How To Prompt Claude Like A Pro: 9 Techniques With Examples
12 min read
Short answer: the best Claude prompts read like a clear brief to a smart new hire: say exactly what you want and why, show an example of good work, separate your materials with labels, let Claude think on hard problems, and ask it to check its work against your sources. Anthropic's own prompt engineering guide is built on the same ideas. Below are nine techniques, each with a before and after from my shop.
I learned prompting the hard way with my bot team: hundreds of vague asks, hundreds of so-so answers. What finally clicked is that a prompt isn't a magic phrase. It's a job description. Everything below is just a way to write a better job description.
1. How Specific Should A Claude Prompt Be?
More specific than feels necessary. Claude can't see what's in your head: your customers, your tone, what you'll do with the answer. Anthropic's first piece of advice is to be clear and direct, like briefing a brilliant colleague who's new to the job (source).
"Write a Facebook post about screen repair."
"Write a Facebook post for people in North Park with a cracked iPhone screen. Mention prices start at $78.95, most repairs take about 45 minutes, and walk-ins are by call or text first. 60 to 90 words, no hashtags, end with the phone number."
2. Why Should You Tell Claude The Reason?
Because a rule with a reason gets applied sensibly in situations you didn't predict. "Never use exclamation points" is followed literally. "Keep it calm, because customers with broken phones are already stressed" is followed in spirit, including in ways you didn't think to list.
"Don't use the word OEM."
"Don't say OEM, genuine or original, because we sell high-quality aftermarket parts and those words would mislead customers. Describe parts by their tier and brand instead."
3. Do Examples Really Help Claude?
Yes, more than almost anything else. One or two examples of the output you want teach tone, length and format faster than a page of rules (source). Show variety, though: if all your examples are about iPhones, Claude will lean iPhone even when the customer has a Pixel.
Here are two replies I liked:
Example 1: "Hi Maria, an iPhone 13 screen is $142.95 in our Standard tier and takes about 45 minutes. Want me to hold a time today? Call or text first so I'm home."
Example 2: "Hey Jon, Galaxy S24 screens are by quote since parts arrive in 1-2 business days. Text me a photo and I'll send exact pricing."
Customer text: "how much for pixel 8 back glass"
4. Why Label Your Material With Tags?
When a prompt mixes instructions, documents and examples, Claude can confuse one for another. Wrapping each piece in simple labels such as <price_list> … </price_list> keeps them apart. Anthropic recommends these XML-style tags for exactly this reason (source). The tag names can be anything that describes the contents.
<our_policy> ...paste warranty policy... </our_policy>
Using only <our_policy>, draft a reply to <customer_email>. If the policy doesn't cover their question, say so and suggest they call.
Bonus: you can ask Claude to answer in tags too ("put the reply in <reply> tags and your reasoning in <notes>"), which makes it easy to copy just the part you need.
5. Should You Give Claude A Role?
Yes, for anything specialized. "You are an experienced bookkeeper for a one-person repair shop" changes what Claude pays attention to: it starts flagging missing receipts instead of just adding columns. Put the role in your Project instructions so you don't repeat it.
6. When Should You Let Claude Think First?
On anything with numbers, logic, several steps or trade-offs. Anthropic's guidance is to give Claude room to reason before answering on complex tasks (source). In the app, that means turning extended thinking on or raising the effort level. In a plain prompt, ask for it: "Think through the options first, then give me your recommendation."
The trade-off: thinking uses more of your allowance. Turn it off for routine rewrites and replies (source).
7. How Do You Prompt Claude With Long Documents?
Two habits from Anthropic's long-document advice (source):
- Put the documents first and your question last. Claude answers better when the ask comes after the material.
- Ask it to quote before it concludes. "First pull the exact lines from the contract about cancellation, then answer my question using only those lines." Quoting first keeps the answer grounded in what the document actually says.
8. How Do You Handle Big Jobs?
Break them into steps and do one step per message, checking each before moving on. Anthropic calls this chaining prompts (source). When Claude rewrote my warranty wording across the site, it went: find every page that mentions it, show me the list, propose the new wording, wait for my answer, change the short descriptions, verify, then the longer pages. Each step could be checked. One giant "fix my warranty everywhere" prompt couldn't.
Even easier: ask Claude to make the plan. "Before you start, give me a numbered plan and wait for my OK."
9. How Do You Make Claude Check Its Own Work?
Ask for the check in the same prompt. Some of my standing check lines:
- "After the table, tell me which file and line each price came from."
- "Before you finish, reread the banned words list and confirm none are in your draft."
- "If you're not sure about a fact, say so instead of guessing."
- "Count the items you changed and compare with the count you found at the start."
That last one is how I knew my store fix was complete: Claude found 154 items, changed 154, and showed both numbers.
Is There A Claude Prompt Template You Can Copy?
Goal: [what done looks like] because [why it matters].
Context: <materials> [paste or name the files] </materials>
Rules: Always [x]. Never [y], because [reason].
Example of good output: <example> [one example] </example>
Format: [length, structure, where to put it].
Check: Before finishing, [how to verify]. If anything's unclear, ask me first.
You won't need every line every time. For a quick rewrite, Goal and Format are enough. For anything going to customers or touching money, use the whole thing.
Your Turn
- ☐ Take a prompt you used this week and rewrite it with the template.
- ☐ Save your two best Claude replies as examples in a Project.
- ☐ Add one check line to your most important routine.
Frequently Asked Questions
What is the best way to prompt Claude?
Write it like a brief to a smart new hire: the goal, why it matters, your materials, an example of good work, the format you want and how to check the result.
Does Claude understand XML tags?
Yes. Anthropic recommends wrapping documents, examples and instructions in simple tags like <document> so Claude keeps them apart. Any descriptive tag name works.
Should I put my question before or after a long document?
After. Put the documents first and your question at the end.
How do I stop Claude from making things up?
Give it the source material, tell it to answer only from that material, ask it to quote the lines it used, and tell it to say "I'm not sure" instead of guessing.
Is prompting different for Claude than for ChatGPT or Grok?
The basics are the same. Claude responds especially well to clear structure, reasons for rules and labeled material.
More In This Series
- How To Set Up Claude As Your AI Assistant
- Claude Plans, Models And Usage Limits Explained
- How Claude Remembers: Projects, Memory, Skills And Instructions
- Let Claude Work In Your Apps: Connectors, Chrome And Scheduled Tasks
- How To Prompt Claude Like A Pro (you're here)
Also see How To Write AI Prompts For Beginners and the Grok Bot series.
Unofficial guide written by a Claude customer. Not affiliated with or endorsed by Anthropic. Checked against Anthropic's documentation on November 17, 2026.



