What you will learn
  • Separate a task from its relevant background.
  • Write a request with a goal, context, constraints, and useful format.
  • Diagnose why an answer missed your intention.
  • Use a focused follow-up to improve a response.
  • Check an answer against evidence and your original request.

Before you begin

Read AI01 for the difference between AI and ordinary software. No programming is required; use a text assistant you are permitted to access.

Start with the result you need

Imagine asking a classmate, ‘Help with my project.’ They would need to know the subject, deadline, and part you are stuck on. An AI assistant faces a similar information problem, although it does not understand your circumstances the way a classmate does. It generates a response from available context and learned patterns. A useful request makes the missing information visible.

The text you send is called a prompt. It can include a question, instructions, examples, or source material. You do not need special punctuation or an impressive job title. Begin with an observable result: explain a difference, propose three options, identify a mistake, or help me practice. ‘Teach me robotics’ names a huge subject. ‘Explain what a sensor does using a desk-lamp example’ identifies a manageable result.

Before sending the message, finish this sentence privately: ‘After reading the answer, I should be able to…’ Put that outcome into the prompt. It becomes the standard you will use to judge the response. A long answer may contain correct statements and still fail to help you complete the actual task.

Give relevant context and boundaries

Context is background that changes what a good answer looks like. For an explanation, include what you already know and what confuses you. For a plan, include available time and equipment. For feedback, supply your draft and identify the part needing attention. Leave out personal information that does not affect the task. ‘I am new to coding’ is usually more useful than your school name.

Constraints are boundaries such as length, audience level, or excluded actions. ‘Keep it clear’ is less useful than ‘Define new terms and use one familiar example.’ A format tells the assistant how to organize the response: three paragraphs, a comparison table, five questions, or a procedure. Pick the format that makes the answer easiest to use.

A role such as ‘patient electronics tutor’ can suggest perspective and tone. It does not give the system qualifications or verified knowledge. Your concrete requirements still matter more. Keep them compatible: asking for an exhaustive textbook in fifty words creates a conflict that no clever phrasing can resolve.

A complete first conversation

Weak prompt: ‘Explain robots.’ The assistant must guess whether you want history, components, programming, or buying advice. A general response may be reasonable but unsuitable for your goal.

Improved prompt: ‘I am beginning robotics and know that computers run programs. Explain how a small obstacle-avoiding robot uses a sensor, controller, and motors. Define those terms, follow one obstacle through the sense–decide–act loop, and finish with two questions I can answer. Use about 250 words. Do not recommend products.’ This supplies starting knowledge, scope, length, and structure.

An appropriate response would explain that a sensor measures the surroundings, a controller reads that measurement and runs a rule, and motors create motion through suitable driving electronics. If the distance falls below a chosen limit, the program might stop. The explanation should acknowledge measurement errors and stopping distance. This is an illustrative expectation, not a recorded product test.

Follow-up prompt: ‘I understand the sensor, but I am confusing the controller with the motor driver. Explain their different jobs using the same robot.’ This targets the gap while preserving useful context. Verification: Compare the definitions with ROB02 and, when selecting hardware, its documentation. Check that the assistant has not suggested powering a motor directly from a signal pin.

Treat conversation as revision

Read the first answer for fit. Is it too advanced, missing a constraint, or answering another question? Name the mismatch. ‘Keep your example, but define actuator before using it’ is more actionable than ‘This is bad.’ You can ask the assistant to list assumptions so you can correct them.

Sometimes your goal changes. Say so directly: ‘I now want a checklist for selecting a beginner project.’ In a long conversation, restate the important decisions in a short working brief. Do not assume that every product remembers every earlier detail forever. Context limits and memory features vary.

For repeated tasks, save a successful prompt with placeholders for topic, audience, and format. Keep it readable. Anthropic’s prompting overview emphasizes defining success and testing instructions. Here, the test is whether the response helps you achieve the outcome you named, rather than whether the prompt sounds technical.

Check before relying on the answer

A fluent answer can contain an error, invented reference, or unnoticed assumption. ‘Are you sure?’ may produce reassurance without evidence. Identify which claims need checking, then perform a check outside the generated response. Consult official documentation for specifications, execute a small code example in an appropriate environment, or compare an explanation with course material.

Also check the request itself. Count requested options, inspect the audience level, and look for assumptions you prohibited. Separate ‘Did it follow the prompt?’ from ‘Is it correct?’ A response can pass one and fail the other. For learning, explain the answer in your own words. A polished paragraph you cannot explain has not yet accomplished its educational purpose.

Do you need perfect English or a long prompt? No. A narrow question often deserves a short request. State your meaning clearly and ask the assistant to confirm its interpretation when ambiguity matters. Ordinary courteous language is fine, but the useful details carry more weight than elaborate greetings.

Important terms

Prompt
The question, instruction, or material supplied to an AI system.
Context
Background relevant to completing the task.
Constraint
A boundary an acceptable answer should respect.
Iteration
Improving a result through successive revisions.
Verification
Checking a claim against independent evidence.

Mini project: Improve one explanation prompt

  1. Choose a familiar topic and write the result you want in one sentence.
  2. Add your starting knowledge, two constraints, and an output format. Send the prompt to an available assistant.
  3. Identify one specific weakness and send a follow-up addressing that weakness.
  4. Check one factual claim using a reliable source, then explain the topic without reading. Finish when your explanation answers your original goal.

Common mistakes and debugging

  • Giving only a subject: add an observable outcome.
  • Sharing irrelevant personal details: keep the context that changes the answer.
  • Changing five requirements at once: revise one important mismatch so you can see what helped.
  • Treating confidence as proof: inspect the underlying evidence.

Independent challenge

Request explanations of one topic for a curious ten-year-old and a beginner programmer. Identify which facts should remain unchanged and which examples should differ.

Check your understanding: 10 questions

  1. Why is a request tied to a particular robot more useful than ‘Explain sensors’?

  2. How does context differ from a constraint?

  3. Does assigning an expert role establish accuracy?

  4. What is a useful follow-up when one term remains confusing?

  5. Why check both prompt compliance and factual correctness?

  6. In your own words, what does “Prompt” mean?

  7. In your own words, what does “Context” mean?

  8. In your own words, what does “Constraint” mean?

  9. In your own words, what does “Iteration” mean?

  10. In your own words, what does “Verification” mean?

Quiz answers

Reveal all 10 answers after your attempt
  1. It supplies a concrete context and purpose instead of leaving the scope entirely open.
  2. Context explains the situation; a constraint limits acceptable results.
  3. No. It guides perspective or style, while factual claims still require verification.
  4. Name the term, say what you understand, and request a focused explanation using the existing example.
  5. An answer can follow the requested format but contain false claims, or be accurate while missing the actual task.
  6. The question, instruction, or material supplied to an AI system.
  7. Background relevant to completing the task.
  8. A boundary an acceptable answer should respect.
  9. Improving a result through successive revisions.
  10. Checking a claim against independent evidence.

Summary

State the outcome, add relevant context and boundaries, and improve the response through focused follow-ups. Check usefulness and correctness separately.

Continue learning

Continue with USE02 to turn a useful individual prompt into a repeatable method you can test.

Choose a connected learning path

Sources and further reading

Prepared 2026-09-18. Draft; conceptual review complete and cited guidance checked. Prompt outputs are illustrative, not recorded product tests.