Skip to Content
đźš§ This site is under development; its content is not final and may change at any time. đźš§
Basics of Prompt EngineeringUse explicit, actionable instructions

Use Explicit, Actionable Instructions

A prompt should tell the model what action to perform. Context without an action often produces a generic continuation; an action with no criteria produces a result that may look useful without meeting the teaching need.1

Turn background into a task

Compare these prompts:

This is a draft activity about supply and demand for first-year students. [activity]
Review the draft activity for first-year economics students. Check whether students must apply both supply and demand, whether the necessary information is present, and whether the task can be completed in 20 minutes. For each problem, quote the relevant wording and propose one revision. Do not add concepts that have not been taught. [activity]

The second prompt identifies the action, criteria, evidence, constraint, and desired revision. These details matter more than polite phrasing or an elaborate persona.

Use instructions at the point of need

  • Introduce source material with an action: “Use the excerpt below to…”
  • State boundaries beside the task: “Use only these sources for factual claims.”
  • Define missing-information behaviour: “Ask if an essential input is absent.”
  • Specify an output that supports review: “Return a table mapping each claim to its source passage.”
  • End with a check: “Compare the draft with every criterion before returning it.”

Avoid vague instructions such as “make it better”, “be accurate”, or “ensure quality”. Name what better means and how you will verify it.

Do not ask a model to score student work against improvised criteria. Supply the approved assignment and rubric, protect student data, and keep the final judgment with the educator. See Providing feedback with AI support.

A compact instruction pattern

Action: [create, compare, extract, critique, transform, or check] Object: [the material or question] Criteria: [what a successful result must contain] Evidence: [what the response may rely on] Constraints: [audience, scope, length, and boundaries] Output: [the most useful structure] Uncertainty: [what to do when information is missing] Verification: [checks before returning the result]

References & Footnotes

Footnotes

  1. Lin, Z. (2024). How to write effective prompts for large language models. Nature Human Behaviour, 8, 611–615. https://doi.org/10.1038/s41562-024-01847-2  ↩

Last updated on