Zero-Shot Prompting
Asking an AI to do a task with no examples — just instructions.
Zero-shot prompting means giving a model a task description with no demonstration examples — relying entirely on the model's pre-trained knowledge to generalise.
"Translate the following text to French: [text]" is zero-shot. You haven't shown a translation example; you've described what you want.
When zero-shot works well:
- Common, well-defined tasks (translation, summarisation, Q&A)
- Standard formats the model has seen extensively in training
- Rapid prototyping before investing in few-shot examples
When zero-shot struggles:
- Unusual output formats the model rarely produced during training
- Domain-specific jargon or niche tasks
- Highly constrained outputs (specific JSON schemas, house-style rules)
Zero-shot chain-of-thought: Simply appending "Let's think step by step" enables zero-shot reasoning on many problems — one of the most cost-effective prompt improvements available.
Capability evolution: Models in 2024–2026 are dramatically better zero-shot reasoners than 2022 models. Tasks that required few-shot examples 18 months ago often work zero-shot on frontier models today.
In plain terms
Asking someone to parallel park without ever demonstrating it — just giving verbal instructions. Works fine if they already understand driving; not if it's entirely new.