Writing a good prompt: the guide for ChatGPT, Claude and Gemini

The quality of an AI answer depends above all on the quality of the request. A vague prompt gives a vague result. A precise prompt gives something you can use. The good news: writing well comes down to a few principles, valid for ChatGPT, Claude or Gemini alike.

A prompt is an instruction, not a keyword

You do not use a generative AI like a search engine. You talk to it like a capable colleague who does not know your context. The more you give, the better the answer.

The four ingredients of a good prompt

• The role : “You are a sales assistant writing to customers in Benin.”
• The context : who, what, for whom, under which constraints.
• The task : precisely what you want, as one clear objective.
• The format : length, tone, expected structure (list, email, table).

“Write an email” gives you a platitude. “Write a short, courteous email chasing a customer who has not paid a 150,000 FCFA invoice for 30 days, firm but respectful in tone” gives you a usable text.

Give examples

Showing an example of what you expect strongly steers the result. A successful email, an extract in the right tone: the AI draws on it to produce in the same spirit.

Iterate rather than expect everything at once

The first draft is rarely perfect. You correct through dialogue: “shorter”, “add a closing line”, “make it more direct”. It is a back and forth, not an oracle. The official documentation from OpenAI and Anthropic both insist on this iterative approach.

What not to put in

Beware of sensitive data. You do not paste confidential information (named customer data, trade secrets) into a consumer tool without a clear framework. We cover this in AI and data protection.

Building the skill with Urban Academy

Prompting well is a craft, and one a team works on together. Our generative AI and prompting course is built for that. Discover it on the Academypage, and see also 10 AI use cases for an SME.