Sustainable Prompting
Craft efficient prompts that reduce energy consumption while maintaining high-quality results.
Learn how to craft efficient prompts that reduce energy consumption while maintaining high-quality results.
Mission & impact
Sustainable prompting is a core part of GreenPT's mission. GreenPT operates on renewable energy and uses smaller, efficient models and lightweight agents to reduce computational requirements. Using sustainable prompt settings reduces token usage by 20–30%, a significant reduction in the amount of text the model processes.
Principles of sustainable prompt design
1. Be precise and concise
- Avoid verbosity and redundancy. Unnecessarily long or polite phrasing wastes tokens. Replacing "Can you please kindly help me write a short summary" with "Summarize this article" reduces token count and achieves the same result. Scope3 notes that removing "please" from a prompt saved about 7 millilitres of water by reducing token processing.
- Define constraints clearly. Specify the desired length or format (e.g., "give a 100-word summary" or "list three key points in bullet form") so the model doesn't generate long responses. Research from the Technology Carbon Standard shows that requesting a minimal answer can reduce response length by 60% and energy usage by 28%.
- Plan your asks. Think through what information you actually need before prompting. Planning your requests and avoiding multiple follow-ups reduces total prompts and energy consumption.
2. Use purpose-driven structure (ROCKS method)
The ROCKS framework from the CoDesignS AI framework helps users craft clear prompts that minimise iterations.
| Component | Guideline | Why it helps |
|---|---|---|
| Role | State who you are (e.g., "I am a lecturer in pathology"). | Provides context so the model need not infer your perspective. |
| Objective | Specify what you want to achieve. | Prevents vague queries and focuses the model. |
| Community (Audience) | Describe your audience (e.g., "300 Year-2 MBBS students"). | Allows tailoring of explanations, avoiding unnecessary detail. |
| Key / Tone | Indicate tone or style (e.g., "friendly and concise"). | Produces targeted responses rather than verbose or misaligned output. |
| Shape | Define the desired format (bulleted list, table, code). | Encourages structured, concise results. |
Using ROCKS turns broad questions into structured prompts that are easier for the model to satisfy on the first attempt, reducing energy use and follow-up queries.
3. Break complex tasks into logical steps
Complex tasks sometimes require multiple steps, but combining everything into one long prompt can force the model to generate overly long outputs. Chain logic by breaking tasks into sequential steps (e.g., brainstorming ideas, then selecting the best three, then expanding them). This reduces computational overhead by focusing each prompt on a specific subtask rather than one giant query.
4. Encourage concise responses
The length of the response is often more impactful than the prompt length. Both prompt and response length influence energy consumption, and instructing a model to provide only the essential answer can dramatically cut tokens.
- Ask the model to "provide the minimal answer necessary".
- Request summaries instead of full reproductions of source material.
- Avoid open-ended "tell me everything about…" questions unless necessary.
5. Monitor and iterate responsibly
Green prompting is an ongoing practice. Monitor token usage and refine your prompts.
- Track tokens and usage. Use GreenPT's dashboard to understand the impact of your interactions.
- Calibrate and test prompts. Test different prompt structures (prescriptive, descriptive, imaginal, illustrative) and measure output quality and speed. Use the structure that yields the desired output with the fewest tokens.
- Educate yourself and your team. Organisations should adopt token budgets and prompt standards as part of ESG policies. Provide training sessions on prompt efficiency and environmental impact.
Examples: less sustainable vs. green prompts
| Scenario | Less sustainable prompt | Sustainable prompt |
|---|---|---|
| Write a presentation | "Can you please help me by creating a very detailed, comprehensive presentation on climate change with all possible information?" | "Prepare a 5-slide presentation summarizing the main causes and impacts of climate change. Use concise bullet points and limit the total text to 300 words." Specifies length and format; avoids extraneous tokens. |
| Ask for coding help | "I need you to write a Python program that will solve the problem of finding the factorial of a number in the most efficient way possible. Could you also explain every line of code in detail?" | "Write a Python function that returns the factorial of an integer. Provide only the code and a brief docstring." Limits explanation to a short docstring. |
| Plan a lesson | "Can you give me ideas to make my lecture more interactive?" | Use ROCKS: "Role: I am a pathology lecturer for year-2 MBBS students. Objective: Make my lecture on head pathology more interactive. Community: 300 in-person students. Tone: engaging and inclusive. Shape: Suggest three easy-to-implement activities in bullet points." |
Broader sustainability practices
Beyond crafting efficient prompts, there are additional ways to reduce your AI carbon footprint.
- Model and data centre selection. Choose services hosted in regions with greener energy grids and cooler climates, and schedule heavy tasks when renewable energy availability is high. GreenPT's data centres run on 100% renewable energy and have low power-usage effectiveness.
- Hardware efficiency. Running models on modern, efficient hardware reduces energy per inference. Upgrading hardware should be balanced against embodied carbon costs and e-waste.
- Batching requests. Group multiple queries into batch inference where possible to share computational overhead.
- Model size and compression. When developing or fine-tuning models, use smaller architectures and techniques such as quantization and pruning to reduce energy use.
Conclusion
Sustainable prompting is about intention and efficiency. By crafting precise, structured prompts, choosing the right model and settings, and limiting output length, you can significantly reduce the environmental impact of your AI interactions without compromising quality. GreenPT's infrastructure amplifies these benefits by running on renewable energy and using optimized models. Every token counts, think before you prompt and help build an AI-powered future that is both intelligent and sustainable.