Claude prompts: 16 to copy, plus Anthropic's own advice

In brief

The best Claude prompts name a role, paste the source text and say what the answer should look like. Claude is free with limited usage every 5 hours, and Pro costs $20 a month, as of November 2026. This page has 16 original prompts for writing, research, analysis, documents and planning, plus Anthropic's own advice.

What is a good Claude prompt template?

A good Claude prompt names the job, pastes the material and fixes the shape of the answer. Here is a template to copy now. Replace everything in [brackets] and paste it into Claude, ChatGPT, Gemini or Whizi:

A fountain pen resting on a black plinth beside a terracotta Claude starburst on its own plinth, with a small stack of blank index cards leaning against the pen, in a charcoal studio under white rim light.

"You are [a role, such as a plain spoken editor]. Here is my material: <document>[paste it here]</document>. Task: [what you want done]. Audience: [who will read it]. Format: [length and layout]. If anything is unclear, ask me one question before you start."

That one habit, asking for a question first, saves most rewrites. The 16 prompts below are the same template filled in for common jobs. They cover writing, research, analysis, documents and planning. For code, see Claude coding prompts, which this page does not repeat.

What does Anthropic say makes a prompt work?

Anthropic's own guide says be plain, add context, use examples, tag your content and set a role. These are the lines worth knowing, quoted from its prompting best practices, read 15 November 2026.

On being direct: "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 is the next line: "Show your prompt to a colleague with minimal context on the task and ask them to follow it. If they’d be confused, Claude will be too."

On context: "Providing context or motivation behind your instructions, such as explaining to Claude why such behavior is important, can give Claude more information about your goals and help it deliver more targeted responses." So add the because. "Keep it short, because it goes on a phone screen" beats "keep it short."

On examples: "Examples are one of the most reliable ways to steer Claude’s output format, tone, and structure." It asks you to make them relevant, diverse and structured, and to "Wrap examples in <example> tags (multiple examples in <examples> tags) so Claude can distinguish them from instructions."

On XML tags: "XML tags help Claude parse complex prompts unambiguously, especially when your prompt mixes instructions, context, examples, and variable inputs." Its own tag names are <instructions>, <context> and <input>. Use the same names every time.

On roles: "Setting a role in the system prompt focuses Claude’s behavior and tone for your use case. Even a single sentence makes a difference." In the chat app you have no system prompt box, so put the role as the first sentence of your message, or in a project's instructions. Claude Projects shows where that goes.

What are the best Claude prompts for writing?

Writing prompts work when you give Claude a sample of your own voice and a rule to follow. Fill the brackets in each.

  1. Rewrite in my voice: "Here are 400 words I wrote: <sample>[paste]</sample>. Describe my voice in five rules, such as sentence length and how I open paragraphs. Then rewrite <draft>[paste]</draft> following those rules. Don't add any fact I haven't given you."
  1. Cut by a third: "Cut this to [word count] words. Keep every number and every name. Remove warm up lines, repeated points and hedging words. Show the result, then list what you removed in three bullets. <draft>[paste]</draft>"
  1. Tone shifter: "Rewrite this email for [recipient and relationship]. It is currently [too stiff, too blunt, too long]. Keep my ask in the first two lines and my deadline of [date] unchanged. <email>[paste]</email>"
  1. Five openings: "Write five different opening paragraphs for an article titled [title] for [audience]. One starts with a number, one with a short scene, one with a question the reader would actually ask, one with a claim they'll disagree with, and one with a plain statement. Under 50 words each."

What are the best Claude prompts for research?

Research prompts work when you ask Claude to separate what it found in your sources from what it is guessing. Claude only knows what you paste, unless web search is on, so paste the sources.

  1. Read these sources: "Below are [number] sources inside <document> tags. Answer this question using only them: [question]. After each claim, give the document number it came from. If the sources don't answer part of it, say so and don't fill the gap."
  1. Find the disagreement: "Compare these [number] articles on [topic]. List three points where they disagree, quote the exact line from each, and say which disagreement matters most to someone deciding [decision]."
  1. Explain it back: "I'm new to [topic] and have a [job or background]. Explain [concept] in three layers: one sentence, one paragraph, then a worked example using [something I know]. Flag the two most common wrong beliefs."
  1. Question list for an expert: "I'm interviewing [type of expert] about [topic] for [purpose]. Write 12 questions, ordered from easy to hard, that can't be answered with yes or no. Mark the three that would surprise them."

What are the best Claude prompts for analysis?

Analysis prompts work when you say what decision the analysis feeds. A numbers question without a decision gets a summary of the numbers.

  1. Read a spreadsheet: "I've pasted a table of [what it holds]: <data>[paste rows]</data>. I need to decide [decision]. Give me the three patterns that matter for that decision, the number behind each, and one thing the data can't tell me."
  1. Argue against me: "Here is my plan: [plan] and my reasons: [reasons]. Act as a skeptical advisor. Give the three strongest objections, then say which one I should answer first and why. Don't soften them."
  1. Pre-mortem: "It is [date a year from now] and [project] has failed. Write the five most likely reasons, from most to least likely, each with an early warning sign I could spot in the first month."

What are the best Claude prompts for documents?

Document prompts work when you hand Claude the file and name the reader. The same prompts give you the text in any app.

  1. Summarize a long document: "Summarize <document>[paste]</document> for [reader] who has two minutes. Give: the decision needed, three key facts, any deadline, and one risk. Under 150 words. Quote anything that is a number or a date exactly."
  1. Turn notes into minutes: "Here are my rough notes from a meeting: <notes>[paste]</notes>. Write minutes with decisions, owners and dates in a table, then a short list of open questions. Where my notes don't say who owns something, write unassigned."
  1. Reply to a hard email: "I got this email: <email>[paste]</email>. Draft a reply that [says no, asks for more time, pushes back on the price]. Be firm and polite. Give me two versions, one under 80 words and one under 150. Don't apologize more than once."

What are the best Claude prompts for planning?

Planning prompts work when you give Claude your real limits first, such as hours, money and deadline. A plan without limits is a wish list.

  1. Weekly plan: "I have [number] hours this week, fixed commitments on [days], and three goals: [goals]. Build a plan by day. Put the hardest goal in my best energy window, which is [morning or evening]. Leave one empty block a day. Ask me one question first if my goals clash."
  1. Break down a big goal: "My goal is [goal] by [date]. I'm starting from [where you are] and can give it [hours] a week. Split it into milestones with a finish line you can check for each, and list the first three tasks I can finish in under an hour."

Keep the best ones. Paste your favorite three into a project's instructions, so each new chat starts with them already loaded.

Which app should you run Claude prompts in?

Run them in Claude if you only want Claude, and in a multi model app if you want to try the same prompt on more than one. The table below shows what each costs as of November 2026, from each company's own page.

WhereFree useSign-up neededPaid priceBest at
Claude"Limited usage" every 5 hours and every week; up to 5 projectsYesPro $20 a month, $17 billed annuallyThe prompts on this page, tuned by its maker
ChatGPTFree plan, $0; everyday text chats listed as "Unlimited*"YesGo $8, Plus $20, Pro from $100 a monthThe same prompts with a different voice
GeminiFree without a plan: standard limits and a 32k token context windowGoogle AccountPaid Google AI plansPrompts that need Google apps
Whizi10 messages a day on Whizi AI, Auto and Claude Haiku, no cardYesPro first week $1.99, then $29.99 a monthTrying one prompt on Claude, GPT and Gemini
Anthropic docsFree to read; examples, not a chatNot statedNot applicableLearning why a prompt works

Whizi loses on one thing. If you will only ever run a prompt on Claude, Claude's own app does the job and Whizi adds nothing. Free Claude and free ChatGPT are enough for most of the 16 prompts here, and need no card. The numbers come from Claude's pricing page and ChatGPT's pricing page, read 15 November 2026, and from Google's Gemini limits page.

Anthropic prints no free message count, only "limited usage", so that limit shows up when you hit it. OpenAI lists everyday text chats as "Unlimited*" on Free, with fair use rules behind the asterisk. If a long document prompt fails on a free plan, the cause is usually the window, not the prompt. What is a context window explains why.

What mistakes ruin a Claude prompt?

Five mistakes cause most bad answers, and each has a one line fix.

  1. Asking with no material. A prompt like "summarize my report" with no report gets a generic reply. Fix: paste the text inside tags.
  2. Leaving out the reader. "Explain this" has no right answer until you say to whom. Fix: add "for [audience]".
  3. Burying the ask. A long prompt with the real request in the middle gets half done. Fix: put the task in one labeled line, and the format in another.
  4. Saying "make it better". Claude has to guess what better means. Fix: name the fault, such as "too formal" or "buries the deadline".
  5. Trusting facts it can't see. Claude can state a wrong number in a calm tone. Fix: ask for the source line for every number, and check the ones that matter.

Anthropic's guide says to ask for extras outright. If you want Claude to do more than the task strictly requires, the guide says to "explicitly request it". Don't make Claude guess.

Can you run all these prompts on one app?

Yes, on Whizi, which publishes this page. Pro carries Claude, GPT and Gemini in one app, and its first week is $1.99. You can paste prompt 10 into Claude, then send the same text to GPT and compare the objections side by side. Pro and Powerhouse have a compare view for that.

The free plan gives you 10 messages a day on Whizi AI, Auto and Claude Haiku, with no card. Word, Excel, PDF and slide files made in chat have no plan gate, so prompts 12 and 13 can end as a file. Whizi projects take custom instructions, so you can pin your favorite prompts once. See how projects work.

What can Claude prompts not do?

A prompt can't make Claude know what you haven't shown it, and it can't make an answer true. Three limits matter.

First, accuracy. Models can write a wrong fact fluently. For legal, medical or money questions, use the answer to prepare questions for a professional, not to replace one.

Second, memory. Claude keeps nothing from one chat to the next unless you store it in a project or paste it again, so keep your best prompts in a note.

Third, ownership. The U.S. Copyright Office's January 2025 report says AI outputs "can be protected by copyright only where a human author has determined sufficient expressive elements." Prompts alone don't count, so edit what you plan to publish. That is what the Office has published, and it isn't legal advice.

Workflow checklist
  • Open with one sentence naming the role
  • Paste the real source text inside tags
  • Name the reader and the decision
  • Fix a length and a format
  • Add one example when the format matters
  • Ask for the source line behind every number
Common questions

Frequently asked questions

What are the best Claude prompts?

The best Claude prompts give a role, paste the real material and fix the format of the answer. The 16 on this page follow that template for writing, research, analysis, documents and planning. Fill the brackets and paste. Anthropic's own guide recommends the same moves, plus tags and examples.

Are Claude prompts different from ChatGPT prompts?

Mostly no, but Anthropic specifically recommends XML tags for Claude. Its guide says they help Claude parse prompts that mix instructions, context, examples and inputs. A plain prompt works in both. Best ChatGPT prompts has a list written for that app.

How do you write a good prompt for Claude AI?

Start with a role, paste the source, say who it is for and fix the length and format. Then ask Claude to question you before it starts. Anthropic's test is to show the prompt to a colleague with no context, and if they'd be confused, Claude will be too.

Is Claude free to use for prompts?

Yes, Claude has a free plan with limited usage every 5 hours and every week, as of November 2026, per its pricing page. Pro costs $20 a month. Anthropic doesn't print a free message count. How to use Claude for free covers the limits.