Jump to a section
The short answer

Three habits lift your prompts from good to great. Show it a few examples of what you want, because examples steer format and tone reliably. Separate your instructions from the material you paste in, using labels or tags. And explain why a rule matters, so it can apply it sensibly. Anthropic's own guide puts it simply: treat the AI like a brilliant but new colleague who doesn't know your context.

Think 'brilliant new colleague'

Anthropic, the company behind Claude, tells developers to 'think of Claude as a brilliant but new employee who lacks context on your norms and workflows. The more precisely you explain what you want, the better the result.' Its test: show your prompt to a colleague with little context, and 'if they'd be confused, Claude will be too.' That works for any chatbot.

How to ask an AI chatbot a good question (the four-part trick)

1. Show, don't just tell

Describing a style is hard. Showing it is easy. Anthropic calls examples 'one of the most reliable ways' to steer format, tone and structure, and suggests three to five, varied enough that it doesn't copy one too closely.

This is perfect for anything you want to sound like you: social posts, product listings, the family newsletter.

Learn my style from examples
I want you to write [WHAT YOU NEED] in my style. Here are three examples I've written before:

Example 1: [PASTE]
Example 2: [PASTE]
Example 3: [PASTE]

Match the length, tone and structure of these, but not the content. Now write one about: [TOPIC].

2. Label the parts of your prompt

When you paste in a long email, a document or notes, the chatbot can confuse your material with your instructions. Anthropic recommends wrapping each kind of content in its own tag, such as <instructions> and <document>, which it says 'reduces misinterpretation'. Plain headings in capitals do a similar job in any chatbot.

Instructions kept separate
<instructions>
Summarise the letter below in three bullet points, then list any deadlines and what I need to do by each one.
</instructions>

<letter>
[PASTE THE LETTER, WITH PERSONAL DETAILS REMOVED]
</letter>

3. Explain why, not just what

Anthropic says giving the reason behind an instruction helps the model 'deliver more targeted responses', because it can generalise from the explanation. Its example: rather than 'never use ellipses', say the text will be read aloud by a text-to-speech engine that can't pronounce them.

So instead of 'keep it short', try 'keep it short: my mum reads these on her phone and gives up on long messages'. You'll get something that fits the real situation, not just the rule.

Bonus: number the steps

When order matters, Anthropic suggests giving instructions as numbered steps. It's handy for anything with stages, like 'read this, then compare, then recommend'.

A step-by-step prompt
Please do these in order:
1. Read the three quotes below and list what each includes.
2. Point out anything one includes that the others don't.
3. Tell me which is the best value and why, in two sentences.
4. Suggest two questions I should ask before choosing.

[PASTE THE QUOTES]

Save the prompts that work

Once a prompt works, don't retype it. Put it in a Project or Gem's instructions, or keep a note on your phone of your best ones.

ChatGPT Projects and Gemini Gems: how to set them up

40 useful AI prompts for everyday life

How to get honest feedback from AI (and stop it just agreeing with you)

Sources

These sources support the guide. Tool settings, prices and policies change, so dated source boxes show when those details were checked.

AI helps with research and drafting. Dan checks the sources and is responsible for what is published.

How we check