Prompt Engineering Essentials
A tool-agnostic beginner course in writing clear prompts, testing and fixing them, checking AI answers, keeping private data safe, and building a reusable prompt library for your own work.
One payment of $39. Instant online access to the full written course. No subscription. Refund policy.
You will build a personal prompt library: a set of tested, reusable prompts for your own work, each with a checklist, a test record and a safety note. You will be able to write clear prompts, test them, fix them one part at a time, and check AI answers instead of trusting them.
Who it is for. Complete beginners who use, or want to use, an AI assistant for everyday work or study: office workers, small-business owners, freelancers, students and career changers. No coding, math or prior AI experience is needed. It also suits people who already use AI but get uneven results and want a repeatable way to improve them.
You finish with. A Prompt Library for Your Own Work. One document or folder that a person who has never seen your work could open and use. It holds a README with your safety rules, an index, at least six tested prompt cards for tasks from your own work or study, and a change log. Every card carries a template, a fill guide, a yes or no checklist, a test record and a stakes level, and every score in it is one you recorded yourself.
Certificate. Finish every lesson, resolve every quiz question with at least 50% right on the first try, and tick the capstone checklist — Apex Flow Academy issues a verifiable Certificate of Completion with a unique ID and a public verification page. It is a certificate of completion, not a degree, licence, accreditation or exam result.
01 · Start Here: How Prompts Work and How to Practise4 lessons
Learn what a prompt is and what the model does with it, set up a free place to practise, learn the routine that makes every later lesson reliable (run it several times, score it against a checklist, keep a log), and get the five-part map you will use for the rest of the course.
- 1.1What a Prompt Is and What the Model Does With ItFREE
- 1.2Set Up Your Practice Space and Prompt Log
- 1.3Why One Try Is Not a Test: The Compare Routine
- 1.4The Five Parts of a Good Prompt
02 · Say What You Want: Task, Context and Audience4 lessons
Write a clear one-sentence task, give the model the facts it cannot guess, name the reader and the role, and let the model interview you when you cannot tell what is missing. These four moves fix most weak prompts.
- 2.1Write the Task in One Clear Sentence
- 2.2Give the Context the Model Cannot Guess
- 2.3Name the Audience and the Role
- 2.4Let the Model Ask You Questions First
03 · Shape the Answer: Format, Limits and Examples4 lessons
Decide the shape of the answer before the model does, set limits you can check, show the model what good looks like with examples, and ask for fixed fields you can paste into a spreadsheet or check with a small script.
- 3.1Ask for the Format You Need
- 3.2Set Limits You Can Check
- 3.3Show What Good Looks Like With Examples
- 3.4Ask for JSON or a Table You Can Reuse
04 · Handle Bigger Jobs: Steps, Chains and Long Documents4 lessons
Ask for visible working when a task has several parts, break a big job into a chain of small prompts, use a draft, critique and revise loop, and work with long documents by putting the text first and asking for quotes before answers.
- 4.1Ask for Steps When the Work Needs Checking
- 4.2Break a Big Job Into a Chain
- 4.3Draft, Critique, Revise
- 4.4Work With Long Documents
05 · Check, Fix and Test Your Prompts4 lessons
Diagnose which part of a prompt failed, build a small test set and score two versions against it, catch made-up facts before they reach anyone, and turn a winning prompt into a reusable template with placeholders.
- 5.1Diagnose a Bad Answer: Which Part Failed?
- 5.2Build a Test Set and Score Two Versions
- 5.3Catch Made-Up Facts
- 5.4Turn a Winner Into a Reusable Template
06 · Limits, Privacy and Your Finished Library4 lessons
Learn the limits of AI answers beyond made-up facts, keep private data out of your prompts, protect yourself from instructions hidden in pasted text, and assemble, version and maintain your prompt library as the capstone.
- 6.1Know the Limits: Old, One-Sided and High-Stakes Answers
- 6.2Keep Private Data Out of Your Prompts
- 6.3Pasted Text Can Give Orders: Hidden Instructions
- 6.4Assemble, Version and Maintain Your Library
What a Prompt Is and What the Model Does With It
You will be able to
- Say in one sentence what a prompt is and what the model does with it.
- Spot the missing facts in a vague prompt before you send it.
Why this matters
Most disappointing answers do not come from a weak tool. They come from a prompt that left out something the model had no way to know. Once you see this, you can fix most problems by adding the missing piece instead of guessing or starting over.
Learn it
A prompt is the text you give an AI tool. It can be a question, a request, a pasted email, or all three. The model is the program behind the tool. It was trained on a very large amount of text, and it works by predicting which words should come next. The text it sends back is the output. One exchange, or a whole thread of them, is a chat.
Three facts follow from this. The rest of the course builds on them.
- The model only has your words. It does not know your business, your customer, your deadline or your taste. Some tools can also search the web or read files you upload. Check what your tool does, but never assume it knows your situation.
- Gaps get filled with the most common guess. Ask for a message to customers and you get a message that would suit almost any business. That is why the output feels generic.
- Fluent does not mean true. The model writes smooth sentences whether or not the facts are right. Modules 5 and 6 cover how to check.
Here is a simple test to run before you send any prompt. Imagine handing it to a smart stranger who has never heard of you. Would they need to ask questions before they could start? Every question they would ask is a gap in your prompt. Anthropic's prompting guide for its Claude models gives the same advice. Show your prompt to a colleague who knows little about the task. If they would be confused, the model probably will be too.
Use these steps:
- Write your prompt the way you normally would.
- Read it as the stranger. Write down every question you would ask.
- Answer each question in a short line and add the lines to the prompt.
- Send it and compare the result with your first try.
See it in action
Example (illustrative numbers). You own a small bakery. You raise the price of a large loaf from $6 to $7. Your first prompt is:
Write a message to my customers about the price change.The stranger's questions: Which business? What changed, and by how much? When does it start? Why? What tone? Where will the message appear, and how long should it be?
You answer each one and send this instead:
Write a short message to customers of my small neighborhood bakery. From next Monday the large sourdough loaf goes from $6 to $7 because flour and energy costs went up. Keep the tone warm and honest, no more than 80 words. It will go on our shop window sign and our Instagram page.Example of what you might get from the first prompt: a friendly note with blanks such as [business name] and [date], and vague words about rising costs. From the second prompt you might get a message that names the loaf, the new price and the start day. Your results will differ, and no tool gives the same answer every time. The point is how much more usable the second one is.
Common mistakes
- Treating the tool like a search engine. A search engine finds pages that already exist. A model writes new text from patterns, so it needs instructions, not keywords. Write full sentences that say what you want done.
- Assuming it remembers you. Many tools start each new chat with a blank slate, and memory settings differ by tool. Put the facts that matter into the prompt.
- Believing a longer prompt is always better. Add what is missing, not what is repeated.
You are done when
You can take any prompt you wrote today and list three questions a stranger would ask about it. You have also added the answers to one of your own prompts.
You finished the free lesson
That is one lesson from the course. The full course gives you every remaining lesson, a quick check and a hands-on task in each one, and the workbook of templates and checklists.
Full course$39
- A computer or phone with an internet connection
- A free account with at least one AI assistant. Examples: ChatGPT, Claude and Gemini each had a free plan when this course was checked in September 2026. Check each tool's current features, usage limits and pricing, because they change often. Every core lesson can be done on a free plan
- A second assistant if you can open one. It is helpful for the comparison exercises but not required
- A place to keep notes: a spreadsheet or a document with tables. Free options exist, such as Google Sheets and Docs or LibreOffice
- A word counter. Most document tools have one built in, and free websites offer them too
- A handful of real tasks from your own work or study, described with made-up or cleaned-up details, never private information
- Optional and never required: Python or Node.js, free to install, if you want to run the small checking scripts described in some lessons
- Optional and never required: a paid plan on any tool. Paid plans usually raise usage limits and unlock extra features, and the course does not depend on them
- Comfort with a web browser, copy and paste, and a document or spreadsheet
- No coding, statistics or prior AI experience needed
- If you plan to practise on work material, permission to use AI tools at work. Otherwise use made-up or cleaned-up material, which Module 1 shows you how to prepare
A Prompt Library for Your Own Work
One document or folder that a person who has never seen your work could open and use. It holds a README with your safety rules, an index, at least six tested prompt cards for tasks from your own work or study, and a change log. Every card carries a template, a fill guide, a yes or no checklist, a test record and a stakes level, and every score in it is one you recorded yourself.
- A README with how to use the library, your context pack, your never-paste list, your stakes ladder, your approval rule and your retest routine
- An index listing each card, when to use it, its stakes level, version and last-tested date and tool
- At least six library cards, including one chain, one critique and revise pair, one structured-output card and one long-document card
- For each card: a template with six or fewer named placeholders, a fill guide with an example for every placeholder, and a yes or no checklist of three to six checks written before testing
- For each card: a test note with the date, the tool and a score from at least two runs, plus a stakes level and a privacy line
- Two-version scorecards on a five-input test set for at least three cards, and at least one card tried on a second tool
- A fact-check table for one real output, showing each claim, where you checked it and the result
- A change log with at least three entries
- Prompt
- The text you give an AI tool. It can be a question, a request, pasted material, or all three.
- Model
- The program behind an AI tool. It was trained on a large amount of text and writes by predicting what words should come next.
- Output
- The text the AI tool sends back to you.
- Chat
- One exchange, or a whole thread of exchanges, with an AI tool. Earlier messages in the same chat are part of what the model reads.
- Temperature
- A setting that controls how adventurous the model is when it picks its words. Lower is more repeatable and higher is more varied. Chat apps often hide it.
- Token
- A small chunk of text, often part of a word, that the model reads and writes. Models count tokens, not words, so exact word counts are hit or miss.
- Context
- The background facts and source text you put in a prompt so the model does not have to guess.
- Context pack
- A short saved paragraph about you, your business or your reader that you paste at the top of many prompts.
- Marker (delimiter)
- A line of dashes or a tag such as <notes> that shows where pasted material starts and ends, so it is not confused with your instructions.
- Zero-shot
- A prompt with no examples in it.
- Few-shot
- A prompt that includes a small number of examples, each with a sample input and the output you want.
- JSON
- A way to write labeled data as names and values inside curly braces, so a spreadsheet or program can read it.
Browse the full Academy encyclopedia
23 checks were run and recorded while writing this course (code, formulas, commands and facts), and it lists 15 official sources it was checked against. Prices, features and policies of outside tools can change, so check each tool's own website.
Created by Apex Flow Academy with AI assistance. For education only; not legal, tax, financial or medical advice. Results depend on your effort and circumstances.
Python From Zero: Write Programs That Work
You write and run real Python programs from the first lesson, and you finish with three small working tools that you built yourself, tested with checks you wrote, and can change without starting over.
Python Automation: Make the Computer Do Your Boring Work
You automate files, spreadsheets, email drafts and web lookups that eat your week. You finish with your own script that plans before it acts, keeps a log, is tested against sample data, and runs on a schedule.
SQL: Ask Any Database Any Question
You will be able to write SQL that answers business questions from a database: choose and sort columns, filter rows, summarise and group, join tables, and check that your numbers are right. You finish with a report you built and checked yourself.