Skip to content
Service status

Type to search

Prompting Course: Basics · 1 min read

Practical tips and summary of the basics course

Last good habits before the advanced course: follow-up questions, structure, step-by-step instructions, and meta-prompting.

Section 5: Practical tips that make a difference

Section titled “Section 5: Practical tips that make a difference”

Here are the final good habits to take with you before you move on to the advanced course.

Dialogue and uncertainty
  • End with a follow-up question (“Do you need X to answer confidently?”) – reduces guessing and surfaces hidden assumptions.
  • Write what should happen if the answer is missing – e.g. ask clarifying questions, or refer to a channel/policy.
Context, steps, and reasoning
  • Context: who the response is for, what has already happened, the goal of the task.
  • Break up large tasks: Step 1 … Step 2 … so everything doesn’t end up as one unstructured blob.
  • Reason out loud for complex questions: ask the model to show its reasoning before the conclusion (more on that in the advanced course).
Structure and meta-help
  • Use clear headings in the prompt: BACKGROUND, TASK, IMPORTANT, FORMAT – the model follows structure more easily.
  • Mark priorities with words like IMPORTANT/CRITICAL or bold where warranted (no guarantee, but often helpful).
  • Stuck? Let a chat review your prompt or suggest a v2 based on what went wrong.

Congratulations – you’ve completed the Prompting Course: Basics.

You can now:

  • Build clear instructions with WHAT–WHY–HOW–WHO–TONE
  • Reinforce with realistic examples
  • Test and refine methodically
  • Use practical habits for safer and more useful assistants

Want to go deeper? Move on to the advanced course on context engineering, advanced structure, reasoning techniques, and RAG from a prompting perspective.

Good luck with your prompting!

Test your knowledge

Question 1 of 2

Question 1 of 2

Which advice most often reduces the risk of the model guessing when it lacks information?

Which advice most often reduces the risk of the model guessing when it lacks information?