System prompts

A system prompt is the first message in the list, with role system, that tells the model its job, its rules and the exact shape of the answer.

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

The ticket sorter starts here: sort an online shop's support tickets into billing, shipping or other, with a priority, as JSON a program can read. A vague instruction goes first.

from groq import Groq

client = Groq()
MODEL = "openai/gpt-oss-120b"


def ask(messages, **settings):
    response = client.chat.completions.create(model=MODEL, messages=messages, **settings)
    return response.choices[0].message.content

Seven labelled tickets

tickets = [
    ("I was charged twice for one order", "billing"),
    ("My parcel has not arrived", "shipping"),
    ("Can I get a refund for the blue mug?", "billing"),
    ("How do I change my password?", "other"),
    ("The parcel arrived but the box was crushed", "shipping"),
    ("The courier asked me to pay a delivery fee at the door", "billing"),
    ("I want to cancel my order before it ships", "other"),
]

Each ticket has the category a person chose. The last two are harder than they look: a fee asked for at the door is a charge, and cancelling an order is not about delivery.

A vague instruction

vague = "Sort this support ticket into billing, shipping or other and give it a priority."

Sorting the tickets with the vague prompt

for text, expected in tickets:
    answer = ask([{"role": "system", "content": vague}, {"role": "user", "content": text}])
    print(f"{expected:9} {answer!r}")

billing   '**Category:** Billing  \n**Priority:** High (duplicate charge requires prompt attention)'
shipping  '**Category:** Shipping  \n**Priority:** High'
billing   '**Category:** Billing  \n**Priority:** High'
other     '**Category:** Other  \n**Priority:** High'
shipping  '**Category:** Shipping  \n**Priority:** High (the package arrived damaged and may require replacement or refund)'
billing   '**Category:** Shipping  \n**Priority:** High'
other     '**Category:** Shipping  \n**Priority:** High   (time‑sensitive request to cancel before the order ships)'

• No answer is JSON. The model chose its own shape, Markdown with bold labels, so a program cannot read it.
• Every priority is High, with no scale to measure against.
• The fee and the cancellation are filed as Shipping.

A structured prompt

structured = """You sort customer support tickets for an online shop.

Categories:
- billing: payments, charges, refunds
- shipping: parcels, delivery, damaged boxes
- other: anything else

Reply with one line of JSON and nothing else, like this:
{"category": "shipping", "priority": 3}
priority is 1 (can wait) to 5 (urgent)."""

It says who the model is, defines each category, shows the exact output with one sample line, and gives the priority a scale. Triple quotes let a string run over several lines.

Sorting the tickets with the structured prompt

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

billing   {"category": "billing", "priority": 5}
shipping  {"category": "shipping", "priority": 3}
billing   {"category": "billing", "priority": 4}
other     {"category": "other", "priority": 2}
shipping  {"category": "shipping", "priority": 3}
billing   {"category": "shipping", "priority": 4}
other     {"category": "shipping", "priority": 4}

• Every answer is one line of JSON in the shape the prompt showed.
• Five of seven categories are right.
• The fee and the cancellation are still shipping. The prompt never said where a fee at the door or a cancellation belongs, so the model filed them by the word delivery and the word ships.

Vague vs structured prompt

What to put in a system prompt

• The role: who the model is and what it is for.
• Definitions: what each category or field means, with the edge cases you know about.
• The output shape, with one sample the model can copy.
• Scales and limits, such as 1 to 5, or one sentence.

Related

• Previous: Chat messages and templates
• Next: Few-shot prompting
• Reference: OpenAI prompt engineering guide