Prompt Engineering for Beginners
Prompt engineering is the skill of writing instructions that get reliable, useful results from AI tools. For beginners it comes down to five habits: define the task and what “good” means, give context and examples, specify the output format, evaluate and improve the answer, and practice on real work. This free course teaches each habit in one short lesson with an exercise and a worked model answer.
On this page
Course overview
- 5 lessons, about 15–20 minutes each, plus a final project
- Works with ChatGPT, Claude, Gemini or Microsoft Copilot
- No coding. You need a free account with any AI assistant
- Each lesson: a principle, a before-and-after example, an exercise and a model answer
Lesson 1: Define the Task and Success Criteria
Most weak results come from unclear tasks, not from the AI. Before you type, answer two questions: What exactly should the AI produce? and How will I know it’s good? The second question is your success criteria, and writing them into the prompt is the fastest upgrade you can make.
| Before | After |
|---|---|
| Write a product description for my candle. | Write a product description for a soy candle that smells like cedar and orange, for an Etsy listing. Success: under 80 words, mentions the 45-hour burn time, no clichés like “cozy vibes,” ends with one line about gifting. |
A simple pattern
Task: [what to produce] Purpose: [what it will be used for] Success looks like: [2–4 checkable criteria]
Task: Write a message to my neighbors about a shared driveway repair. Purpose: get everyone to agree on a date and split the $1,200 cost. Success looks like: - under 120 words - states the cost per household for 4 households - proposes two possible dates - friendly, not pushy
Model answer: what to look for
The message should show $300 per household, give two specific dates, and ask for a reply by a deadline. If the AI wrote 200 words or skipped the math, your criteria weren’t followed: say “You missed criteria 1 and 2. Fix them.”
Why it works
Checkable criteria (word count, a number, two dates) let you, and the AI, verify the result. “Make it good” can’t be checked.
Lesson 2: Add Context and Examples
The AI knows a lot about the world but nothing about your situation. Context is background it can’t guess: your audience, your constraints, what you’ve tried. Examples show the style or structure you want. Giving one to three examples is often called “few-shot prompting,” and it is one of the most reliable techniques there is.
Useful context to include
- Who the reader is and what they already know
- Your role (“I’m a new team lead,” “I run a two-person bakery”)
- Constraints: budget, deadline, tone, words to avoid
- Source material the answer must be based on, pasted in full
I write short tips for our clinic’s Instagram. Here are two I liked: Example 1: “Drinking water before a blood test? Yes, it actually helps. Plain water only.” Example 2: “Running late? Call us. We can often shift your slot instead of rebooking.” Write 5 more tips in the same style about preparing for a dental cleaning. Same length, same tone, no emojis, no medical claims we can’t back up.
Model answer: what to look for
Five one- or two-sentence tips that start with a question or a hook, like the examples. If they’re longer or use emojis, the AI ignored your examples: point to the difference.
Lesson 3: Specify the Output Format
Tell the AI exactly what shape the answer should take. Format instructions save editing time and make answers easier to check.
- Length: “under 100 words,” “exactly 5 bullet points,” “one paragraph.”
- Structure: “a table with columns Task, Owner, Due date,” “headings for each day.”
- Style: “plain English, 8th-grade reading level,” “no jargon.”
- Machine-friendly output: “CSV I can paste into Excel,” “JSON with keys name and email.”
Turn these notes into a project plan. Format: a table with columns Step, Owner, Start, Due, Depends on. Use dates in MM/DD format. After the table, list the 2 biggest risks in one sentence each. Notes: website relaunch, Sam does copy (1 week), Lee does design after copy (2 weeks), dev by Ana after design (2 weeks), launch June 30, today is May 12
Model answer: what to look for
Steps in dependency order, with dates that leave enough time before June 30. Check the date math yourself: AI often miscounts days across months.
The same technique in different tools
Format instructions work in every major assistant, with small differences:
- ChatGPT: can create downloadable files (for example, a spreadsheet) when you ask, depending on your plan. See the beginner ChatGPT tutorial.
- Claude: responds well to clearly separated sections. Wrapping source text in tags such as
<notes>…</notes>helps it tell instructions from material. See how to use Claude. - Gemini and Copilot: handle tables well and can export or insert results into Google or Microsoft apps on eligible plans.
Lesson 4: Improve and Evaluate Answers
No prompt works perfectly the first time. Skilled users evaluate the answer against their criteria and iterate with specific feedback. AI models still make mistakes, including confident ones, so evaluation isn’t optional.
A four-step review
- Criteria check: Does it meet each success criterion from Lesson 1?
- Fact check: Are names, numbers, dates and quotes correct? Verify anything you’ll publish or act on.
- Gap check: Ask the AI, “What’s missing or weak in your answer?”
- Targeted fix: Give feedback on specific parts instead of “make it better.”
| Vague feedback | Specific feedback |
|---|---|
| Make it better. | The second paragraph repeats the first. Cut it and add one concrete example instead. |
| Too formal. | Use contractions and shorter sentences, like you’re talking to a coworker. |
| That’s wrong. | The deadline is June 30, not June 13. Recalculate the schedule from that date. |
Review your previous answer against these criteria: 1) under 150 words 2) includes the price 3) clear call to action. For each criterion, say pass or fail and why. Then rewrite the answer to pass all of them.
Lesson 5: Complete a Practical Project
Put all five habits together on one realistic task. Choose the version that fits you best.
You are helping me write my weekly status report for my manager, who reads it on her phone. Context: I'm a marketing coordinator. This week: launched the spring email (open rate 41%), the landing page is delayed 3 days waiting on legal, and I need approval for a $500 ad test. Example of a report she liked: "Done: ... / In progress: ... / Blocked: ... / Need from you: ..." Format: the four headings above, max 2 bullets each, under 120 words total. Success: numbers are exactly as given, the approval request is the last line. Before writing, ask me one question if anything is unclear.
Model answer: what to look for
Four short sections, the 41% open rate unchanged, the 3-day delay under “Blocked,” and “Approve $500 ad test?” as the final line. Under 120 words.
Your deliverable
- Run the prompt and get a report that passes every criterion
- Save the final prompt as a template with the highlighted parts as blanks
- Use it again next week with new facts and compare the time it takes
Where to go after this course
- Build a library of templates: start with our practical ChatGPT prompt examples.
- Apply prompting to real job tasks: how to use AI at work.
- Want a formal certificate? Compare AI courses, including prompt engineering courses from universities and AI companies.
Sources
Techniques in this course are consistent with official prompting guides from AI developers. Checked November 18, 2026.