Writing Clear Prompts

Be specific

Models do best with the instructions you would give a capable new colleague: context (who the users are, what the data is), the task, constraints and the output format. Vague prompts get generic answers.

Structure long prompts

Separate instructions from data with clear markers such as XML-style tags, so the model knows what is an instruction and what is content:

<ticket>...customer text...</ticket>
Summarise the ticket above in 3 bullet points: problem, impact, what the customer tried.

def build_prompt(ticket):
    return (f"<ticket>\n{ticket.strip()}\n</ticket>\n"
            "Summarise the ticket above for the on-call engineer in exactly 3 bullet points:\n"
            "- Problem\n- Impact\n- What the customer already tried\n"
            "If something is unknown, write 'unknown'.")

print(build_prompt("Checkout fails with error 502 since 9am. All EU customers affected. Cleared cache, no change."))

<ticket>
Checkout fails with error 502 since 9am. All EU customers affected. Cleared cache, no change.
</ticket>
Summarise the ticket above for the on-call engineer in exactly 3 bullet points:
- Problem
- Impact
- What the customer already tried
If something is unknown, write 'unknown'.

See Prompt engineering overview and Use XML tags.