Basics of Prompting

Prompting is the practice of describing a task so that a generative AI system can produce a useful draft or analysis. Clear prompts improve the odds of a good result, but they do not guarantee accuracy. The most reliable workflow combines a well-specified task, appropriate source material, explicit success criteria, and human verification.¹

Prompting guides

• Assign a useful role or perspective explains when a role helps—and why it does not create expertise.
• Use explicit, actionable instructions turns background information into a checkable task.
• Separate instructions, sources, and output uses headings, tags, or other boundaries to organise long prompts.
• Plan, verify, and explain replaces generic “think step by step” advice with checkable criteria and validation.
• Provide examples to guide the output uses examples to demonstrate the intended structure, tone, or classification.
• Further resources collects additional prompting material.

A practical default

Start with a compact prompt containing:

1. Goal: what you need and why.
2. Context: audience, level, and relevant circumstances.
3. Source material: the information the response should use.
4. Success criteria: what must be true of the result.
5. Constraints: scope, tone, length, and boundaries.
6. Output format: headings, table columns, or another useful structure.
7. Verification: checks to perform and uncertainty to report.

For example:

Draft three formative quiz questions for first-year economics students using
only the supplied reading. Each question must test a different learning outcome.
Return a table with the question, expected answer, learning outcome, difficulty,
and the passage that supports the answer. Flag any learning outcome that the
reading does not support; do not invent missing material.

Review is part of prompting

Treat the first response as a draft. Check claims and citations, compare the result with your criteria, look for missing perspectives, and revise the prompt based on observed failures. For consequential or repeated workflows, test several representative inputs and record what the human reviewer changed.

Never include personal or confidential information merely to improve a prompt. See When to use generative AI and Responsible AI use before using real student material.

References & Footnotes