AI can only work with what you tell it. Give it the context you’d give a capable new colleague.
Two owners of similar guesthouses both ask an AI tool to help with their website. The first types: “Write a description of my guesthouse.” She gets a bland paragraph about “a cosy retreat with warm hospitality” that could describe any guesthouse anywhere.
The second writes several sentences. Her guesthouse is a restored 1920s farmhouse near Clarens with five rooms, mountain views and a hearty breakfast made from local produce. Her guests are mostly couples from Johannesburg looking for a quiet weekend away. She wants a warm, unpretentious tone, about 120 words, and no clichés like “hidden gem”. She pastes in two reviews guests have written and asks the AI to capture what they loved.
Her result is specific, appealing and sounds like her. Same tool, very different output. The difference came down to the quality of the brief.
AI can only work with what you tell it. Give it the context you’d give a capable new colleague.
What the AI companies recommend
The companies that build AI models publish guidance on how to get the best results from them, and their advice is strikingly consistent. Anthropic’s prompt engineering guide and OpenAI’s equivalent both emphasise a handful of principles.
Be clear and specific. Say exactly what you want, including the format, length and audience. Vague requests produce generic answers.
Give context. Explain who you are, what the task is for, who will read it and why it matters. AI tools don’t know your business, your customers or your standards unless you tell them.
Provide examples. Showing an example of what good looks like, such as a previous email you liked, a review that captures your voice or a sample format, is often more effective than describing it.
Say what to avoid. If you don’t want jargon, clichés, a particular tone or certain claims, say so.
Break big tasks into steps. Complex requests often work better as a sequence: first outline, then draft, then refine.
Iterate. Treat the first answer as a draft. Ask for changes, more detail, a different tone or alternatives.
A useful way to think about it, often suggested in these guides, is to imagine you’re briefing a very capable new colleague who knows nothing about your business. Whatever you’d need to tell them to do a good job, the AI needs too.
AI tools don’t know your business, your customers or your standards unless you tell them.
A simple prompt structure
You don’t need special techniques for most business tasks. A simple structure covers the essentials. Before asking AI for help, think through five things:
- Role and context: who you are and what your business does. “I run a five-room guesthouse near Clarens…”
- Task: exactly what you want. “Write a 120-word description for our website’s home page.”
- Audience: who it’s for. “Couples from Johannesburg looking for a quiet weekend away.”
- Style and constraints: tone, length, format and what to avoid. “Warm and unpretentious. No clichés like ‘hidden gem’.”
- Examples or material: anything that shows what you mean. “Here are two guest reviews that capture what people love.”
Then read the result critically and ask for improvements: “Make it more specific about the breakfast”, “Give me three alternative opening lines”, “Shorten it to 80 words”.
It’s also worth saving prompts that work well. A shared document of good prompts for common tasks, such as writing product descriptions, replying to reviews or summarising meetings, can save your team a great deal of time.
INFOGRAPHIC GOES HERE
Two prompts, two results. ” → generic result. Right, a structured prompt with five labelled parts: Context · Task · Audience · Style · Examples → specific, on-brand result. Caption: Based on published prompting guidance from Anthropic and OpenAI.
Context → Task → Audience → Style → Examples.
Good prompts start with good thinking
There’s a quiet benefit to writing better prompts: it forces clearer thinking. To brief an AI well, you have to know who your audience is, what you want them to feel or do, and what makes your business different. Those are exactly the questions explored throughout this series, from What Are Customers Really Hiring You to Do? to Say One Thing Well. If you struggle to write the brief, the problem may be the strategy, not the tool.
And remember that a good prompt improves the draft, not the facts. As explored in AI Sounds Sure of Itself, always check anything factual before using it.
A few questions can help:
- Do our prompts include the context a new colleague would need?
- Do we show examples of what good looks like?
- Do we treat the first answer as a draft and refine it?
- Have we saved prompts that work well for tasks we repeat?
Which brings us back to the two guesthouse owners. They used the same tool, but only one gave it something real to work with: the farmhouse, the mountain views, the couples from Johannesburg, the guests’ own words. Her description sounded like her guesthouse because she had told the AI what her guesthouse was really like.
If you struggle to write the brief, the problem may be the strategy, not the tool.
One Like Research & Further Reading
- Anthropic, Prompt engineering overview, Claude documentation.
- OpenAI, prompt engineering guide, OpenAI platform documentation (general reference).