Lessons0 of 20 done

Module 3: Using AI well. Lesson 9 of 20, 7 minutes.

The anatomy of a good prompt.

The six parts of a prompt that change the answer, the two most people leave out, and why saying your reason helps.

Most prompts people type look like this: write something about our new product. The reply is fine. Generic, a bit long, in a voice that belongs to no one. The person reads it, sighs, and decides the tool is overrated.

The problem is not the model. The prompt left out nearly everything the model needed in order to make choices. A good prompt is not long or clever. It just has its parts. This lesson names them.

The parts of a prompt

A prompt (the text you send to the model) does its work through a handful of components. Not every prompt needs all of them, but when a reply disappoints, one of these is usually missing:

  • The task: what to do, as a specific verb.
  • The material: the text, data or facts it should work from.
  • The audience: who the result is for.
  • The format: the shape the answer should take.
  • The limits: length, tone and what to leave out.
  • An example: a sample of what good looks like, when you have one.

Think of it as a brief you would give a capable new colleague who knows nothing about your situation. You would not say "write something about the product." You would say what, for whom, how long, in what tone, and you would hand over the notes.

Build a prompt from its parts

Fill in what you have. The prompt assembles itself on the right, and the line underneath says which parts are still missing.

Your prompt

Choose a task to begin.

Start with a specific verb

"Write something about" is not a task. Summarize, compare, rewrite, list, draft, explain, critique, translate, extract, rank: these are tasks. Each tells the model a different kind of answer is wanted. Summarize this article and Critique this article produce completely different replies from the same material, and "tell me about this article" lands somewhere vague between them.

The material is whatever the task operates on: the article, the email thread, the figures, the facts about your product. If the model does not have it, it fills the gap with generic text, because generic text is all it can produce about a product it has never heard of. The next lesson is entirely about handing over material well, so here the rule is just: include it.

The two parts people leave out

Ask people what their prompt was missing and they rarely guess these two, yet between them they account for most disappointing replies.

Who it is for. The same facts should be written differently for a customer, a board, a new employee or a ten-year-old. Without an audience, the model picks a default: a polite, mid-length, mildly corporate voice aimed at nobody. One clause fixes it: for existing customers who already know the product, or for my manager, who has two minutes.

What shape the answer should take. A paragraph, a table, five bullets, a subject line plus body, a numbered checklist, a single sentence. If you do not say, the model guesses, and its guess is usually longer and more padded than you wanted. Saying reply as a table with three columns: option, cost, risk or three sentences, no bullet points saves a round of editing.

Limits and an example

Limits are where you head off the predictable problems. Length: under 120 words. Tone: plain and friendly, no marketing language. Exclusions: do not mention pricing. Models lean toward more: more words, more enthusiasm, more caveats. A limit or two pulls them back.

An example beats a description. If you have a past email that struck the right tone, paste it and say match this style. If you want a particular layout, show one line of it. The model is better at copying a pattern than at following a paragraph that describes the pattern. You will see this idea again in the next lesson.

Say why

One more piece costs a sentence and pays back well: the reason. I am writing this to calm a customer who is annoyed about a delay tells the model something no list of instructions quite captures. Every reply involves dozens of small decisions you did not spell out: whether to apologize first or explain first, whether to use the customer's name, whether a light touch fits. Given the purpose, the model makes those decisions in your direction. Without it, it makes them by default, and the defaults belong to nobody in particular.

Before and after

Before

Write something about our new scheduling feature.

After

Draft a short announcement email about our new scheduling feature for existing customers who already use the product every day. The goal is to get them to try it this week. Three short paragraphs, under 150 words, plain and warm, no sales language. Do not mention pricing. The feature: pick a time slot from inside the chat window, invite others, see everyone's availability. Here is a past announcement whose tone I like: [pasted email].

The second prompt is longer, but every added clause is one fewer guess for the model and one fewer edit for you. Task, material, audience, format, limits, example and the reason, all in five sentences.

With practice you start to feel the missing part before you press enter. Until then a checklist helps, and tools exist for this: Alide, the extension this course comes from, points out the gaps in a prompt as you type it. Either way the habit is the same. Before sending, ask who this is for and what shape the answer should take.

Try it

Take the last prompt you sent to a chatbot, or use write something about our new product. Send it as it is and keep the reply. Now rewrite it with all six parts: a specific verb, the material (a few real facts), the audience, the format, two limits, and a sentence saying why you want it. Send that in a fresh chat. Put the two replies side by side. The difference is the lesson.

Key terms

Prompt
The text you send to a model, including the instructions, any material and any examples.
Task
The specific action you want, stated as a verb such as summarize, compare or draft.
Audience
The person or group the answer is for, which sets its vocabulary, tone and depth.
Format
The shape of the answer: a table, a list, a paragraph, an email, a set number of sentences.
Limits
Constraints on the answer, such as length, tone and things to leave out.

Check yourself

1.Which of these is a complete task in the sense this lesson uses?
2.Which two parts do people most often leave out of a prompt?
3.Why does telling the model why you want something help?

A good prompt is not clever, it is complete: a specific verb, the material, who it is for, the shape you want, the limits, and when you can, an example and a reason.