The short version: you learn prompt engineering by using an AI tool on real work every single day for two to three weeks, not by memorising formulas from a course. The four things that matter are giving the AI a role, a task, context about your business, and the format you want back. Everything else is decoration.
Why I even bother writing this
Three years ago I sat in front of ChatGPT with a blank cursor and no idea what to type beyond "write me a blog post about marketing." It gave me back something so generic I could have found it on any content mill site in 2015. I remember thinking the tool was useless. It wasn't. I was useless at talking to it.
That gap between "this tool is rubbish" and "I don't yet know how to ask" is where most people give up on AI. They try one lazy prompt, get a lazy answer, and decide prompt engineering is either too technical for them or a scam word invented by consultants (myself included some days) to sell courses. Neither is true. It's a skill, and like any skill, it's built through repetition on things you care about getting right.
The uncomfortable bit nobody selling a course wants to say out loud
Here's the truth I don't see written plainly enough: prompt engineering as a technical discipline, the kind you'd put on a CV, is already fading. The models are getting better at understanding messy, human instructions, which means the clever formulas people sold in 2023 (the seven-part frameworks, the "magic words" that supposedly triple output quality) matter far less in 2026 than the basic discipline of knowing your own business well enough to explain it clearly. I've watched clients spend £400 on a "prompt engineering masterclass" and come out able to recite an acronym but still unable to write a brief that a junior employee could follow either. The skill was never really about the AI. It's about being able to explain what you want, in writing, to anyone. AI just exposes whether you can do that.
So the honest starting point isn't a framework. It's this question: could you hand this exact instruction to a smart 22 year old on their first day and get roughly what you wanted? If not, no prompt trick will save you.
The four building blocks that do the heavy lifting
Strip away the jargon and every decent prompt has four parts. I use these with every client I train, and I use them myself dozens of times a day.
- Role: tell the AI who it's pretending to be. "You are a direct response copywriter who writes for busy SME owners" gets a different output than no role at all.
- Task: the specific thing you want done, in one clear sentence. Not "help with marketing," but "write three subject lines for an email announcing a price increase."
- Context: the facts it needs and couldn't guess. Your industry, your audience, your tone, your constraints (word count, deadline, what's already been tried and failed).
- Format: how you want it back. A table, a numbered list, three options, plain text with no headers. Say it, don't assume it.
Miss context and you get generic slop. Miss format and you'll spend more time reformatting the output than it would have taken to write it yourself. I've timed this: a prompt with all four parts takes me maybe forty seconds longer to write than a lazy one, and it saves me on average ten to fifteen minutes of back-and-forth editing. That ratio is the entire business case for learning this (sorry, I mean thoroughly, we're not allowed that word).
A real example from my own desk
Last month I was writing a LinkedIn post about a client win and wanted it to sound like me, not like a corporate press release with my name stapled to it. My first prompt was lazy: "write a LinkedIn post about helping a client get more leads." What came back was fluent, polished, and utterly forgettable. Nobody talks like that, including robots that are pretending to.
I rewrote it with role, task, context and format: "You are me, Lilach Bullock, a blunt British marketing consultant who writes short punchy sentences and never uses corporate buzzwords. Write a LinkedIn post (under 200 words, no hashtags, one line per idea) about a client whose lead volume went up 34 percent in six weeks after we fixed their email sequence. Include one specific number and one line of self-deprecating humour."
That version needed two small edits. The lazy version would have needed a full rewrite. I counted, out of curiosity, how many prompt versions it took me to land on a formula I now reuse for every LinkedIn post: 11. Eleven attempts, over about ninety minutes, spread across a single afternoon two years ago. That's the actual learning curve. Not a weekend course. An afternoon of trial, annoyance, and small adjustments, repeated until a pattern clicked.
Step by step: how to learn this in two weeks
This is the process I give clients who ask me to teach their team, condensed into something you can do without paying anyone.
- Day 1 to 3: pick one tool (ChatGPT, Claude, or Gemini, doesn't matter which for basics) and use it for one recurring task you already do weekly, like drafting emails or summarising meeting notes. Don't experiment with ten different use cases. One task, repeated, teaches faster than ten tasks tried once.
- Day 4 to 7: start adding the four building blocks deliberately to every prompt. Write them out in full sentences at first, even if it feels clunky. You're building a habit, not writing poetry.
- Day 8 to 10: practise giving feedback inside the same conversation instead of starting over. Type "make this shorter and cut the corporate language" rather than closing the chat and beginning again. This single habit shift is where most people speed up dramatically, because AI tools remember context within a conversation and get better with each correction.
- Day 11 to 14: save your five best prompts somewhere, a note, a doc, whatever. These become templates. I still reuse a client-onboarding prompt I wrote eighteen months ago, tweaked maybe four times since.
Fourteen days, using one tool, on real work, with no course fee. That's the whole method. If you want ready-made starting points rather than building from nothing, I've put together a set of AI prompts for marketers that are built from actual campaigns rather than hypothetical examples, which saves you the eleven-attempt afternoon I described above.
The frameworks are fine, just don't worship them
You'll see acronyms floating around: RACE (Role, Action, Context, Execute), CO-STAR (Context, Objective, Style, Tone, Audience, Response), and a dozen others. They're all reasonable memory aids and they all boil down to the same four building blocks I listed above with different letters attached. Learn one if it helps you remember to include context and format. Don't buy a £200 course to learn an acronym you could write on a Post-it note.
What the frameworks don't teach you, and what separates someone who's good at this from someone who isn't, is domain knowledge. If you don't know your own customer well enough to describe them in a sentence, no amount of prompt structure fixes that. The AI can't invent your business context for you.
Common mistakes I see constantly
- Treating the first output as final. It's a first draft. Push back on it the way you'd push back on a junior writer's first attempt.
- Being vague about tone and then blaming the tool for sounding "AI-generated." If you don't specify a voice, it defaults to the blandest possible middle ground.
- Not giving examples. If you have a piece of writing you like, paste a paragraph of it into the prompt and say "match this style." This works better than any adjective you could choose.
- Starting a new chat every time instead of correcting within the same thread, which throws away useful context the tool has already built up.
- Assuming one prompt should do everything. Break big tasks into steps: research, then draft, then edit, rather than asking for a finished article in one shot.
When it's worth getting someone to teach you directly
Self-teaching over two weeks works well for individuals. It works less well when you're trying to get an entire team using AI consistently, because everyone develops their own sloppy habits and nobody corrects them. This is where I'd tell a client honestly: if you're one person learning for your own output, save your money and follow the steps above. If you're trying to move a team of eight from "nobody uses AI" to "everyone uses it the same sensible way," that's a different job, and it's faster with structured help than with everyone muddling through separately. I've written about what it looks like to hire someone to teach your business AI, including what to expect to pay and what a decent session covers, if that's the stage you're at.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
The terminology itself trips people up too. Half the confusion around "prompt engineering" comes from mixing it up with adjacent terms like fine-tuning, RAG, or agents, which are different things entirely. If jargon is slowing you down, the AI marketing glossary is worth ten minutes, purely so you stop nodding along in meetings when someone says "context window" and you're not sure what it means.
What comes after basic prompting
Once you're comfortable writing a decent prompt without thinking about it, which for most people takes those two weeks I mentioned plus maybe a month of regular use to feel automatic, the next useful skill isn't a fancier prompt. It's learning to chain tasks together, or set up something that runs without you typing a fresh prompt each time. That's the territory of building your first AI agent, which sounds far more technical than it is; it's really just prompting, but with steps linked together so the output of one becomes the input of the next.
And if you find you enjoy this enough that you're now the person your friends and colleagues ask for help, it's worth knowing there's a real path from "good at prompting" to paid work in this space. I get asked constantly whether prompt skills alone are enough to consult professionally. The honest answer sits in how to become an AI consultant, which is blunter about the actual requirements than most of what you'll find on this topic.
What I'd tell you if we were having a coffee about this
Stop looking for the perfect course. Pick ChatGPT or Claude today, open it, and write down one task you do this week that's boring and repetitive. Give it a role, tell it the task, give it context, tell it the format. Read the output critically, the way you'd read a first draft from a new hire. Correct it in the same chat. Do that on the same task five more times this week. You'll be noticeably better by Friday than you are right now, and you won't have spent a penny getting there.
Frequently asked questions
How long does it take to learn prompt engineering basics?
Most people get comfortable with the fundamentals in about two weeks of daily use on real tasks, roughly fifteen to twenty minutes a day. Fluency, where prompting feels automatic, tends to arrive after a month of regular use rather than occasional dabbling.
Do I need to pay for a course to learn prompt engineering?
No, not for the basics. The core skill (role, task, context, format) can be learned free by using any major AI tool on your own work daily. Paid help becomes worth it when you're training an entire team consistently or want faster progress with someone checking your work.
What is the biggest mistake beginners make with prompts?
Accepting the first response as final and blaming the tool when it sounds generic, instead of specifying tone, giving an example to match, and correcting the output within the same conversation.
Is prompt engineering still a useful skill in 2026?
Yes, but it looks different than in 2023. The formulas matter less because models understand plainer instructions now. What matters more is knowing your business well enough to explain it clearly in writing, which is really a communication skill wearing an AI costume.
Useful references
Related reading: How Many Prompts Can You Send ChatGPT in a Day? (The Real Numbers for 2026) and What Is an AI Agent, and How Do You Build a Simple One?.
Want the complete version? Read where I break down AI marketing.
