The 5-step framework behind prompts that actually work.

Most people use AI the way they use a search bar: type a wish, hope for magic. Then they judge the tool by the first mediocre answer it returns. The tool was never the problem. The prompt was — because a prompt isn't a wish, it's a specification. And specifications can be engineered.

Every prompt we ship — in client work and in the playbooks we publish — goes through the same five steps. None of them are complicated. Together, they're the difference between output you delete and output you use.

01Persona

Tell the model who it is before you tell it what to do. "Write a listing description" gets you average internet copy, because that's the average of what the model has read. "You are a luxury real-estate copywriter who has sold $50M homes with words alone" pulls from a far sharper distribution. The persona is a lens — it focuses everything that follows.

02Context & references

The model can't read your mind, your market, or your brand voice — unless you hand them over. Who is the audience? What's the situation? What does good look like? Paste an example of the tone you want. One honest reference beats three paragraphs of adjectives. This step is where most prompts quietly die: people give the task and withhold the world the task lives in.

03Chain of thought

For anything beyond a one-liner, make the model think before it writes: "First analyze the audience's top three objections. Then outline. Then draft." Forcing intermediate steps does for AI what showing your work does for a student — it catches the lazy leap before it becomes the final answer. The quality jump from this one instruction is consistently the largest of the five.

04Configuration

Not every task wants the same dial settings. Factual work — policy summaries, compliance language, data extraction — wants low temperature: predictable, careful, boring on purpose. Creative work — hooks, headlines, campaign angles — benefits from turning it up. If you only ever use defaults, you're driving one gear. Know when you want precision and when you want surprise.

05Evaluate & refine

The first output is a draft, not a verdict. Tell the model what's wrong with its answer and make it fix it: too generic, wrong tone, weak opening — say so, specifically. We call the habit ABI: Always Be Iterating. The professionals getting remarkable results from AI aren't writing magic first prompts. They're running tight, honest feedback loops — usually two or three rounds — while everyone else gives up after one.

That's the whole framework. No secret words, no "jailbreaks," no 3,000-prompt swipe files. Five deliberate steps, applied every time — engineered, not guessed.