Legal Document
Running an AI-powered service uses real resources. The data centres that power AI and host websites consume electricity, and many use water for cooling. We think it would be wrong to pretend otherwise, so this statement sets out honestly what we understand about our impact, what we do to reduce it, and what we do not do yet.
Revision Genie is a small but growing company. We have deliberately written this statement to be candid rather than promotional. We would rather be honest about the gaps in what we measure than overstate our environmental credentials.
Because we teach students about technology, we think it helps to explain the basics plainly.
2.1 Training vs. answering: Large AI models go through two very different stages. Training a model from scratch is enormously energy-intensive and happens once. Answering your questions (called "inference") uses a much smaller amount of computing power each time.
2.2 Electricity: The computers in data centres draw electricity both to run AI models and to power and cool the buildings they sit in.
2.3 Water: Many data centres use water as part of their cooling systems, which is why water use is part of the conversation about AI's footprint.
2.4 Everyday digital use: Hosting the website, storing data, and sending emails also use energy, even when AI is not involved.
3.1 The most energy-intensive part of AI is training large models from scratch. We do not do this.
3.2 We use models that have already been trained, accessed through Microsoft Azure OpenAI, and we only run inference - that is, generating answers to your questions. This avoids the very large one-off energy and water cost of training a model ourselves.
3.3 Neither we nor Microsoft use your data to train models, as set out in our AI Ethics & Usage Policy.
We do not own data centres. We run on large, shared cloud platforms, which are generally more efficient per unit of work than small operators running their own servers.
Our main providers are:
Microsoft Azure (UK / EU regions) - AI inference (Azure OpenAI) and text-to-speech (Azure Speech Services).
Vercel - website hosting and serverless functions.
MongoDB Atlas - our primary database.
Upstash Redis - caching and rate limiting.
Vercel Blob - file storage for uploads.
Stripe - payment processing.
We rely on these providers' own published environmental commitments. For example, Microsoft has publicly stated goals to be carbon negative, water positive and zero waste by 2030. We do not control these providers and cannot independently guarantee their performance against their own targets.
Efficiency is part of how we build, not an afterthought. Every unnecessary AI call or database query is both a financial cost and an energy cost, so reducing waste is something we do as a matter of normal engineering practice.
5.1 Lighter models where we can: When marking exam answers, we send typed and tick-box answers to a lightweight text model rather than image analysis, and reserve the heavier image-based marking for answers that genuinely need to be seen (such as diagrams and graphs). This uses considerably less computing power.
5.2 Caching: We store slow-changing data (such as subject information and a student's learning context) in a fast cache so we do not repeatedly recompute or re-fetch the same things.
5.3 Minimal context: We send AI models only the information needed to give a useful answer, rather than large amounts of unnecessary history.
5.4 Storing less: AI conversations are not retained by default, and uploaded chat files are automatically cleaned up after a limited period, which reduces the data we store. See our Data Retention Policy.
5.5 Efficient data handling: We use database indexing, pagination and batching so our database does less work to serve the same result.
In the interest of honesty, the following are things we do not currently do:
We do not publish a measured carbon or water figure for our own service.
We have not completed an independent third-party environmental audit.
We do not currently purchase carbon offsets, and we do not claim to be carbon neutral.
We rely on our providers' renewable-energy and efficiency commitments rather than our own on-site measures.
These are areas we expect to revisit as we grow and as better measurement tools and greener options become available to a company of our size.
7.1 Normal study use of Revision Genie has a small footprint, and nothing you do here is wasteful by default.
7.2 As with any online service, using features purposefully - rather than, for example, regenerating the same content many times unnecessarily - is a small, positive habit that applies across all of the internet.
We will review this statement as our service grows, as the tools to measure digital impact improve, and as greener options become practical for us. We will update the "Last Updated" date when we make material changes.
If you have questions or suggestions about our environmental impact, please contact us at support@revisiongenie.com. For more about how we use AI responsibly, see our AI Ethics & Usage Policy.