Effective prompting techniques

Q: How do you write an effective AI prompt?

Write an effective AI prompt by treating it like briefing a capable new colleague: say clearly what you want, why you want it, and who you are, then check what comes back. The techniques that do most of the work are giving context and constraints, showing an example of what good looks like, and, for anything with reasoning, asking it to think the problem through before it answers. Pick two or three, use them until they are automatic, and refine when the first attempt is not quite right.

Q: What makes a good AI prompt?

A good AI prompt is specific and clear about what success looks like, rather than leaving the model to guess. The moves that matter most are these. Give context and constraints by saying what it is for and setting boundaries such as length, format, and what to leave out. Show one good example, because the model copies structure and tone well. Ask it to work step by step for anything involving reasoning. Plan first by asking for its approach or an outline, checking that, then asking for the final version. Set a role and tone, like a cautious compliance reviewer, which changes the output more than people expect.

Q: How can beginners write better AI prompts?

Beginners can write better AI prompts by adding the two changes that make the biggest difference: paste one example of what good looks like, and ask it to work through the task step by step. Take a prompt you use often, rewrite it with those two added, and run the old and new versions side by side. The difference is usually obvious. Prompting is iterative, so when a response is not quite right, refine your approach rather than expecting the first attempt to be perfect.

Q: What should you do when an AI response is not what you wanted?

When an AI response is not what you wanted, treat the first answer as a draft and refine, rather than assuming the prompt failed. Add more context or specificity, give an example of the output you want, break the task into smaller steps, or ask for a few different versions to choose from. If the conversation has drifted off track, starting a fresh one often works better than trying to correct it. When you are stuck on how to ask, describe your goal to the AI and ask it to write or improve the prompt for you.

This is the practical companion to Description: the specific techniques that make a brief land. None of them are complicated, and a handful will cover most of what your team does day to day.

Video: Anthropic’s AI Fluency: Framework & Foundations · CC BY-NC-SA 4.0 · watch on YouTube

The techniques worth knowing

The video walks through the full set. These are the ones I would make sure your team actually uses:

Give context and constraints. Say what it is for and set the boundaries: length, format, what to leave out.
Show an example. Paste one good sample of what you want. The model copies structure and tone well.
Ask it to work step by step. For anything with reasoning, tell it to think it through before answering. The output gets noticeably better.
Plan first, then write. Ask for its approach or an outline, check that, then ask for the final version.
Set the role and tone. 'You are a cautious compliance reviewer' changes what you get back more than people expect.

Pick two or three of these and use them until they are automatic. That is most of practical prompting.

Try this

Take one prompt your team uses a lot and rewrite it with two of these added: an example and a step-by-step instruction. Run the old and new versions side by side. The difference is usually obvious, and it makes the case better than I can.

Common questions about AI prompting

How do you write an effective AI prompt?

Treat it like briefing a capable new colleague: say clearly what you want, why you want it, and who you are, then check what comes back. The techniques that do most of the work are giving context and constraints, showing an example of what good looks like, and, for anything with reasoning, asking it to think the problem through before it answers. Pick two or three, use them until they are automatic, and refine when the first attempt is not quite right.

What makes a good AI prompt?

A good prompt is specific and clear about what success looks like, rather than leaving the model to guess. The moves that matter most are:

Context and constraints: say what it is for and set the boundaries, such as length, format, and what to leave out.
An example: paste one good sample of what you want, because the model copies structure and tone well.
Step by step: for anything involving reasoning, ask it to work through the problem before answering.
Plan first: ask for its approach or an outline, check that, then ask for the final version.
Role and tone: telling it who to act as, like a cautious compliance reviewer, changes the output more than people expect.

How can beginners write better AI prompts?

Start by adding the two changes that make the biggest difference: paste one example of what good looks like, and ask it to work through the task step by step. Take a prompt you use often, rewrite it with those two added, and run the old and new versions side by side. The difference is usually obvious. Prompting is iterative, so when a response is not quite right, refine your approach rather than expecting the first attempt to be perfect.

What should you do when an AI response is not what you wanted?

Treat the first answer as a draft and refine, rather than assuming the prompt failed. Add more context or specificity, give an example of the output you want, break the task into smaller steps, or ask for a few different versions to choose from. If the conversation has drifted off track, starting a fresh one often works better than trying to correct it.

Note: when you are stuck on how to ask, describe your goal to the AI and ask it to write or improve the prompt for you.