-based measurement technology (Hilton, 2025). This is a good example of when environmental and business interests align directly: less food ends up as waste, while procurement costs are also reduced.

Energy and water usage

Intelligent systems can help regulate heating, cooling, and ventilation. They can detect unusual consumption, leaks, malfunctioning equipment, or devices running unnecessarily.

The goal is not to compromise guest comfort but to ensure the system does not use energy where and when it is not needed.

Preparing sustainability reports and certifications

A sustainability certification or audit involves significant documentation. Policies, invoices, consumption data, photos, and other evidence must be collected, organised, and linked to specific requirements.

AI helps not by "greenwashing" the provider but by recognising and organising available evidence, highlighting missing data, and supporting compliance tracking.

The final compliance decision, however, must still be made by a qualified professional or auditor.

When AI makes things worse

One issue with AI recommendations is that they often make already popular places even more visible.

Lee and Pennington-Gray's research pre-study, based on one million modelled domestic trips in the US, found that the examined large language models created travel flows that were more seasonal, uneven, and less diverse than actual tourism patterns. The systems particularly reinforced the pairing of already popular destinations and periods (Lee–Pennington-Gray, 2026).

Since this is a pre-study, the findings need to be confirmed by further research, but the identified risk deserves attention.

If many people seek advice from the same few systems, and these systems consistently recommend the same places, it could further increase overcrowding.

The other problem is invisibility.

Chen, Weng, and Moro's 2026 pre-study examined restaurant recommendations by three major language models in five US cities across 304 neighbourhoods. Even when the models could choose from a verified list of real establishments, 47.5% of the places were never recommended. Of the businesses that remained invisible, 31.9% were the same across all three model families (Chen–Weng–Moro, 2026).

This shows that AI does not simply aggregate available information. It also selects from it. Some places are regularly prioritised, while others are barely or never shown.

This can particularly disadvantage smaller, less-reviewed, or weaker digital presence local businesses.

What can travellers do?

Don’t just ask where to go

Also ask when it’s best to travel and which nearby alternative is less crowded. A useful question could be:

“When is this place least crowded, and what similar programme can I find nearby?”

Often, we do more for sustainability by travelling to the same place at a different time than by seeking another distant destination.

Check practical information

Always verify opening hours, schedules, prices, booking conditions, and certificate validity at the original source. AI can convincingly present outdated or incorrect data.

Plan fewer activities

If the itinerary suggests ten places for a single day, ask it to create three or four instead.

Fewer locations generally mean less local travel, a more relaxed programme, and more genuine encounters. In terms of environmental impact, transportation usually has a much greater footprint than the few AI queries used to plan the trip.

Question sustainability claims

When accommodation or a programme claims to be sustainable, it’s worth asking: Does it have an independent certification? Who issued it? How long is it valid? What data does it measure? What specific results has it achieved?

There is a big difference between a certification issued based on an independent set of requirements and a business’s self-created “green” label.

What can tourism businesses do?

First, get the data right

AI can only work with what is digitally accessible and interpretable.

If the website has inaccurate opening hours, lacks an address or location, does not include accessibility information, and the service descriptions are unclear, then AI systems will also provide incorrect or incomplete information.

Particularly important are:

accurate name, address, and contact details;
current opening hours;
prices and booking conditions;
transport options;
accessibility information;
appropriate categorisation of the service;
and structured data on the website.

Well-maintained, machine-readable data is no longer just a matter of search engine optimisation. It can also be a prerequisite for appearing in AI-based recommendations.

Measure first, then communicate

Energy, water, waste, local procurement, guest nights, transport. These are the basis for calculating real sustainability outcomes.

AI can help organise data, identify discrepancies, and prepare reports. However, it should not be used to create convincing environmental claims without measurement.

Good starting points could be:

food waste;
heating and cooling schedules;
laundry processes;
detecting water leaks;
inventory management;
or procurement planning.

These often yield business results relatively quickly.

AI should suggest, humans should decide

All guest-facing texts, pricing suggestions, translations, and environmental claims should be checked by a human.

A poorly set price, opening hours, or booking condition can cause much more damage than the time saved through automation.

Don’t use a larger system than necessary

Not every task requires the largest language model, and not every communication needs AI-generated video.

For short text corrections, data classification, or simple summaries, smaller, more efficient systems are often sufficient. A good real photograph can often be more valuable than an energy-intensive, AI-generated video.

This is not just an environmental issue. It is also a cost and efficiency consideration.

How do we use AI at I-DEST?

We do not wish to pretend we are observing this process from the outside. In the development of I-DEST and our professional work, we also regularly use artificial intelligence. We do not view it as an independent decision-maker but as a tool that can speed up certain processes, help uncover connections, and reduce errors caused by human oversight.

Creating forecasts and scenarios

We analyse demand time series, seasonality, capacity loads, and various development scenarios.

AI is useful here not because it provides infallible answers. It helps us examine many variations in a short time and notice connections that would be harder to see through manual processing.

However, the quality of input data determines everything. From incorrect or incomplete base data, the system can quickly produce flawed forecasts. Therefore, AI does not replace data verification but makes it even more important.

Software development and debugging

We also use artificial intelligence in the technical development of I-DEST. It helps review longer code sections more quickly, identify certain logical errors, inconsistencies, and security risks, and create test cases more easily. The goal is not for AI to program the system without human oversight. We use it to review developments from multiple perspectives, identify problems earlier, and ideally fix them before users encounter them. This is particularly important for a complex system like I-DEST, where various data sources, languages, user interfaces, evaluation modules, and recommendation functions are interconnected. A single modification can impact the operation of several other functions.

AI-based code analysis and testing add an additional layer of verification to traditional developer and user tests. However, using AI does not guarantee error-free software. Machine-suggested code can also contain errors, outdated solutions, or security issues. Therefore, every significant modification is reviewed by a developer, and the system's operation is tested before implementation.

The goal is not to replace the developer but to reduce the number of errors, support testing, and make development work more reliable.

Translation and terminology checks

We use AI for multilingual processing of strategies, grant documents, tourism descriptions, and standards. Machine translations are followed by human review for all professionally significant documents. In tourism, sustainability, and EU-specific terminology, even one poorly translated term can change the meaning of a sentence or requirement. AI, however, helps ensure consistent use of terms and names throughout a longer document. Terminological consistency is particularly important for international standards, certification requirements, and multilingual digital platforms. Additionally, AI ensures I-DEST's multilingual accessibility so that destinations and providers do not have to upload data in multiple languages into the system—everyone only works as much as minimally necessary.

Marketing and content creation

AI assists in preparing initial text drafts, multilingual content, and communication tailored to different target groups.

We follow a few simple rules in its use:

we communicate in our own tone;
we work with verifiable, real data;
final content is approved by a human;
sustainability claims are tied to evidence;
and we do not consider clicks or bookings as the sole measure of success.

In marketing, the sustainability impact of AI largely depends on the goals set for the system. If only short-term conversions matter, the algorithm may easily reinforce already popular periods and locations. However, if seasonality, spatial distribution, or the visibility of local providers are also included among the objectives, different types of campaigns and recommendations can be created. Occasionally, though rarely, artificial intelligence helps create various posters and summaries for advertisements.

The development direction of I-DEST

At I-DEST, we strive to ensure that AI bases its responses on verified data and identifiable sources, not solely on the internal knowledge of the language model.

We link sustainability assessments not to personal impressions but to internationally accepted criteria. This is why we process the Global Sustainable Tourism Council (GSTC) criteria for destinations and tourism providers in machine-readable form.

The goal of self-assessment is not to automatically present a favourable image of the user. The system should show:

which criteria are met;
what evidence supports this;
which data or document is missing;
and where further action is needed.

In recommendations, timing, seasonality, environmental impact of transport, and spatial concentration are also important considerations.

Researchers at the Technical University of Munich have developed a tourism recommendation system that incorporates CO2 emissions of transport modes, destination popularity, and seasonality into recommendations alongside user preferences (Banerjee et al., 2025).

A related user study found that clearly displayed emissions, popularity, and seasonality information can influence travellers' choices. Participants were more likely to choose lower-emission transport modes or less crowded destinations better suited to the time of year when this information was clearly presented (Banerjee–Mahmudov–Wörndl, 2024).

This is an important lesson. Sustainability is not necessarily a limitation. With the right data and well-defined goals, AI can reveal opportunities that benefit travellers, local communities, and businesses alike.

It’s not AI but us who decide what kind of tourism we build

The environmental impact of artificial intelligence is real. However, the biggest environmental burdens of tourism remain primarily related to travel, transportation, building operations, consumption, and the seasonal overloading of popular locations.

The benefits of AI are also real but not automatic.

The same system can help avoid a crowded location or send even more visitors there. It can help find a local business or make it invisible. It can reduce food waste or produce a convincing but data-less sustainability campaign.

The key question is not AI or sustainability.

What matters is the data the system works from, what we ask of it, what goals we set, and who checks the results.

This is not just a technological issue. It is a professional and managerial decision.

That is why destinations, tourism businesses, and travellers can shape whether artificial intelligence serves further concentration in tourism or a more balanced, accessible, and resource-efficient operation.

It’s not AI but us who decide what kind of tourism we build

Declaration

No new images were generated with artificial intelligence for this blog post. All images used—documented in the source code with the source indicated—come from previously published sources, as at I-DEST, we only use AI when it truly adds value.

Sources used

Banerjee, A. – Mahmudov, T. – Adler, E. – Aisyah, F. N. – Wörndl, W. (2025): Modeling sustainable city trips: integrating CO₂e emissions, popularity, and seasonality into tourism recommender systems. Information Technology & Tourism, 27, 189–226. DOI: 10.1007/s40558-024-00303-1.

Banerjee, A. – Mahmudov, T. – Wörndl, W. (2024): A User Interface Study on Sustainable City Trip Recommendations. arXiv:2405.11243.

Chen, L. – Weng, G. – Moro, E. (2026): Large language models create an uneven informational layer over cities. arXiv:2607.06260. Research pre-study.

Elsworth, C. – Huang, K. – Patterson, D. – Schneider, I. – Sedivy, R. – Goodman, S. – Townsend, B. – Ranganathan, P. – Dean, J. – Vahdat, A. – Gomes, B. – Manyika, J. (2025): Measuring the Environmental Impact of Delivering AI at Google Scale. arXiv:2508.15734.

Gössling, S. – Mei, X. Y. (2025): AI and sustainable tourism: an assessment of risks and opportunities for the SDGs. Current Issues in Tourism. DOI: 10.1080/13683500.2025.2477142.

Hilton (2025): Hilton Scales Up Green Ramadan Initiative in 2025, Achieves 26% Reduction in Plate Waste During the Holy Month. Results of a programme implemented in collaboration with Hilton, UNEP West Asia, and Winnow.

International Energy Agency – IEA (2025): Energy and AI. Analysis of data centre energy consumption and expected growth by 2030.

Lee, S. – Pennington-Gray, L. (2026): Large Language Models Yield Unsustainable Tourist Flows: Testing Algorithmic Biases Using the Baseline-Rescaling-Outcome Model. DOI: 10.31235/osf.io/kqczf. Research pre-study.

Majid, G. M. – Tussyadiah, I. – Kim, Y. R. – Pal, A. (2023): Intelligent automation for sustainable tourism: a systematic review. Journal of Sustainable Tourism, 31(11), 2421–2440. DOI: 10.1080/09669582.2023.2246681.

Global Sustainable Tourism Council – GSTC: GSTC Industry Criteria and GSTC Destination Criteria. International sustainability criteria for destinations, accommodations, and tourism providers.