- Asking an LLM to perform a task without providing any examples in the prompt.
- Instruction-tuned models like GPT-4o and Claude reliably perform tasks without labeled examples.
- Few-shot provides examples and often improves accuracy on ambiguous tasks, while zero-shot is simpler and uses fewer tokens.
- Adding 'Let's think step by step' dramatically improves performance on reasoning tasks without examples.
- Choose zero-shot for clearly specified tasks with high-quality models when simplicity or token cost matters.
What Is Zero-Shot Prompting?
Zero-shotZero-ShotThe ability of a model to perform a task correctly from just a description, with no examples — demonstrating that it has generalized the concept from training.Learn more → prompting is asking an LLMLLMLarge Language Model — a neural network trained on vast amounts of text data that can understand and generate human-quality language across a wide range of tasks.Learn more → to perform a task without providing any examples in the promptPromptThe input text sent to a language model — the question, instruction, or context that triggers a response.Learn more →. You describe what you want — classify this text, summarize this document, translate this sentence — and trust the model to generalize from its training. 'Zero-shot' refers to zero demonstrations provided in context.
Modern instruction-tuned models are remarkably capable zero-shot performers. When you ask GPTGPTGenerative Pre-trained Transformer — the model architecture and family name behind OpenAI's most famous models, from GPT-2 to GPT-5.Learn more →-4o or Claude to 'classify the sentiment of this review as positive, negative, or neutral', it reliably does so — without needing a single labeled example. This is the power of large-scale pretrainingPretrainingThe initial large-scale training phase where language models learn patterns from vast text datasets through next-token prediction.Learn more → and instruction tuning: the model has internalized diverse tasks from training and can apply them immediately.
Zero-Shot vs. Few-Shot
The contrast to zero-shot is few-shotFew-ShotA prompting technique where you include a small number of input-output examples in the prompt to demonstrate the desired task format and behavior to the model.Learn more → prompting, which provides a small number of input-output examples before the task. Few-shot often improves accuracy on ambiguous or specialized tasks — the examples clarify exactly what format, tone, or edge-case handling you want. But zero-shot is simpler, uses fewer tokens, and often works just as well for clearly specified tasks.
When to choose zero-shot: tasks with clear descriptions, simple formats, high-quality frontier models, and when tokenTokenThe basic unit of text that an LLM processes — roughly corresponding to a word or word-piece. Models read input and produce output in tokens, which is also how API usage is measured and billed.Learn more → cost or simplicity matters. When to add few-shot examples: specialized classification schemes the model might not know, desired output formats that are hard to describe precisely, tasks where quality is notably inconsistent zero-shot, and edge cases you want explicitly handled.
Zero-Shot Chain of Thought
A powerful variant is zero-shot chain-of-thought prompting: adding 'Let's think step by step' (or 'Think carefully') to your prompt, without providing any examples. This single phrase dramatically improves performance on reasoning tasks by eliciting the model's internal chain-of-thought before committing to an answer.
The effectiveness of this technique — discovered by Kojima et al. in 2022 — reveals that the instruction to reason is more powerful than the content of specific examples in many cases. For math problems, logic puzzles, and multi-step analysis, zero-shot CoT often outperforms complex few-shot prompts and requires minimal effort to implement.