Skip to content

Prompting: how to talk to an AI without offending it

Five simple habits to get useful answers on the first try.

A “prompt” is simply the instruction you give an AI. Good news: it doesn’t get offended. Bad news: if your instruction is vague, its answer will be too. Here are five habits to make yourself understood.

A real case: chasing a customer

Say you run a small business and want to write a reminder for an unpaid invoice. Compare:

  • “Write a reminder.”: the AI guesses everything, and often guesses wrong.
  • “You are an administrative assistant. Write a polite but firm reminder email, in three sentences, for an invoice past its due date. Tone: friendly, formal address.”: now we’re getting somewhere.

The five habits

  1. Give a role and a context: who the AI is, who it writes for, in what situation.
  2. Say exactly what you expect: format, length, tone, language.
  3. Give an example: a sample sentence or table beats ten explanations.
  4. Break complex tasks down: one step at a time, rather than a novel of instructions.
  5. Read and correct: answer the AI like a colleague (“shorter”, “less blunt”). The first draft is only a draft.

The limits

  • A good prompt doesn’t guarantee a correct answer: always proofread what goes out to a customer.
  • Don’t paste sensitive data (contracts, customer data) into a tool when you don’t know where the information goes.

The takeaway: the more your instruction looks like the brief you’d give a colleague, the better the answer.

An AI agent project?

We design it with you, seriously. Well, almost.

Talk to the team
Octopote Avatar

Written by