How we think about AI

KindScore uses artificial intelligence as part of how we research and compile a rating.

We believe AI has enormous potential benefits as well as risks, depending on how it's used. This page covers why we use AI, and why we believe our use is responsible.

AI is uniquely powerful as a tool for reading large amounts of public information from search results and distilling it into the most important points.

It means we can provide an estimated sustainability impact rating for every travel activity with an online presence.

Our firm belief is that enabling travellers to choose more sustainable operators is a sustainability benefit that vastly outweighs the cost of running the KindScore ratings.

AI is one of several digital tools in the pipeline that creates a KindScore rating for any activity.

We endeavour to keep our use of AI extremely efficient. This matters because it's a technology with the potential to use a lot of energy. Staying efficient also keeps our costs manageable, so as independent founders we have a real incentive to be responsible. We also use prompt caching and result caching to further reduce AI use.

Running a KindScore rating uses less energy than watching 1 minute of videos on TikTok or Instagram. We measure the tokens used per rating to estimate the energy usage to ensure we stay accountable for our energy usage. As of 12th September 2026, an average KindScore rating used around 33,000 input tokens and produced around 1,000 output tokens.

AI is not the only technology with high potential energy consumption. A lot of energy is used in digital infrastructure, including data centers, when we watch Netflix, TikTok or Instagram, make video calls or even search the web.

There's a lot of headlines and controversy about data centers in the media currently. AI uses data centers, as does any activity on the internet. However, there is currently huge demand for more data centers globally for AI.

There are two big factors driving this data center demand. The first is corporations using irresponsibly massive amounts of AI. An example is Meta, who are building a 5-gigawatt data center almost the size of Manhattan, largely to serve their social media algorithms and their push for users to generate AI videos, which is extremely energy-intensive.

The second factor is in training new and more powerful AI models, which has risks beyond the environmental impact of data centers. This is covered below.

KindScore works using Claude Haiku, the most efficient existing Claude model. It's a very effective tool already and has clear, beneficial use cases.

While the topic of frontier AI model development is much too large to discuss here, we will state that we don't support this development continuing at its current pace, due to legislation failing to keep up and the increasing dangers of more powerful models being used irresponsibly.