# How do I…? Answers on cutting AI's impact

> Short, sourced answers to the questions people ask about reducing the environmental impact of using and building AI.

Source: https://greenai.guide/how-do-i/

97 straight answers to the questions people ask about using AI with less energy, water and waste: for everyday users, developers, organisations and schools. Each answer comes first, then the steps and the sources.

## Everyday use of AI

- [How do I avoid replacing my devices just to use AI?](https://greenai.guide/how-do-i/avoid-buying-new-devices-for-ai/): Most AI tools run in the cloud and work through a browser or app on the devices you already have, so you rarely need new hardware to use them. Keeping devices in use longer avoids the footprint of making new ones, which includes electronic waste that researchers expect generative AI to add to.
- [How do I cut the impact of AI image and video generation?](https://greenai.guide/how-do-i/cut-impact-of-ai-image-and-video-generation/): Generate fewer images and videos, plan the prompt before you start, keep video short and at the resolution you need, and reuse good results. Image generation used over a thousand times more energy per request than text classification in one study, and a short AI video can cost far more again.
- [How do I decide whether I need AI for a task at all?](https://greenai.guide/how-do-i/decide-if-i-need-ai-for-a-task/): Ask whether a simpler tool already does the job reliably: a search engine, calculator, map, dictionary, template or spreadsheet formula. Use generative AI where it saves real time or improves the result, such as drafting, explaining or reworking text, and skip it for look-ups and sums that other tools handle well.
- [How do I help my family use AI more sustainably?](https://greenai.guide/how-do-i/help-my-family-use-ai-sustainably/): Agree a few simple household habits rather than rules: use the simplest tool that works, ask clear questions, refine instead of regenerating, and treat AI images and video as occasional rather than endless entertainment. Keeping phones, tablets and laptops in use for longer also matters, because making devices has a footprint too.
- [How do I know when to use an AI "thinking" or reasoning mode?](https://greenai.guide/how-do-i/use-reasoning-modes-only-when-needed/): Use a reasoning or "thinking" mode for genuinely hard problems, such as multi-step maths, tricky code or careful analysis, and a standard or fast model for everyday questions, drafting and summaries. Reasoning modes write long hidden chains of working before answering, and one benchmark found they used on average 30 times more energy.
- [How do I offset the carbon of my AI use?](https://greenai.guide/how-do-i/offset-the-carbon-of-my-ai-use/): Reduce first, and treat offsets as optional extras rather than a way to make AI use "carbon neutral". For most people the footprint of everyday text AI is small, and cutting the heavy uses, such as image and video generation and endless regeneration, does more than buying credits. Standards for software carbon do not let offsets reduce the measured footprint.
- [How do I put my AI energy use in perspective?](https://greenai.guide/how-do-i/put-my-ai-energy-use-in-perspective/): One ordinary text question to a chatbot uses a fraction of a watt-hour on published figures, roughly what an oven uses in about a second. Images, video, long documents and reasoning modes can use many times more, and the global total is growing fast, so both "it's nothing" and "it's a disaster" miss the point.
- [How do I reduce the energy my ChatGPT use consumes?](https://greenai.guide/how-do-i/reduce-energy-of-chatgpt-use/): Ask for what you need in one clear request, keep answers short, start a new chat for a new topic, and go easy on image and video generation. A typical text question uses roughly a third of a watt-hour on published estimates, but long inputs, repeated regeneration and media generation can multiply that many times over.
- [How do I search the web without AI answers I don't need?](https://greenai.guide/how-do-i/get-search-results-without-ai-answers/): Look for a search option or filter that shows plain web links, choose a search engine or setting without automatic AI answers, or go straight to sites you trust using bookmarks. An AI answer generated for every search adds model work you may not need, especially for simple look-ups such as opening hours or a known website.
- [How do I stop long AI chats wasting energy?](https://greenai.guide/how-do-i/keep-long-ai-chats-efficient/): Start a new chat when you change topic, and when a long conversation is still useful, ask for a short recap and carry on from that in a fresh chat. Chatbots typically reread the conversation so far to answer each new message, so very long threads make every reply more expensive than it needs to be.
- [How do I stop regenerating AI answers over and over?](https://greenai.guide/how-do-i/stop-regenerating-ai-answers/): Work out why the first answer missed, then tell the AI what to change in a short follow-up instead of pressing regenerate. Each regeneration is a full new answer, so a habit of rolling the dice five or ten times multiplies the energy of a task, especially for images and video.
- [How do I talk to my children about AI and the environment?](https://greenai.guide/how-do-i/talk-to-children-about-ai-and-the-environment/): Explain that AI runs on powerful computers in big buildings called data centres, which use electricity and often water to keep cool, and that some uses, like making pictures and videos, take far more than others. Keep it calm and practical: the aim is to use AI thoughtfully, not to feel guilty about it.
- [How do I turn off AI features I don't need?](https://greenai.guide/how-do-i/turn-off-ai-features-i-dont-need/): Open the settings of each app, phone, browser or office suite you use, look for sections named after its AI assistant or "smart features", and switch off what you don't use. Features that run automatically on every email, page or document use energy whether you read the result or not, so defaults matter.
- [How do I use AI agents and deep research tools without waste?](https://greenai.guide/how-do-i/use-ai-agents-and-deep-research-sparingly/): Save agents and deep research modes for tasks that genuinely need many steps, such as researching a broad topic across many sources, and use an ordinary chat for quick questions. Agents call a model many times per task: Anthropic reported agents using about four times the tokens of a chat, and multi-agent systems about fifteen times.
- [How do I use AI on my phone without wasting energy?](https://greenai.guide/how-do-i/use-ai-on-my-phone-efficiently/): Switch off AI features you don't use, prefer quick on-device features for simple tasks, and save heavy jobs such as image or video generation for when you really need them. Small on-device models can avoid sending data to a data centre, but they still use your battery, so the biggest saving is not running AI that adds nothing.
- [How do I write prompts that use less energy?](https://greenai.guide/how-do-i/write-prompts-that-use-less-energy/): Write short, specific prompts that say exactly what you want, give only the context the model needs, and set a length for the answer. Energy rises with the amount of text a model reads and writes, so trimming both, and getting a usable answer first time, saves the most.

## Choosing AI tools and models

- [How do I ask an AI company for its energy figures?](https://greenai.guide/how-do-i/ask-an-ai-company-for-energy-figures/): Ask for energy, carbon and water per typical request for the model you use, plus the method, the system boundary and where it runs. Ask for both location-based and market-based carbon figures. Many providers cannot yet answer in full, but a specific question gets a better reply than a general one, and repeated asking builds demand.
- [How do I choose a smaller AI model?](https://greenai.guide/how-do-i/choose-a-smaller-ai-model/): Define what "good enough" means on real examples, then test models from smallest to largest and stop at the first that passes. Small and task-specific models can use a fraction of the energy of large general ones: UNESCO and University College London report savings of up to 90% on suitable tasks.
- [How do I choose the greenest AI chatbot?](https://greenai.guide/how-do-i/choose-the-greenest-ai-chatbot/): No one can yet rank the big chatbots reliably on energy, because providers publish different figures, measured in different ways, and some publish none. Choose a service that is open about its energy and water use, offers lighter models for simple questions, and lets you control features, then use it well: how you use a chatbot matters more than which one you pick.
- [How do I compare the energy use of different AI models?](https://greenai.guide/how-do-i/compare-energy-use-of-ai-models/): Use a standard benchmark where one covers your task, such as Hugging Face's AI Energy Score, and otherwise measure the candidates yourself on the same task, hardware and settings. Compare energy per successful task alongside quality, because a cheap model that fails costs more in the end.
- [How do I decide between running AI on my device or in the cloud?](https://greenai.guide/how-do-i/choose-between-device-and-cloud-ai/): Run small, frequent, private or latency-sensitive tasks on the device, and send large or occasional tasks that need big models to an efficient cloud service. Some studies find large energy savings on the device for suitable models, but well-run data centres are a hard benchmark, so measure for your own case where you can.
- [How do I decide whether classic machine learning will do instead of an LLM?](https://greenai.guide/how-do-i/choose-classical-ml-instead-of-an-llm/): If your inputs are structured data or short labelled text and the output is a number, a category or a ranking, try a classic model first: a linear model, gradient-boosted trees or a small classifier. They are far cheaper to run, faster, easier to explain and often as accurate. Reach for a large language model when the task genuinely needs open-ended language.
- [How do I find a task-specific model instead of using a general chatbot?](https://greenai.guide/how-do-i/find-a-task-specific-model/): Name the task precisely, then search a model hub such as Hugging Face by task type and test two or three small candidates on your own examples. A model built for one job can be orders of magnitude cheaper to run than a general generative model doing the same job, so it is worth an afternoon of testing before you settle on a chatbot.
- [How do I know if an AI tool is green?](https://greenai.guide/how-do-i/know-if-an-ai-tool-is-green/): Look for evidence rather than labels: published energy or emissions figures per request with a stated method, a choice of smaller models, information on where it runs, and independent ratings such as the AI Energy Score. Few tools publish all of this yet, so treat "green" or "sustainable" without numbers as a claim, not a fact.
- [How do I read an AI Energy Score rating?](https://greenai.guide/how-do-i/read-an-ai-energy-score-rating/): An AI Energy Score rating gives a model one to five stars for energy efficiency compared with other models tested on the same task, based on measured GPU energy per 1,000 queries. More stars means less energy for that task, but the stars are relative within a task, so compare models only against others rated for the same job.
- [How do I run an AI model locally?](https://greenai.guide/how-do-i/run-an-ai-model-locally/): Install a local runner such as Ollama or llama.cpp, download a small open-weight model in a quantised format such as GGUF, and run it on your own computer. Pick the smallest model that does your task: quantised versions need far less memory, so an 8-billion-parameter model that needs about 16 GB in 16-bit form fits in about 4–5 GB at 4 bits.
- [How do I spot greenwashing in AI marketing?](https://greenai.guide/how-do-i/spot-greenwashing-in-ai-marketing/): Be wary of words like "green", "sustainable" or "carbon-neutral AI" without figures, methods and boundaries behind them, and of claims that rest on offsets or annual renewable certificates. Honest claims say what was measured, against what baseline, and what was left out; in the EU, generic environmental claims that cannot be proved are banned.
- [How do I use an AI coding assistant without a big footprint?](https://greenai.guide/how-do-i/use-an-ai-coding-assistant-with-a-smaller-footprint/): Use the assistant where it clearly pays back, give it focused context rather than whole repositories, pick a smaller or faster model for routine edits, and break hard tasks into simple steps you review. One study found AI-assisted coding had around 33 times the carbon footprint of the manual approach on the tasks tested, so habits matter.

## Measuring energy and emissions

- [How do I account for the embodied carbon of AI hardware?](https://greenai.guide/how-do-i/account-for-embodied-carbon-of-ai-hardware/): Embodied carbon is the emissions from making, shipping and disposing of hardware. To account for it, take the hardware's total embodied emissions, share them out by how long and how much of the kit your AI work uses, and add that share to the operational figure, as the SCI specification's M term does. Because the data is often thin, state your assumptions; and remember that keeping hardware longer spreads the same emissions over more work.
- [How do I calculate a Software Carbon Intensity score for an AI feature?](https://greenai.guide/how-do-i/calculate-an-sci-score-for-an-ai-feature/): The Software Carbon Intensity (SCI) score, published as ISO/IEC 21031:2024, is ((E × I) + M) per R: energy used, times the location-based grid carbon intensity, plus a share of the hardware's embodied emissions, divided by a functional unit such as one request or one document. Choose the unit, draw the boundary, gather or estimate each term, and recalculate after every change.
- [How do I check a claim about AI's energy use?](https://greenai.guide/how-do-i/check-a-claim-about-ai-energy-use/): Ask six questions: per what (prompt, image, year?), who measured it, what is included, how old the figure is, what it is compared with, and whether it is about efficiency or total use. Most misleading claims, alarming or reassuring, fail one of these. Published per-prompt figures differ mainly because they count different things.
- [How do I estimate the energy of my AI API calls?](https://greenai.guide/how-do-i/estimate-energy-of-ai-api-calls/): You can't meter hardware you don't control, so estimate instead: log input and output tokens per call and use a tool such as EcoLogits, which wraps your AI client and reports estimated energy and emissions per request. Label the figures as estimates, state the method, and replace them with supplier-reported numbers when you can get them.
- [How do I estimate the water footprint of AI?](https://greenai.guide/how-do-i/estimate-the-water-footprint-of-ai/): Water use follows energy use: multiply the energy of your AI work by the data centre's water usage effectiveness (WUE) for on-site cooling, and add the water used to generate the electricity if it is in your boundary. Published per-prompt figures range from a fraction of a millilitre to tens of millilitres, mostly because they count different things. Where the water is used matters as much as how much.
- [How do I find the carbon intensity of my electricity grid?](https://greenai.guide/how-do-i/find-grid-carbon-intensity-data/): Use regional, time-matched grid data for where and when your AI work ran. In Great Britain the Carbon Intensity API from the National Energy System Operator gives regional estimates and forecasts; Electricity Maps covers more than 350 zones worldwide. Fall back to national yearly averages only when nothing finer exists, and say which you used.
- [How do I measure the carbon footprint of training a model?](https://greenai.guide/how-do-i/measure-carbon-footprint-of-training-a-model/): Wrap your training script in an emissions tracker such as CodeCarbon, which estimates energy from your CPU, GPU and memory and converts it to carbon using the local grid. Count every run, including tuning and failures, record hardware, location and duration, and publish the result with the model. If you have no logs, a calculator such as Green Algorithms can estimate it afterwards.
- [How do I measure the energy AI uses on my own computer or GPU?](https://greenai.guide/how-do-i/measure-energy-of-ai-on-my-own-computer/): On your own machine you can measure directly: use CodeCarbon or a power tool that reads processor and GPU counters, or put a metering smart plug between the computer and the wall. Record the idle baseline first, sample for the whole job rather than taking snapshots, and repeat the run a few times before trusting the number.
- [How do I report the energy use of my AI research?](https://greenai.guide/how-do-i/report-energy-use-in-a-research-paper/): Report hardware, GPU-hours for all runs including tuning and failures, energy in kWh with the tool used, the location and grid carbon intensity, emissions in kg CO₂e, and inference energy at a stated batch size. Say what the boundary includes and how uncertain the figures are. Measuring as you go, with a tool such as CodeCarbon, makes this a table rather than a reconstruction.
- [How do I set a baseline to prove AI energy savings?](https://greenai.guide/how-do-i/set-a-baseline-for-ai-energy-savings/): Measure the current system before you change anything, on the same hardware, load and inputs you will use afterwards, and repeat the runs to see the normal spread. Report energy per unit of useful work as well as in total, and only claim a saving bigger than the noise and achieved without a drop in quality.
- [How do I set a compute budget for a research project?](https://greenai.guide/how-do-i/set-a-compute-budget-for-a-research-project/): Estimate the GPU-hours your main experiments need, add a margin for exploration, and track use against that budget like money. Pilot ideas at small scale, stop poor runs early, reuse pretrained models and checkpoints, and report results against the compute they took. A budget makes trade-offs visible and stops endless tuning.
- [How do I track AI energy per task rather than in total?](https://greenai.guide/how-do-i/track-ai-energy-per-task/): Pick a functional unit that matches the value the AI delivers, such as per document summarised or per successful task, and divide energy by it. Track energy per unit, quality per unit and the total side by side. Per-unit figures show whether the system is getting more efficient; the total shows whether growth is outrunning it.
- [How do I work out my own AI carbon footprint?](https://greenai.guide/how-do-i/work-out-my-personal-ai-carbon-footprint/): Count roughly how many text questions, long documents and images you use each week, multiply by published per-request figures, and you have a rough range. A typical text prompt is a fraction of a watt-hour on the companies' own figures, while long inputs and images cost far more, so your heaviest habits usually dominate the total. Treat the result as an estimate, not a precise number.

## For developers and data scientists

- [How do I avoid training a model from scratch?](https://greenai.guide/how-do-i/avoid-training-a-model-from-scratch/): Search for a pre-trained model close to your task, then adapt it: fine-tune it, train a small adapter such as LoRA, or add knowledge through retrieval instead of retraining. Where you already have several models fine-tuned from the same base, averaging their weights can improve accuracy at no extra inference cost. Training from scratch is rarely necessary outside foundation-model research.
- [How do I batch AI requests to save energy?](https://greenai.guide/how-do-i/batch-ai-requests/): Serve live traffic with an inference engine that uses continuous batching, so concurrent requests share the accelerator. Queue work that doesn't need an instant answer, such as overnight document processing, and run it in large batches, ideally when the grid is cleaner. An accelerator serving one request at a time is underused but still draws substantial power.
- [How do I choose between retrieval and fine-tuning to save energy?](https://greenai.guide/how-do-i/choose-between-rag-and-fine-tuning/): Use retrieval when the model needs facts that change or are specific to you: put them in a searchable index and fetch only the relevant passages. Use fine-tuning when the model needs a skill, format or style it lacks. Retraining to keep facts up to date repeats training costs, while pasting whole documents into every prompt wastes inference, so each approach has its place.
- [How do I cut the energy of hyperparameter searches?](https://greenai.guide/how-do-i/cut-the-energy-of-hyperparameter-searches/): Replace large grid searches with random or adaptive search, stop poor trials early, run pilots on small data or small models first, and start from published defaults for your architecture. Track energy for every trial, not just the winner, and set a compute budget before you start. Tuning runs often cost more than the final training run, and they rarely get reported.
- [How do I cut the energy of serving a model in production?](https://greenai.guide/how-do-i/cut-the-energy-of-serving-a-model/): Serve the smallest model that meets your bar, on a runtime built for inference, with batching turned on, weights in half precision or quantised, and the library's optimised attention and cache paths enabled. Then benchmark two or three runtimes and hardware options at your real traffic, and pick on energy per request as well as latency.
- [How do I decide whether a mixture-of-experts model will save energy?](https://greenai.guide/how-do-i/decide-if-mixture-of-experts-saves-energy/): A mixture-of-experts model routes each token to a few specialist sub-networks, so only part of the model does work per token. Mixtral, for example, has 47 billion parameters but uses 13 billion per token. That can cut computation per answer compared with a dense model of similar quality, but every expert must still sit in memory, so compare quality, latency and energy per request on your own hardware.
- [How do I design AI features so they don't run when nobody needs them?](https://greenai.guide/how-do-i/design-ai-features-that-dont-run-needlessly/): Make AI features run when a user asks, not automatically on every page load, message or file. Generate once and reuse shared results, use plain code for decisions a rule can make, and remove AI features that few people use. The W3C Web Sustainability Guidelines ask teams to evaluate the need for AI before implementing it, and that applies to each trigger as well as each feature.
- [How do I distil a large model into a smaller one?](https://greenai.guide/how-do-i/distil-a-large-model-into-a-small-one/): Check first whether a distilled version of your model already exists. If not, use the large "teacher" model to label or generate examples for your task, train a small "student" model on them, and test the student on your own real cases. DistilBERT, an early example, was 40% smaller and 60% faster than BERT while keeping 97% of its language understanding.
- [How do I fine-tune a model without huge compute?](https://greenai.guide/how-do-i/fine-tune-a-model-with-less-compute/): Start from a pre-trained model and train a small adapter instead of all its weights, using LoRA, or QLoRA with 4-bit weights. LoRA cut trainable parameters by 10,000 times and GPU memory by three times compared with full fine-tuning of GPT-3, and QLoRA fine-tuned a 65-billion-parameter model on a single 48 GB GPU. Compare the result with your quality bar before considering anything bigger.
- [How do I get AI to write energy-efficient code?](https://greenai.guide/how-do-i/get-ai-to-write-energy-efficient-code/): Ask explicitly for efficient code and name the constraint, such as memory, allocations or algorithmic complexity, then run the tests and measure the code's energy before accepting it. Research found prompt changes could make generated code up to 91.9% more energy-efficient, but often at the cost of correctness. Give the assistant energy principles to follow, and keep a person checking the result.
- [How do I limit how much an LLM writes?](https://greenai.guide/how-do-i/limit-how-much-an-llm-writes/): Set a maximum output length in the API call, tell the model the format and length you want in the prompt, and use stop sequences or structured output so it ends when the job is done. Energy grows with the number of tokens a model generates, so capping output is one of the simplest savings in any application.
- [How do I make a model small enough for an edge device?](https://greenai.guide/how-do-i/make-a-model-small-enough-for-an-edge-device/): Start from the smallest model that could meet your quality bar, then shrink it further with distillation, quantisation and pruning, testing each step on the target device. Measure energy per inference, latency and accuracy together on the real hardware, and let the model step down when battery is low or the device is hot.
- [How do I make vector search and RAG more efficient?](https://greenai.guide/how-do-i/make-vector-search-and-rag-more-efficient/): Use an approximate nearest-neighbour index instead of comparing every query with every stored vector, retrieve a few short, relevant passages rather than many long ones, and re-embed only content that has changed. Then use the smallest generation model that answers well from good context. Most of the waste in retrieval systems comes from brute-force search and oversized prompts.
- [How do I prune a neural network?](https://greenai.guide/how-do-i/prune-a-neural-network/): Prune a trained network by removing the weights, neurons or layers that contribute least, then fine-tune briefly so it recovers, and test it against your quality bar. Pruning cut AlexNet's parameters ninefold with no loss of accuracy, and very large language models have been pruned to at least 50% sparsity in one shot. Prefer structured pruning if you want ordinary hardware to run faster.
- [How do I quantise a model?](https://greenai.guide/how-do-i/quantise-a-model/): Quantising stores a model's weights with fewer bits, typically 8 or 4 instead of 16 or 32, so it needs less memory and often less energy per answer. Load a ready-quantised version or quantise with your library's built-in methods, then test it on your own examples against the full-precision model. Keep it if quality holds and energy per request falls.
- [How do I reduce the energy of training a model?](https://greenai.guide/how-do-i/reduce-the-energy-of-training-a-model/): Start from a pre-trained model where you can, choose the smallest architecture that could meet your bar, train in mixed precision, stop when validation stops improving, and run on efficient hardware in a low-carbon region and time. Patterson and colleagues found that the choice of model, data centre and processor can reduce the carbon footprint of training by up to around 100 to 1,000 times. Measure every run so the trade-offs are visible.
- [How do I route requests to smaller models?](https://greenai.guide/how-do-i/set-up-model-routing/): Put a router in front of your models: a rule, a small classifier or a cascade that tries a small model first and escalates when confidence is low or a check fails. Published routers have cut costs by more than half in some tests without lowering quality. Tune the threshold on your own data and measure cost per successful task, not per call.
- [How do I speed up a large model with speculative decoding?](https://greenai.guide/how-do-i/use-speculative-decoding/): Pair your large model with a small draft model from the same family: the draft proposes several tokens and the large model checks them in one pass, keeping only those it agrees with. The output is the same as the large model alone, and the original study reported 2 to 3 times faster inference. Check your serving stack supports it, then measure energy per request, not just speed.
- [How do I stop AI agents wasting tokens?](https://greenai.guide/how-do-i/stop-ai-agents-wasting-tokens/): Give every agent a testable goal that ends the run, hard limits on steps, tokens, time and spend, and loop detection that stops it when it repeats itself. Trim the context each step carries, use plain code for decisions a rule can make, and send simple sub-tasks to small models. Agents can use many times the tokens of a chat, so these limits pay off quickly.
- [How do I stop my chatbot resending the whole conversation?](https://greenai.guide/how-do-i/stop-resending-the-whole-conversation/): Most chat features send the whole conversation back to the model on every turn, so input grows with every message. Cap how many past turns you send, summarise older ones into a short note, retrieve only the earlier messages or documents that matter, and cache fixed system prompts. Long inputs cost far more energy than short ones.
- [How do I store AI training data so pipelines read less?](https://greenai.guide/how-do-i/store-training-data-so-pipelines-read-less/): Store training and feature data in a columnar format so jobs read only the columns and partitions they use, and convert raw text to a binary format once instead of re-parsing it on every run or epoch. Measure a typical job before and after to check the conversion pays for itself. Data handling is easy to overlook next to the model, but it runs on every experiment.
- [How do I use caching to cut the energy of LLM calls?](https://greenai.guide/how-do-i/cache-llm-calls/): Use two layers. Provider prompt caching reuses the processing of a fixed prefix, such as a long system prompt or shared document, so put stable content first. Your own response cache stores answers to identical requests so the model is not called again. Both cut computation, cost and latency, as long as you set sensible expiry times.

## Data centres, cloud and hosting

- [How do I check whether my AI runs on renewable energy?](https://greenai.guide/how-do-i/check-whether-my-ai-runs-on-renewable-energy/): Ask where your AI actually runs, then look at two things for that region: the grid's carbon intensity and the share of electricity matched with clean power hour by hour. "100% renewable" usually means annual certificates or contracts that match yearly totals, which don't change what the local grid supplies at a given hour. Ask for both location-based and market-based figures.
- [How do I choose a sustainable data centre or host for AI?](https://greenai.guide/how-do-i/choose-a-sustainable-data-centre-for-ai/): Ask for evidence rather than labels: the grid carbon intensity and hourly clean-power share where your work runs, the site's PUE and water use, how long hardware stays in service, and whether you will get energy and emissions data for your own account. In the EU, larger data centres must report energy and water use, so the data often exists. Choose clean hosting, but keep cutting the work you send there.
- [How do I choose energy-efficient hardware for AI?](https://greenai.guide/how-do-i/choose-energy-efficient-hardware-for-ai/): Benchmark your actual model on two or three hardware and runtime options and compare energy per request or per training run at realistic batch sizes, alongside latency and cost. Patterson and colleagues found that the choice of model, data centre and processor together can cut the carbon of training by up to around 100 to 1,000 times. Keep good hardware busy and in service, because making new kit has its own footprint.
- [How do I find my cloud provider's carbon data?](https://greenai.guide/how-do-i/find-my-cloud-providers-carbon-data/): The big cloud providers publish customer carbon data in their billing or sustainability consoles: the AWS Sustainability console, Microsoft's Azure emissions tools and Google Cloud's Carbon Footprint dashboard. They break emissions down by service, region and scope, but are monthly and estimated, so treat them as a trend and pair them with your own per-workload measurements.
- [How do I have a say on a data centre planned near me?](https://greenai.guide/how-do-i/have-a-say-on-a-data-centre-near-me/): Find the planning application, read the developer's statements on energy, water, noise and heat, and comment within the consultation period with specific, evidence-based questions. Ask how much grid capacity and water the site will use, whether waste heat will be reused, and what energy and water figures will be reported once it runs. Specific questions carry more weight than general objections.
- [How do I make sense of PUE and WUE figures?](https://greenai.guide/how-do-i/understand-pue-and-wue/): Power usage effectiveness (PUE) is a data centre's total energy divided by the energy used by its computing equipment: closer to 1.0 means less lost to cooling and power conversion. Water usage effectiveness (WUE) is the litres of water used on site per kilowatt-hour of computing: lower is better. Neither says how much computing is done, how clean the power is, or whether the work was needed.
- [How do I pick a cloud region with low carbon?](https://greenai.guide/how-do-i/pick-a-low-carbon-cloud-region/): Compare regions on grid carbon intensity and the share of carbon-free energy, using your provider's published region data or a source such as Electricity Maps, then weigh latency, data-residency rules and cost. Batch work such as training, fine-tuning and overnight processing is the easiest to move; user-facing inference usually needs to stay close to users.
- [How do I reduce the footprint of AI experiments on a research cluster?](https://greenai.guide/how-do-i/reduce-footprint-of-ai-on-a-research-cluster/): Start by measuring: many clusters record energy per job in the Slurm scheduler, and CodeCarbon or the Green Algorithms calculator can fill gaps. Then request only the GPUs, memory and time you need, pilot ideas at small scale, checkpoint long jobs and schedule flexible ones when the grid is cleaner. Report the energy of whole projects, not just final runs.
- [How do I reduce the water footprint of AI?](https://greenai.guide/how-do-i/reduce-the-water-footprint-of-ai/): Water use follows energy use, so the first step is the same: use smaller models, shorter outputs and fewer regenerations. Beyond that, choose hosts that publish water use and avoid water-stressed regions for heavy work, and ask suppliers for site-level water figures. Per text prompt the water is small; totals and local effects are what matter.
- [How do I run AI jobs on spare cloud capacity?](https://greenai.guide/how-do-i/run-ai-jobs-on-spare-cloud-capacity/): Cloud providers sell spare capacity cheaply on condition they can take it back at short notice, as spot or preemptible instances. Training, evaluation, batch inference and data processing can use it if they save checkpoints and retry cleanly. Microsoft's guidance is to use it only for interrupt-tolerant work, because the interruptions can add instability.
- [How do I schedule training when the grid is greener?](https://greenai.guide/how-do-i/schedule-training-when-the-grid-is-greener/): Find the jobs that can wait, such as training, fine-tuning, evaluation and batch inference, and start them when a carbon-intensity forecast shows the grid is cleaner. Save regular checkpoints so long jobs can pause when intensity rises and resume when it falls, within a firm deadline. Carbon-aware scheduling cuts carbon, not energy, so do it on top of efficiency work.
- [How do I stop GPUs and notebooks sitting idle?](https://greenai.guide/how-do-i/stop-gpus-sitting-idle/): Make on-demand the default: pipelines and training jobs should start resources when triggered and release them when done. Shut down idle notebooks and development GPUs automatically after a set time, scale non-production environments to zero out of hours, and report idle GPU-hours per team next to cost so the waste is visible.

## For organisations

- [How do I check an AI project is worth its environmental cost?](https://greenai.guide/how-do-i/check-an-ai-project-is-worth-its-footprint/): Estimate the AI's own footprint (training or fine-tuning, plus running it for its expected life) and set it against the benefit, measured over the same period. Compare it with the simplest non-AI option, not with doing nothing. If the project is meant to save energy or emissions, the saving has to clearly exceed the AI's own cost, or it is a net loss.
- [How do I encourage greener AI as a policymaker?](https://greenai.guide/how-do-i/encourage-greener-ai-as-a-policymaker/): Start with transparency: require or encourage comparable energy, water and emissions reporting from AI providers and data centres. Use public procurement to reward efficient AI, plan grid and water capacity for data centres, and fund research and skills in efficient methods. Watch total demand as well as efficiency, because cheaper AI tends to be used more.
- [How do I keep my small business's AI footprint low?](https://greenai.guide/how-do-i/keep-my-small-business-ai-footprint-low/): Use AI for the tasks where it clearly saves time, keep requests short and specific, refine rather than regenerate, and go easy on images and video. Turn off AI features you are paying for but not using, and keep a simple record of which tools you use so you can answer customers who ask. For a small business, the same habits cut subscription and staff costs.
- [How do I make honest claims about our green AI?](https://greenai.guide/how-do-i/make-honest-claims-about-green-ai/): Claim only what you have measured, be specific about what improved and against what baseline, and state the method, boundary and limits. Separate efficiency per task from total use, and don't rest a claim on offsets or annual certificates. Vague words such as "green AI" or "carbon-neutral AI" are risky, and in the UK and EU can breach consumer law.
- [How do I make my company's AI use more sustainable?](https://greenai.guide/how-do-i/make-my-companys-ai-use-more-sustainable/): Start by mapping where your company uses AI, then make the efficient choice the default: the smallest model that meets the quality bar, short outputs, AI features off unless needed. Ask suppliers for energy per request and where they run, measure per useful task, and put it in a short policy with an owner. Most of these steps cut cost and delay as well as energy.
- [How do I make the business case for green AI?](https://greenai.guide/how-do-i/make-the-business-case-for-green-ai/): Make the case in cost, speed and risk, because the things that cut AI's energy use (smaller models, fewer tokens, fewer wasted calls) also cut bills and waiting. Show one measured example from your own work, add the reporting and regulatory reasons, and keep the environmental benefit as a real but secondary point if that suits your audience.
- [How do I map where my organisation uses AI?](https://greenai.guide/how-do-i/map-where-my-organisation-uses-ai/): Build a simple inventory with four kinds of entry: AI tools staff use directly, AI features built into software you already have, automated workflows and agents, and suppliers whose services run AI for you. For each, note the owner, how often it runs, the model or provider if known and what data goes in. The inventory is the starting point for cutting waste, managing risk and reporting.
- [How do I meet the EU AI Act's energy documentation rules?](https://greenai.guide/how-do-i/meet-eu-ai-act-energy-documentation-rules/): If you provide a general-purpose AI model in the EU, the AI Act's Annex XI requires your technical documentation to include the model's known or estimated energy consumption, along with the computational resources used for training. Track training compute and energy from the start, record the method, and estimate where you cannot measure. If you only use AI, the duty falls on your supplier, but you can ask for the figures.
- [How do I put sustainability into AI procurement?](https://greenai.guide/how-do-i/put-sustainability-into-ai-procurement/): Ask suppliers for energy and emissions per request or per task, with the method and boundary, where the workload runs, and both location-based and market-based carbon figures. Score the quality of that evidence in the tender, and write reporting, model choice and notice of model changes into the contract. Many suppliers cannot yet answer fully, but asking shows who takes it seriously.
- [How do I report AI emissions in our sustainability report?](https://greenai.guide/how-do-i/report-ai-emissions-in-a-sustainability-report/): Report AI inside your existing greenhouse gas inventory, not as a separate figure: electricity for hardware you run goes in Scope 2, and cloud and AI services usually go in Scope 3 purchased goods and services. Use supplier data where you can get it, estimate the rest from usage with a stated method, and give the boundary and uncertainty. Show energy per useful task alongside the total.
- [How do I set targets for AI energy and emissions?](https://greenai.guide/how-do-i/set-targets-for-ai-energy-and-emissions/): Measure a baseline first, then set two kinds of target together: an intensity target (energy or emissions per useful task) and an absolute target or cap on the total. Intensity shows efficiency; the absolute figure shows whether growth in use is eating the savings. Review them at least yearly as models and usage change.
- [How do I stop AI efficiency gains being cancelled out by more use?](https://greenai.guide/how-do-i/stop-efficiency-gains-being-cancelled-out/): Track total AI energy as well as energy per task, so growth is visible, and decide in advance what happens to the savings: retire freed capacity, keep a budget or cap, or reinvest deliberately. Pair efficiency with sufficiency by asking whether each new use is worth it. Without these, cheaper AI tends simply to be used more.
- [How do I switch off unneeded AI features across my organisation?](https://greenai.guide/how-do-i/switch-off-unneeded-ai-features-across-my-organisation/): Use your inventory to find AI features that are on by default, then decide feature by feature: keep what people use, switch off or restrict what they don't, and give access on request. Make the change in admin consoles rather than asking users, tell staff what changed and why, and review usage data each quarter. Fewer unused AI licences also saves money.
- [How do I train staff to use AI efficiently?](https://greenai.guide/how-do-i/train-staff-to-use-ai-efficiently/): Teach a handful of habits in a short session with real examples from your own work: check whether AI is the right tool, write clear and specific prompts, ask for short answers, refine rather than regenerate, and go easy on images and video. Back the training with efficient defaults in company tools, because defaults reach people who skip the course.
- [How do I use AI sustainably in a charity?](https://greenai.guide/how-do-i/use-ai-sustainably-in-a-charity/): Use AI where it frees staff and volunteer time for your mission, choose the lightest tool that does the job, and avoid heavy uses such as bulk image generation. Write a short policy that covers environmental impact alongside data protection, ask suppliers simple questions about energy and location, and be honest with supporters about how you use it.
- [How do I write a green AI policy?](https://greenai.guide/how-do-i/write-a-green-ai-policy/): Keep it short and practical: say when AI is worth using, set efficient defaults (smallest adequate model, output limits, features off unless needed), require energy questions in procurement, and commit to measuring per task. Give it an owner, a review date and a few measurable commitments, and attach it to processes you already run.

## Schools, universities and training

- [How do I find a PhD in green AI?](https://greenai.guide/how-do-i/find-a-phd-in-green-ai/): Look beyond projects labelled "green AI": efficient AI research sits in computer science, mathematics, electrical engineering and domain sciences, often under names like efficient machine learning, sustainable computing or low-power hardware. Search doctoral training programmes and project listings, read recent papers to find active supervisors, and ask how energy will be measured in the work.
- [How do I include green AI in a university course?](https://greenai.guide/how-do-i/include-green-ai-in-a-university-course/): Add a short, practical strand rather than a whole new course: explain where AI's energy goes, have students measure the energy of their own code or model runs, and ask them to report compute and energy alongside accuracy. In technical courses, teach efficient methods such as smaller models, quantisation and reuse; in others, focus on sensible use and on judging claims.
- [How do I learn about green AI?](https://greenai.guide/how-do-i/learn-about-green-ai/): Start with a short explainer on where AI's energy goes, then try one free tool on your own use, such as an energy tracker for code or an estimator for AI API calls. Follow with a course at your level and a community or research group in your field. Learning by measuring something real sticks better than reading alone.
- [How do I set school rules for sustainable AI use?](https://greenai.guide/how-do-i/set-school-rules-for-sustainable-ai-use/): Add a short environmental section to your existing school AI policy rather than writing a separate one. Say when AI is the right tool, prefer short text over image and video generation, set light defaults in school accounts, and teach pupils why. Keep it alongside the rules on safeguarding, data protection and assessment, which still come first.
- [How do I teach pupils about AI's environmental impact?](https://greenai.guide/how-do-i/teach-pupils-about-ais-environmental-impact/): Start with the idea that every AI answer runs on real computers that use electricity and water, then use a few reliable figures and a hands-on activity, such as comparing text with image generation. Link it to computing, science and geography, and keep it balanced: the aim is to help pupils use AI thoughtfully, not to make them feel guilty for using it.
- [How do I use AI for my studies without wasting energy?](https://greenai.guide/how-do-i/use-ai-for-studying-without-waste/): Use AI where it helps you learn, such as explaining a hard idea, quizzing you or giving feedback on a draft, and use your notes, textbooks and library search for facts and sources. Ask focused questions, paste only the passage you need, and avoid regenerating or generating images for fun: those habits save energy and usually help learning too.
