Using AI services with a clear task in mind
Plan an AI request, check its result and avoid unnecessary repeats without claiming a fixed footprint per prompt.
2 min read · Oct 9, 2026
AI services can be useful for drafting, summarizing or exploring options. They also run on computing infrastructure, but there is no universal energy figure for one prompt. The model, task, response length and service all change the calculation. ECOA's methodology treats its estimates as scenarios, not measurements of an AI account.
Give the tool one well-defined job
Before opening a service, write down the decision you need. For example: “Summarize this public meeting agenda in five points and list any dates I should verify.” Then include the relevant public text, the audience and a useful output format. A good first request may reduce back-and-forth, but it does not guarantee fewer computations in every system. If the answer is wrong or incomplete, correct it rather than accepting it to avoid another call.
Do not paste private client files, passwords, medical details or other sensitive material into an AI service without checking its terms and your organization's rules. For some work, a local note, search result or direct calculation is a better tool. For other work, AI may save time. Choose by task, not by a slogan about carbon.
Check the output before reuse
An AI summary can invent a citation, misread a number or omit a condition. Open the original source before repeating a fact. If you need a regulation, current price or service setting, use the responsible source for that place and date. In a U.S. workplace or classroom, the same verification rule applies even if a draft sounds polished.
For repeated tasks, save a prompt outline that has worked: purpose, source, format and what to check. Do not feed the entire prior conversation back into a new request when a short, accurate context will do. The goal is a clearer workflow, not a promised emissions saving.
A seven-day test
Choose one recurring, non-sensitive task. Note the number of attempts, how long checking takes and whether the final answer is usable. Revise the task description once. These are your own observations, not a measured CO2 reduction.
Source on data-center energy context: U.S. Department of Energy, Lawrence Berkeley National Laboratory, United States Data Center Energy Usage Report. It does not provide an energy value for your prompts: Source page
Estimated savings from sticking to these habits are explained on the methodology page. Focus on the habit change itself rather than on absolute numbers. Read the methodology
Keep reading
Guides that overlap with this one or make a good next step.
- Streamlining AI Workflows at WorkBatch work-related AI calls such as meeting summaries, code generation, and translation to cut retries and wasted context.
- Using Slack, Teams, and Collaboration Tools AsynchronouslyLower notification load by cutting real-time threads and pushes and collaborating around document links instead.
