Research experiment
Green AI
Testing whether a small behavioural intervention can reduce the computational waste hidden inside everyday AI use.
The question
Most conversations about AI sustainability focus on data centres, model architecture and energy supply. Those questions matter. But there is also a quieter question on the demand side: can ordinary users reduce unnecessary output through better defaults?
The experiment
The first pilot tested one short custom instruction designed to encourage concise, direct answers. Eleven participants used ChatGPT, Claude and Gemini without changing models or receiving prompt-engineering training.
The early result is not a final scientific conclusion. It is a signal worth testing more carefully: small behavioural interventions may help reduce computational demand while preserving usefulness.
What comes next
The next stage expands the sample, improves measurement and tests whether the intervention remains useful across different tasks and user groups. The larger goal is to make sustainable AI use practical rather than abstract.