What are the three common types of prompt engineering?
A common three-part explanation of prompt engineering uses zero-shot, one-shot, and few-shot prompting. Zero-shot gives instructions without an example. One-shot supplies one demonstration. Few-shot supplies several demonstrations that show the desired pattern or clarify difficult boundaries.
These are three example-based prompt patterns, not the complete field. Production systems may also use retrieval, tools, multi-stage routing, structured outputs, fine-tuned models, or human approval. The zero-one-few taxonomy remains useful because it shows how examples change a model's interpretation of a task.
Why the three common types of prompt engineering matter
Examples can teach format and judgment faster than another paragraph of instructions. They can also introduce bias. A few-shot set that contains only easy enterprise leads may make the classifier unreliable for small companies, nonprofits, or incomplete records. The examples should cover meaningful boundaries rather than repeat the same obvious case.
Begin zero-shot when the task and labels are clear. Add one example when the required shape is unusual. Use a small diverse set when the model needs to learn distinctions, tone, or edge-case handling. Evaluate all versions on the same held-out cases so the added examples earn their space.
How to use the three common types of prompt engineering in practice
Put the three common types of prompt engineering under the same change discipline as other business logic. The owner should know the prompt version, data sources, model, tools, expected behavior, and the consequences of an incorrect output. Review a sample of real cases before relying on an aggregate score. The sample should include incomplete, conflicting, and high-value records because averages often hide the failures with the largest business cost. Place the documentation for the three common types of prompt engineering beside the workflow or decision it affects. That keeps definitions, evidence, ownership, and later corrections available to the people who need them.
Example
A prompt classifies customer comments as bug, feature request, billing, or unclear. The zero-shot version handles direct comments. The team then adds four examples showing ambiguous requests and mixed issues. Few-shot accuracy improves, but the owner keeps the test cases separate from the demonstrations to avoid scoring the prompt on examples it has already seen.
Use the fewest examples that produce stable behavior. A larger prompt costs more, can anchor the model too tightly, and becomes harder to maintain when the business labels change.