ChatGPT and Claude can write emails, plan campaigns, draft proposals, and summarise documents in seconds. Yet most business owners who try them walk away disappointed, because the output is generic, wordy, or just not quite right. The problem is almost never the AI. It is the prompt.
Prompt engineering is simply the skill of telling an AI exactly what you want. You do not need to be technical to learn it. In fact, the best prompters are people who are clear communicators, which is exactly what a business owner should be. Here is how to get consistently better results.
The Core Formula: Role, Task, Context, Constraints
Every great prompt has four ingredients. When your prompt feels too vague, one of these is usually missing. A strong prompt looks like this:
- Role — who the AI should act as. Example: 'You are an experienced marketing copywriter for Sri Lankan restaurants.'
- Task — exactly what you want. Example: 'Write a WhatsApp announcement for a weekend seafood special.'
- Context — background it needs. Example: 'We are a family restaurant in Mount Lavinia, price range LKR 800–1,500 per person, customers are mostly local families.'
- Constraints — the rules. Example: 'Keep it under 80 words, friendly but not pushy, include the booking number, no emojis.'
Here is a full example you can copy and adapt: 'You are a marketing copywriter. Write a Facebook post promoting our new lunch delivery service. We are a small bakery in Colombo, delivering within 5 km, orders before 11 a.m. delivered by 1 p.m. Keep it under 120 words, warm and casual, end with a call to action, do not use hashtags.'
Ask for One Thing at a Time
New users often ask for too much in a single prompt, such as 'write me a website, a logo concept, and an email sequence.' The AI will try to do all of it and deliver a shallow result. Break the work into steps. First generate ideas, then pick one, then refine it, then ask for variations. Each step with a focused prompt gives dramatically better results.
Use 'Then' and Iterate Like a Manager
Treat the AI like a capable but literal new employee. It will not read your mind, and it will not ask questions unless you ask it to. Instead of starting over when output is wrong, direct it with follow-ups: 'That is too formal, make it warmer. Now shorten it to 50 words. Now make it sound like a person from Colombo, not a textbook.'
You can also ask the AI to ask you questions before starting. Add this line to any important prompt: 'Before you begin, ask me up to three clarifying questions.' This single trick often produces the biggest quality jump.
Time-Saving Prompt Templates for Owners
- Email replies: 'Draft a polite, professional email replying to a customer complaint about a late delivery. Our delivery partner was at fault. Apologise, explain briefly, and offer a 10% discount on their next order.'
- Social captions: 'Write 5 Instagram captions for a photo of our new product. Each under 60 words, varied tone: one funny, one practical, one storytelling.'
- Meeting summaries: 'Summarise this conversation into action items with owner, deadline, and status.'
- Content ideas: 'List 20 blog post ideas for a small accounting firm serving Colombo startups, ranked by how likely they are to generate client enquiries.'
- Proposal drafting: 'Draft a proposal outline for a website redesign project worth LKR 500,000, including scope, timeline, and payment milestones.'
Give It Your Voice and Data
The best AI output comes from your own materials. Paste your business description, a few of your best emails, or your brand tone notes into the chat and say 'Match this tone.' If you have customer reviews, use them: 'Write a new testimonial-style Facebook post using language like these reviews.' The AI cannot know your voice until you show it.
Common Mistakes to Avoid
- Giving no context and expecting magic. The AI does not know your business exists until you tell it.
- Accepting the first draft. The first version is usually the worst; iterate at least three times.
- Asking for facts it might not have. For numbers, dates, and statistics, verify or paste your own data.
- Filling in every field with generic fluff. Specific, real details beat polished vague language every time.
Final Thoughts
Prompt engineering is not a technical skill reserved for programmers. It is a communication skill: say who the AI is, what you want, what it should know, and what the rules are. Then treat the output as a first draft and direct revisions like a good manager.
Spend twenty minutes practising the role, task, context, constraints formula and the difference will be visible in the very next email, post, or proposal you produce. Your competition is still typing 'write something about our business.' You can do better.