Few-shot prompting

Few-shot prompting is a way of teaching a task inside the prompt by adding a few solved examples as earlier user and assistant turns.

Last updated: 30 Sep, 2026 · groq 1.7 · gpt-oss-120b on Groq

Describing a task only goes so far. Showing solved examples often works better. This lesson continues the file from System prompts: ask, tickets and structured come from there.

Three solved examples

examples = [
    {"role": "user", "content": "You took money from my card two times"},
    {"role": "assistant", "content": '{"category": "billing", "priority": 4}'},
    {"role": "user", "content": "Where is my delivery? It is a week late"},
    {"role": "assistant", "content": '{"category": "shipping", "priority": 3}'},
    {"role": "user", "content": "Can I change the email on my account?"},
    {"role": "assistant", "content": '{"category": "other", "priority": 2}'},
]

Three made-up tickets, one per category, each followed by the answer you want. They go between the system prompt and the real ticket, so the model reads them as a conversation it has already had. None of them is one of the seven test tickets.

Syntax: [system, *examples, user]. *examples spreads the list's items into the new list.

Sorting with the examples in front

for text, expected in tickets:
    messages = [{"role": "system", "content": structured}, *examples, {"role": "user", "content": text}]
    print(f"{expected:9} {ask(messages)}")

billing   {"category": "billing", "priority": 4}
shipping  {"category": "shipping", "priority": 3}
billing   {"category": "billing", "priority": 4}
other     {"category": "other", "priority": 2}
shipping  {"category": "shipping", "priority": 4}
billing   {"category": "shipping", "priority": 4}
other     {"category": "shipping", "priority": 4}

• The same five tickets are right, in the same JSON shape.
• The fee and the cancellation are still shipping. The examples show a double charge, a late delivery and an account change. None is about a fee or a cancellation, so they gave the model nothing to go on for those two.
• Two priorities changed from the structured-prompt run, and the categories did not.

Why examples change the answer

In Next-token prediction, every token in front of the answer changes the next token's scores. Solved examples put answers in front of the model: after a double charge came billing, so billing becomes likelier after the next charge. The model is not trained by them; its weights do not change. It continues a pattern it can see.

Instructions vs examples

When examples help most

• When the output format is fiddly and easier to show than to describe.
• When the model keeps making the same mistake: an example of exactly that case.
• With smaller models, which follow patterns better than long instructions.

Related

• Previous: System prompts
• Next: Structured output
• Reference: OpenAI prompt engineering guide