What is zero-shot and few-shot prompting?
Zero-shot prompting means giving the AI a task without examples. Few-shot prompting includes 1–5 examples to demonstrate the desired output format or tone. Few-shot delivers significantly better results for specific formats.
DEFINITION
Zero-shot and few-shot are two fundamental prompting strategies that differ in the number of examples provided.
Zero-shot prompting: You describe the task. Without an example. The model solves it solely based on its training understanding. This works well for common, clear task types (summarising, translating, answering questions) but fails for specific formats without a template.
Few-shot prompting: You include 1–5 examples in the prompt (input → desired output). The model learns from these examples what the answer should look like — format, style, depth, tone. Significantly more effective for a specific output format or quality level.
Example difference:
- Zero-shot: “Write a LinkedIn summary for this text.”
- Few-shot: “Here is an example text and its LinkedIn summary. Now write a summary for this new text in the same style.”
Another related technique: one-shot (exactly one example). As a rule of thumb: the more specific the desired format, the more examples help.
CONNECTIONS
Leadership
Leaders who use few-shot can calibrate AI outputs precisely to their communication style — for example by providing a sample email and then having many similar ones generated automatically. That is feedback culture in the prompting sense.
Agility
User stories have a defined format. Few-shot prompting with a sample user story enables fast, format-compliant generation of further stories. More consistent than a pure text description.
Project Management
Project charters, status updates or risk analyses have standard formats. Few-shot prompting with a sample charter significantly reduces the effort of creating new charters.
KEY POINTS
- Zero-shot: task without examples. Good for simple, clear requirements.
- Few-shot: include 1–5 examples, significantly better for specific formats.
- One-shot: exactly one example. Often sufficient as a compromise.
- The more specific the desired format, the more examples help.
- Few-shot reduces iteration loops: less correction, more result.
EXAMPLE
Zero-shot: “Write a short summary of this meeting.” → Result: generic, possibly not in the right format.
Few-shot: “Here is a sample meeting protocol and its summary in our house format: [example]. Now write a summary for this meeting: [protocol].” → Result: format correct, style consistent, immediately usable.
MISCONCEPTIONS
Do I always need to provide examples to get good results?
No. For common, clearly defined tasks (translate, summarise, answer questions) zero-shot works well. Few-shot pays off especially for specific formats, custom style requirements or when the zero-shot answer does not fit.
The more examples, the better?
Not necessarily. Too many examples can fill the context window, make the prompt unwieldy or force the model into too narrow a template. 2–3 good examples usually beat 10 mediocre ones.