Prompting Claude Basics

10 examples to get you started with prompting Claude - 7 basic and 3 intermediate.

Every one of these examples uses the same three-part shape: state a clear goal, add the context Claude needs to hit it, and read the response critically before you trust it.

That shape works whether you are asking a quick question in Claude.ai or running a longer task in Claude Code, and whether you are on Claude Haiku 4.5 for a fast draft or Claude Opus 4.8 for something that needs careful reasoning.

Basic Examples

1. State a single, clear goal

The simplest fix for a weak prompt is naming exactly one outcome instead of a vague topic.

Weak:  "Marketing ideas."
Better: "Give me 5 subject-line ideas for an email announcing
a 20% off weekend sale to existing customers."

• A vague topic forces Claude to guess what "help" should look like.
• Naming the outcome (5 ideas, subject lines, a specific sale) removes that guesswork.
• One clear goal per prompt is easier to evaluate than several bundled together.

2. Add the context Claude can't see

Claude only knows what is in the prompt and the conversation, so background you're carrying in your head has to be written down.

"I run a 3-person landscaping business in Ohio. I need a short
reply to a customer asking if we work in the winter. We do
snow removal but not lawn care from November to March."

• The business type, size, and season all shape what a good answer looks like.
• Without this context, Claude would have to answer generically and probably wrong.
• Context is most useful when it changes the answer, not when it's just extra detail.

3. Say what format you want back

Claude defaults to a reasonable format when you don't specify one, which is often not the format you actually need.

"List the 4 steps as a numbered list, one short sentence each,
no extra commentary before or after."

• "Numbered list" and "one sentence each" both remove a decision Claude would otherwise make for you.
• "No extra commentary" heads off a preamble you'd have to skim past every time.
• Format instructions are cheap to write and save real editing time later.

4. Set a length constraint

An open-ended prompt often gets an open-ended, longer-than-needed answer.

"Explain what a deductible is, in 2 sentences, for someone who
has never had insurance before."

• A number ("2 sentences") is unambiguous in a way that "briefly" is not.
• Pairing the length with an audience ("never had insurance") shapes vocabulary too.
• Tight constraints are especially useful for text you'll paste directly into a UI or message.

5. Give Claude a role

Assigning a role narrows the tone, vocabulary, and depth Claude reaches for before it writes a single word.

"You are a patient high-school chemistry tutor. Explain why ice
floats on water to a student who is struggling with density."

• The role ("patient tutor") sets a tone that a bare question wouldn't.
• The audience detail ("struggling with density") keeps the answer at the right level.
• A role is a shortcut for a paragraph of instructions about tone and depth.

6. Ask a follow-up instead of starting over

Claude keeps the conversation's context, so a follow-up can build on the first answer rather than repeating it.

"That's close, but make it shorter and drop the technical jargon
in the second paragraph."

• Claude still has the original answer and your original goal in view.
• A short, targeted follow-up is faster than rewriting the whole prompt from scratch.
• This only works within the same conversation; a new chat starts with no memory of it.

7. Read the response against your original goal

The last step of every prompt is checking the response, not just accepting it.

Checklist after Claude replies:
- Did it answer the actual question I asked?
- Is the format what I asked for?
- Is anything stated as fact that I should verify myself?

• Claude can misread an ambiguous prompt even when it sounds confident.
• Checking format and scope catches most mismatches in a few seconds.
• Treat the first response as a draft you can refine, not a final answer by default.

Intermediate Examples

8. Combine goal, context, and format in one prompt

Once each piece feels natural on its own, the real skill is stacking them into a single, well-organized request.

"I'm writing a Slack message to my team announcing a new
deploy freeze during the holidays.
 
Context: we had 2 production incidents last December caused by
rushed deploys.
 
Write it in 4-5 sentences, friendly but firm tone, ending with
the exact freeze dates: Dec 20 - Jan 2."

• Goal, context, and format each appear as their own clearly labeled piece.
• Labeling the pieces ("Context:") helps Claude weigh them correctly instead of guessing which sentence is background versus instruction.
• This is the same instinct that later leads into structuring a prompt with formal sections or tags for longer, more complex requests.

9. Anchor the style with a short example

Describing a tone in words is harder than just showing Claude one example of it.

"Write 3 more product taglines in this style:
 
'Coffee that doesn't apologize for being strong.'
 
Same length, same slightly cheeky tone."

• One good example often communicates tone more precisely than a paragraph describing it.
• Claude pattern-matches the rhythm and voice of the example, not just its topic.
• This is a lightweight version of few-shot prompting, useful even with just a single example.

10. Iterate across two or three turns instead of one perfect prompt

Trying to get everything right in a single first prompt often takes longer than refining across a couple of turns.

Turn 1: "Draft a 1-paragraph bio for my conference speaker page."
Turn 2: "Good start - make it more casual and mention I'm
         based in Austin."
Turn 3: "Perfect, just trim the last sentence."

• Each turn only asks for one adjustment, which keeps feedback easy to give and easy for Claude to apply correctly.
• Iterating is usually faster than trying to anticipate every requirement up front.
• This turn-by-turn habit is the basis of iterative refinement for more complex, higher-stakes prompts.

Related: How Claude Interprets Your Instructions - the mental model behind why these examples work | Why Specificity Beats Cleverness in Prompts - going deeper on examples 1-4 | Iterative Refinement - a full walkthrough of example 10

Stack versions: Written against the Claude model lineup current as of ~June 2026 - Claude Fable 5, Claude Opus 4.8, Claude Sonnet 5 (the default), and Claude Haiku 4.5. Model names, pricing, and product features move quickly - verify current specifics at platform.claude.com/docs before relying on them.