Best Prompts for Claude AI: 15 Proven Examples

The best prompts for Claude AI share one trait: they tell Claude exactly what job to do, what to work from, and what the output should look like. That’s it. No magic words, no secret phrases. In this guide I’ve collected 15 prompts that follow that pattern, each one tested on real work: writing, admin, analysis, coding, and planning. Copy them as they are, or swap in your own details. Either way, you’ll get more useful output than a one-line request ever produces.

Key Takeaways

• The best prompts for Claude AI define three things: the task, the source material, and the output format.
• One real example inside a prompt beats a paragraph of description. Claude copies patterns better than it follows adjectives.
• Wrap pasted content in simple tags so Claude never confuses your material with your instructions.
• All 15 prompts here work on the free plan. Longer documents may need a paid tier for the bigger usage limits.
• Save the prompts that work for you. A small library of tested prompts beats retyping from memory every time.

I share prompts and automation builds like this in a short weekly email. Join the newsletter if that sounds useful.

What Makes the Best Prompts for Claude AI Work?

Every strong prompt in this list is built from the same five parts. Learn them once and you can write your own versions for any task.

1. A role. One line telling Claude who it is for this task. “You are an editor for a small-business blog” changes the output more than most people expect.
2. A clear task. Name the action, the audience, and the goal. “Write a 150-word update for existing customers” beats “write about our update.”
3. Separated material. Wrap anything you paste in simple tags, like <notes> and </notes>. Claude then knows what’s an instruction and what’s just material.
4. An example. Show one input and one ideal output. This is the highest-leverage habit in prompting.
5. A pinned format. Say exactly what the response should look like, including what to leave out.

None of this is my invention. It matches what Anthropic teaches in its own prompt engineering documentation: be explicit, provide reference material, use examples, and structure the prompt so nothing gets mixed up.

There’s a reason to take output structure seriously. Anthropic’s Economic Index report from June 2026 found that 93% of Claude conversations produce some kind of artifact, most often explanations, documents, and step-by-step guidance. In other words, people don’t chat with Claude for fun. They come to leave with a finished thing. A good prompt decides what that finished thing is before Claude starts typing.

One more note before the list. These 15 are curated, not exhaustive. If you want a bigger library after this, my 30 Claude prompt templates for solopreneurs collection goes wider. This page goes deeper on fewer prompts, with the reasoning behind each one.

Best Claude AI Prompts for Writing and Content

1. The blog post outliner

Use this before you draft anything long. It forces the structure conversation to happen first, when changes are cheap.

You are a content editor for a practical, hands-on blog aimed at small-business owners. Create an outline for a post titled “[TITLE]” targeting the search phrase “[KEYWORD]”. Include: a one-sentence direct answer to open the post, 5-7 H2 sections in a logical order, and 2-3 bullet points under each section noting what it must cover. Flag any section where I’d need first-hand experience or data to be credible. Do not write the post itself.

Why it works: the role sets the voice, the task names the audience, and the last line pins the format by saying what not to do. That final instruction stops Claude from racing ahead into a full draft.

2. The tone-match rewriter

The fix for stiff, robotic output is not a better adjective. It’s a real sample of your own writing.

Here is a paragraph I wrote that sounds like me: <sample>[PASTE A PARAGRAPH YOU LIKE]</sample> Now rewrite the following text to match that tone, sentence rhythm, and vocabulary. Keep all facts unchanged. Do not add new claims. <text>[PASTE THE TEXT TO REWRITE]</text>

Why it works: this is the example principle in its purest form. Claude imitates a demonstrated pattern far more reliably than it follows a description like “casual but professional.”

3. The newsletter drafter

You write a short weekly email for [AUDIENCE]. From the notes below, draft a newsletter of 250 words or fewer. Structure: one-line hook, one useful idea explained plainly, one practical takeaway the reader can apply today, one-line sign-off. No hype words, no exclamation marks. <notes>[PASTE YOUR ROUGH NOTES]</notes>

Why it works: the word cap and the four-part structure remove the two most common newsletter failures, which are rambling and burying the point.

4. The social repurposer

Below is a blog post. Extract 5 standalone insights from it and turn each into a short social post of 280 characters or fewer. Each post must make sense with zero context, contain no hashtags, and end with a plain statement rather than a question. <post>[PASTE THE ARTICLE]</post>

Why it works: “standalone” and “zero context” are the load-bearing words. Without them you get teaser fragments that only work next to the original article.

Best Claude Prompts for Business and Admin

This is where I get the most day-to-day value. Admin prompts are unglamorous, and they pay off every single week.

5. The polite-but-firm email

Write an email to [RECIPIENT] about [SITUATION]. My goal: [WHAT YOU WANT TO HAPPEN]. Tone: warm, direct, professional. Constraints: under 120 words, no apologizing more than once, no filler phrases like “I hope this finds you well,” and end with one specific ask and a date.

Why it works: naming the goal separately from the situation makes Claude write toward an outcome, not just a summary of events. The constraints kill the padding that makes drafts sound generic.

6. Meeting notes to action items

Here are my raw meeting notes: <notes>[PASTE NOTES]</notes> Produce three sections. 1) Decisions made, as single sentences. 2) Action items, as a table with columns for task, owner, and deadline. Mark anything unassigned as “OWNER?”. 3) Open questions nobody answered. Do not invent owners or dates that are not in the notes.

Why it works: the “do not invent” line matters. Without it, Claude will helpfully fill gaps with plausible guesses, and plausible guesses in a task list cause real problems.

7. The scope drafter

You are a freelance consultant writing a project scope. Based on the details below, draft a one-page scope with: objective, deliverables as a bullet list, what is explicitly out of scope, timeline, and payment terms. Write the out-of-scope section with the most care, since that is where projects go wrong. <details>[PROJECT DETAILS, BUDGET, DATES]</details>

Why it works: telling Claude where to invest its effort (“the most care”) shapes attention the way a good brief shapes a human writer’s attention.

8. The decision helper

I’m deciding between [OPTION A] and [OPTION B] for [CONTEXT]. First, list what would need to be true for each option to be the right call. Then give the strongest honest case against each. Then recommend one, and state the single factor that most drove the recommendation. If you’re missing information that would change the answer, ask for it instead of guessing.

Why it works: asking for reasoning before the recommendation produces a noticeably better answer than asking for the recommendation alone. The permission to ask questions stops Claude from papering over gaps.

Best Claude Prompts for Analysis and Research

9. The tagged summarizer

Summarize the document below for a busy reader. Output: a two-sentence overview, then 5 key points as bullets, then one section titled “What this means for [YOUR SITUATION]”. Base the summary only on the document. If the document doesn’t support a claim, leave it out. <document>[PASTE THE DOCUMENT]</document>

Why it works: the tags keep long pasted material from blending into your instructions, and the “only on the document” line keeps the summary grounded instead of drifting into general knowledge.

10. The data explainer

Below is a table of data pasted as text. Identify the 3 most decision-relevant findings. For each: state the finding in one plain sentence, show the numbers behind it, and note one caveat about what the data can’t tell us. Skip obvious observations. <data>[PASTE CSV OR TABLE]</data>

Why it works: “decision-relevant” and “skip obvious observations” push past the surface-level readout you’d otherwise get, like “sales varied across months.”

11. The devil’s advocate

Here is a plan I’m fairly confident about: <plan>[PASTE YOUR PLAN]</plan> Your job is to argue against it. Give the 3 strongest objections a smart, informed skeptic would raise. For each objection, rate how damaging it is out of 10 and say what evidence would settle it. Do not soften the criticism with compliments.

Why it works: Claude defaults to agreeable. Explicitly assigning the critic role, and banning the compliment sandwich, gets you the pushback a good business partner would give. I run drafts of my own plans through this one, and the objection I least want to read is usually the useful one.

Best Claude Prompts for Coding and Automation

You don’t need to be a developer for these two. They’re the prompts I recommend to non-technical business owners first. If you do write code, my separate roundup of Claude prompts for coding covers that ground in full.

12. The code explainer

Explain what the following code does to a smart person who has never programmed. Use a real-world analogy for the overall structure, then walk through what happens step by step when it runs. End with: one sentence on what could break, and one sentence on whether this looks safe to run. <code>[PASTE THE CODE]</code>

Why it works: forcing an analogy first makes the explanation land for a non-technical reader, and the “what could break” line turns a lecture into a practical review.

13. The automation spec writer

I want to automate this process: [DESCRIBE THE MANUAL PROCESS, STEP BY STEP]. Write a build spec for it: the trigger, each step as an input-process-output line, what data moves between steps, the failure cases and what should happen in each, and which single step is hardest to automate reliably. Format as a spec I could hand to a developer or build myself in a tool like n8n or Zapier.

Why it works: most automation projects fail at the edges, not the happy path. Making failure cases a required section means they get designed instead of discovered.

Best Claude Prompts for Learning and Strategy

14. The practitioner tutor

Teach me [TOPIC] the way a practitioner would, not a textbook. Structure: what it actually is in two sentences, the 20% of the topic that covers 80% of real use, the mistakes beginners make, and a small exercise I can do in under 15 minutes to test my understanding. Assume I’m smart but completely new to this.

Why it works: “the way a practitioner would” plus the 80/20 framing filters out the encyclopedia answer. The exercise at the end converts reading into an actual check on your understanding.

15. The plan stress-tester

Here are my goals for the next quarter: <goals>[PASTE YOUR GOALS]</goals> Evaluate them as a tough but fair advisor. For each goal: is it specific enough to know if I hit it, what’s the biggest dependency outside my control, and what would I have to drop to actually achieve it? Then tell me which single goal you would cut, and why. Be direct.

Why it works: the “what would you cut” question is the one most of us avoid asking ourselves. Delegating it to Claude, with permission to be direct, surfaces the trade-off you’re already half-aware of.

All 15 Prompts at a Glance

Prompt	Best for	The one detail that makes it work
1. Blog post outliner	Content planning	Bans writing the draft itself
2. Tone-match rewriter	Sounding like you	A real sample of your writing
3. Newsletter drafter	Weekly emails	Hard word cap and fixed structure
4. Social repurposer	Repurposing posts	“Standalone, zero context” rule
5. Polite-but-firm email	Awkward emails	Goal stated separately from situation
6. Notes to action items	Meetings	“Do not invent owners or dates”
7. Scope drafter	Client work	Out-of-scope section gets the care
8. Decision helper	Two-option choices	Reasoning before the recommendation
9. Tagged summarizer	Long documents	Summary grounded in the document only
10. Data explainer	Spreadsheet data	“Skip obvious observations”
11. Devil’s advocate	Testing plans	Bans the compliment sandwich
12. Code explainer	Non-developers	Analogy first, risk note last
13. Automation spec writer	Process automation	Failure cases are a required section
14. Practitioner tutor	Learning fast	80/20 framing plus an exercise
15. Plan stress-tester	Quarterly planning	“Which goal would you cut?”

How to Adapt These Prompts to Your Own Work

A prompt list is a starting point, not a finish line. Three habits turn these 15 into a system that compounds.

Test on real material first. Run each prompt you like on one genuine task before trusting it. Where the output falls short, edit the prompt, not just the output. Usually the fix is adding one constraint or one example. If you’d rather have Claude do that tuning for you, Anthropic ships tools for exactly this, and I walked through them in my guide to Claude prompt generators.

Give recurring prompts a permanent home. Any prompt you use weekly belongs in a Claude Project as custom instructions, so you stop pasting it every time. Set up one Project per recurring job: one for the newsletter, one for client emails, one for planning reviews. My walkthrough on how to use Claude Projects covers the setup in about five minutes.

Change one variable at a time. When a prompt underperforms, resist rewriting the whole thing. Adjust the single weakest part, often the output format line, and rerun it. You’ll learn what each part contributes, which is the skill that eventually lets you write strong prompts cold.

And a budget note: every prompt on this page runs on Claude’s free plan. The practical limit is volume and document length, not capability. If you’re pasting long documents daily, the free tier’s usage caps will pinch, and my breakdown of the Claude free tier covers where the paid line actually sits.

Frequently Asked Questions

What are the best prompts for Claude AI beginners?

Start with the tagged summarizer (prompt 9) and the polite-but-firm email (prompt 5). Both produce an obviously useful result on the first try, and both teach the core habits: separate your material with tags, state the goal, and pin the output format. Once those feel natural, the rest of the list is variations on the same pattern.

Do these prompts work on Claude’s free plan?

Yes, all 15 work on the free plan. The free tier limits how much you can use Claude per day and how much text fits in one go, but it doesn’t restrict prompt quality or structure. Heavy daily use, very long documents, or big data pastes are where a paid plan starts to earn its cost.

Why do Claude prompts use tags like <notes> and </notes>?

The tags mark where your pasted material begins and ends, so Claude treats it as content to work on rather than instructions to follow. Anthropic recommends this structure in its own documentation because Claude was trained to recognize it. Without tags, a sentence inside a pasted document can get mistaken for something you asked Claude to do.

Are Claude prompts different from ChatGPT prompts?

The principles transfer almost entirely: clear task, separated material, an example, a pinned format. The main practical difference is that Claude responds especially well to XML-style tags for structure. If you’re weighing the two tools for work use, my honest comparison of Claude vs ChatGPT for business goes through where each one fits.

How long should a Claude prompt be?

As long as the task needs, and no longer. A quick question needs one sentence. A reusable work prompt often runs a paragraph or two once it includes a role, the task, tagged material, and format rules, and that’s fine. Length only hurts when it buries the actual ask under unrelated context.

Can I combine several of these prompts?

Yes, and chaining them is often the best workflow. Outline a post with prompt 1, draft it yourself or with Claude, then run the tone-match rewriter (prompt 2) and the social repurposer (prompt 4) on the result. For decisions, run the decision helper (prompt 8) first and feed its recommendation to the devil’s advocate (prompt 11). Each prompt does one job well, and small chains beat one giant prompt.

Sources: Anthropic, “Claude prompting best practices,” Claude Platform Docs. Anthropic, “Economic Index report: Cadences,” June 2026.