Write prompts that get reliable, usable results from ChatGPT, Claude and Gemini.
Read the guide: How to Learn Prompt Engineering: A Free 6 Week Plan
Direct Answer: Write prompts that get reliable, usable results from ChatGPT, Claude and Gemini: structured instructions, examples, reasoning, exact output formats, custom assistants, testing and safety, for writing, data, code and images.
Understand what prompt engineering is, how models break text into tokens and hold it in a context window, why they invent facts and drift off task, and how randomness affects the answers you get.
Build prompts from four parts, role, task, context and format, write instructions and constraints the model follows, organise long prompts with delimiters, and use examples to show exactly what good output looks like.
Improve answers to problems that need thought: ask for step-by-step reasoning, split big jobs into chains of smaller prompts, have the model critique and revise its own work, and choose when a reasoning model is worth its cost.
Get output in exactly the shape you need: tables, lists and fill-in templates, valid JSON for software, the right length, tone and reading level, and checker prompts that confirm an output meets the rules before you use it.
Use AI on real documents without losing accuracy: write summaries that keep what matters, extract data from PDFs, emails and forms, answer questions with supporting quotes, and compare or combine several sources.
Write the standing instructions behind custom assistants: what a system prompt controls, how to structure one for a GPT, Project or Gem, how to set personas, guardrails and refusals, and what knowledge files and memory can and cannot do.
Apply prompting to content work: capture a brand voice that stays consistent, produce social posts and ad variations at scale, draft long-form articles that rank and read well, and use AI as an editor rather than a ghostwriter.
Use AI as a coding and data assistant: specify code requests precisely, debug and understand unfamiliar code, turn plain language into spreadsheet formulas and SQL, and analyse data with AI while guarding against confident mistakes.
Prompt image, video and voice tools with intent: structure image prompts by subject, setting, style and composition, iterate towards a consistent look, write prompts for video and voice generation, and respect rights, likeness and disclosure rules.
Treat important prompts like products: build a test set of realistic inputs, score outputs with clear rubrics, compare prompt versions fairly before switching, and keep a versioned prompt library your team can trust.
Protect assistants and workflows from prompt injection, keep confidential and personal data out of places it should not go, design limits that hold against misuse, and review AI output properly before it reaches customers or the public.
Turn prompting into professional value: work across ChatGPT, Claude and Gemini, build a portfolio that proves results, sell AI content and automation services as a freelancer, and keep your skills current as tools change.
Topic checks
Each one opens the next topic
Chapter quizzes
Each one opens the next chapter
50-question final
50 of 100 marks to pass
Certificate
Issued instantly, verifiable
One attempt, 60 minutes. If you don't pass, you can retake once after 30 days.
To learn prompt engineering, stop collecting "magic prompts" and start practising one habit: write instructions the way you would brief a capable new colleague. Give the model a clear goal, the background it needs, an example of good work and the exact shape of the answer you want. Then test that prompt on several real inputs, not just one. Learn how models read text, learn four building blocks, practise on your own daily work, and measure whether your prompts actually get better. That is the whole skill, and you can learn it for free in about six weeks of steady study with our free Prompt Engineering course.
Prompt engineering is writing instructions that get a language model to produce the output you need reliably, not once by luck. The prompt is everything the model sees: your request, any examples, pasted documents, and the hidden instructions the app adds behind the scenes.
The difference shows up quickly. Compare these:
| Weak prompt | Strong prompt |
|---|---|
| Write a post about our sale | Write a 60 word Instagram caption for a 30% Eid sale on women's lawn suits at our Lahore store. Friendly tone, one emoji, end with "Shop now, link in bio". |
| Summarise this report | Summarise this report for a busy CEO in five bullet points, each under 20 words, focusing on costs and risks. |
| Fix my code | This Python function should return the total of paid orders but returns 0. Explain the bug, then give the corrected function. |
None of that is a trick, just a better brief. Clear communication is what transfers between ChatGPT, Claude, Gemini and whatever comes next.
You do not need maths, but you do need a mental model. Language models read text as tokens, not letters, which is why they are unreliable at counting characters or reversing words. They hold your conversation in a context window, so very long chats can push early instructions out of focus. They predict likely text, which is why they can invent facts, quotes and references that look real. And they have randomness built in, so the same prompt can give different answers.
If you need a meta description under 150 characters, ask the model to draft it, then check the length with a spreadsheet formula like LEN instead of trusting its count.
Almost every good prompt has four parts:
Add delimiters (tags or clear markers around pasted text) so the model can tell your instructions apart from your data. Then add examples. One or two examples of the output you want usually do more than a paragraph of description. Models copy examples closely. If every example starts with "Introducing", most outputs will too.
For anything that needs thought, ask the model to work step by step before giving its answer. For big jobs, split the work into a prompt chain: one prompt for an outline, another for the draft, another for the edit. Then try a self critique loop, where the model reviews its own draft against your criteria, lists the problems and rewrites.
Real work needs a fixed shape: a table, valid JSON, a template, or text at the right length, tone and reading level. A concrete description of the reader ("a shop owner with no technical background, short sentences, no jargon") sets the reading level far better than "write simply".
Use prompting on your actual tasks: summarising long PDFs, extracting data from emails, drafting social posts in a consistent brand voice, writing spreadsheet formulas and SQL from plain language, or debugging code. Go deep on two areas that match your job.
Most self taught people skip this step. Build a small test set of realistic inputs, including the awkward ones. Score each output with a simple rubric, ideally pass or fail criteria rather than a vague 1 to 10 scale. Change one thing at a time and compare versions on the same test set. Keep your best prompts in a versioned library so you know which one is current.
Plan on roughly an hour a day, five days a week, split between lessons and practice on your own tasks.
| Week | Focus | Practice task |
|---|---|---|
| 1 | How models work, the four building blocks | Rewrite five prompts you used this month using role, task, context and format |
| 2 | Examples, reasoning, prompt chains | Build a three step chain for one recurring job, such as a weekly report |
| 3 | Output control, documents, long context | Extract fields from ten real emails or invoices into a table |
| 4 | System prompts and custom assistants | Write instructions for a simple assistant that answers questions about one policy |
| 5 | Writing, marketing, code, data and images | Produce a batch of on brand posts or a set of working spreadsheet formulas |
| 6 | Testing, safety, career | Score two versions of one prompt on 20 inputs and write it up as a case study |
The Prompt Engineering course is rated Beginner to Intermediate, is planned for six weeks, and has 12 chapters with four lessons each. Every lesson ends with a short quiz.
The course suits students, office workers, marketers, freelancers, support staff and developers who already use a chatbot and want reliable results instead of lucky ones. You need no coding background.
If you have never used an AI chatbot at all, start one level lower with AI Tools for Everyone and our guide on how to learn how to use AI tools for free. If you want to go further and connect prompts to real systems, AI Automation and Agents and AI Engineering are the natural next steps.
Six weeks is the planned pace. Taking longer because you practise on your own work is the better outcome.
Each lesson has a 90 second reading time before its quiz unlocks, so you cannot just click through. The certificate comes from a final exam of 50 questions in 60 minutes, taken in one timed attempt. If it does not go well, one retake opens 30 days later. Leaving the exam tab submits it.
Be honest with yourself about what a certificate proves. The course itself ranks a certificate with no project attached below a tested case study. Pair it with two or three case studies: the problem, your prompt, your test set, and how the results improved between versions. We cover this in are free online certificates worth anything.
05 questions
Yes. The free plans of ChatGPT, Claude and Gemini are enough to practise the core text skills, and our course, quizzes and certificate are free.
No. Most prompt engineering is clear writing and careful testing. Coding helps if you want to connect prompts to software or output JSON for an app, but beginners can learn the core skill without it.
Plan for about six weeks at an hour a day to cover the fundamentals and apply them properly. Real fluency keeps growing as you use the skill on your own work.
The job title may come and go, but the skill keeps its value. Turning a messy request into a clear brief, then testing whether the output holds up, is useful in marketing, operations, analysis, support and development roles.
Build a small portfolio. Show a real problem, your prompt, a test set, and how pass rates improved from your first version to your last. That says far more than a list of clever prompts. Ready to start? Open the free [Prompt Engineering course](/learn/prompt-engineering) and finish the first chapter this week.
