Instructions and context

📘What it really is

• A prompt is everything you give the model: instructions, context, examples, the question.
• Good instructions are:
• clear and specific: say exactly what you want ("summarise in 3 bullet points for a busy manager"), not "summarise this";
• about the audience and purpose: who is it for, and why;
• explicit about format and length;
• explicit about limits: what not to do, and what to do when unsure ("If the policy doesn't cover it, say so");

• positive where possible: "write in plain English" works better than a long list of "don't"s.

• Context is the information the model needs that it doesn't have: your policies, the customer's history, today's date, the product list. The model only knows what's in its training and its context (Module 82).

• System prompts (in apps and APIs) hold standing instructions: the assistant's role, rules and tone, separate from each user's message.

• Structure long prompts with headings or tags, so the model can tell instructions from material:

You are a staff assistant for Sharma Textiles' customer team.
Answer only from the documents below. If they don't contain the answer,
reply: "I couldn't find that in our documents. Please check with a manager."

<documents>
...policy text...
</documents>

<question>
Can a customer return a tailored item after 10 days?
</question>

• Think of it as briefing a capable colleague who knows nothing about your business: what would they need to be told?