Most Australian business owners try ChatGPT once, get a generic wishy-washy response, and decide AI is overhyped. The problem is almost never the AI. It's the brief. A good prompt is less about clever wording and more about giving the tool the same context you'd give a new staff member on their first day — who they are, what they're working on, what you want at the end, and what to avoid. Get those four things right and the output stops feeling like a novelty and starts feeling like a team member.
Stop asking questions, start giving briefs
The single biggest mistake in AI prompting is treating it like Google. You type a question, expect an answer, and judge the tool by how good that answer is. AI tools reward specificity the way a contractor rewards a clear scope. If you wouldn't accept "write me a quote for a kitchen" from a tradie without any measurements, don't expect a useful email draft from a one-line prompt.
Instead, treat every prompt like a mini brief. Who is this for? What do you want at the end? What tone or format? What should it avoid? A prompt with those four elements will outperform a clever-sounding one every time, even if yours is three times longer and written in plain English with typos.
The four-part structure that works for anything
After helping Australian SMEs set up their first AI workflows, we've landed on a simple pattern that works whether you're writing a follow-up email, a Facebook ad, a job ad, or an internal memo. Role. Context. Task. Format. Four words, in that order, and the quality of the output changes completely.
Role tells the AI who it is playing. "You are an experienced Brisbane plumber writing quotes for small business clients" lands better than no role at all. Context gives it the background. "The client runs a cafe in Bulimba. The job is an emergency blocked drain on a Sunday." Task is the actual ask: "Write a three-paragraph email explaining the call-out fee, what work is likely required, and when we can be on site." Format specifies the output: "Plain email, no subject line, Australian English, under 150 words, warm but professional."
Those four lines transform the result from something you'd throw away into something you might actually send. Try it today on a task you'd normally outsource — a customer follow-up, a supplier email, a social post. Time yourself: the prompt takes two minutes, the output takes ten seconds, and the editing takes another two. That's a job done in under five minutes that probably took you twenty before.
Give examples of good and bad
AI models learn very fast from examples. If you have two or three previous emails or social posts you wrote that really worked, paste them into the prompt as reference material. Say "match the voice and structure of these examples, but not the content." You'll get something that sounds like you, not like a chatbot that read a marketing textbook.
The same trick works in reverse. If you've seen output you hated — too salesy, too American, too much jargon — include it and say "avoid this style." This is one of the cheapest quality wins available. It takes 60 seconds and changes the output meaningfully. Most business owners skip it because they're still thinking of AI as a magic answer machine rather than a tireless junior staffer who copies what you show them.
Be specific about what NOT to do
This is the trick nobody mentions. AI tools have strong default habits. They love the word "leverage," they default to American spelling, they sprinkle em-dashes and exclamation marks through everything, and they end every email with "please let me know if you have any questions!" If you don't explicitly tell them to stop, they will keep doing it, no matter how good the rest of your prompt is.
A simple guardrail line at the end of every prompt solves most of this. Something like: "Use Australian English. No em-dashes. Do not use the words leverage, seamless, or cutting-edge. No exclamation marks. Do not end with 'let me know if you have any questions.'" Paste that into every prompt you write for the next week and notice how much less editing you're doing. You'll also notice how quickly you start trusting the output, which is the bigger prize.
Save your best prompts, don't rewrite them
The operators getting the most out of AI tools aren't the ones writing perfect prompts on the fly. They're the ones who've built a small library of prompts that work — a quote reply template, a social post generator, a follow-up email writer, a meeting summariser — and reuse them with light tweaks each time.
Start a simple document today. Call it "Prompts that work." Every time you get output you actually use, paste the prompt that produced it. Within a month you'll have a personal playbook that's worth more than any AI course you could buy. It's also the raw material for automating these workflows properly, because once a prompt is reliable, it can be plugged into a tool like Zapier, Make, or a custom GPT and run without you lifting a finger.
That's the gap between businesses who dabble with AI and businesses who build a real edge from it. If you'd like help turning the prompts that already work for you into proper workflows — the kind that run while you're at lunch — we do exactly that at ProjxAI. Start with our AI Readiness Audit and we'll map out where the fastest wins are hiding in your business.