Prompt

How a prompt works

An AI tool like ChatGPT, Claude or Gemini starts every task knowing nothing about your business. The prompt is its only briefing. Think of it like a brief to a freelance copywriter you have never met. "Write something about our sale" gets you something generic. A brief that names the audience, the product, the tone and the length gets you something you can use.

A good prompt usually covers four things:

• Task: what to produce, for example three Instagram captions.

• Context: who the customer is and what matters to them.

• Examples: one or two texts you liked before.

• Format: length, language, tone and structure.

In practice for a small business

A family bakery in Athens wants posts for a new sourdough loaf. A weak prompt: "Write a post about our new bread." A stronger prompt: "Write three Instagram captions in Greek for our new sourdough loaf, made with Greek flour and a 24-hour rise. Our customers are local families. Tone: warm and simple, no exaggeration. Maximum 40 words each, with one call to visit the shop. Here is a caption we liked: …"

The second version takes one extra minute to write and usually saves several rounds of editing. Once a prompt works well, save it in a shared document so the whole team can reuse it. The same approach works for AI product descriptions.

Why it matters and what to watch out for

• Personal data: what you put in a prompt is sent to the provider. Do not paste customer names, phone numbers or employee records into tools without business-grade data terms. GDPR still applies.

• No guarantee: even a perfect prompt can produce a wrong fact. Always check numbers, prices and claims.

• Same prompt, different answer: AI tools vary their output. Test a prompt on several real cases before you rely on it.

• Confidential information: treat a prompt like an email to an outside supplier. If you would not email it, do not paste it.

Writing prompts systematically, with tests and templates, is called prompt engineering.