--- title: "Prompts for Claude: 7 Templates You Can Paste Today" description: "Seven copy-ready prompts for Claude, plus notes on system prompts, Claude Code and resumes, and how to fix a prompt that keeps giving you bad output." url: https://www.promptpal.app/blog/prompts-for-claude-templates author: "Nwokemodo Chibuike Daniel" published: 2026-11-04T08:00+00:00 updated: 2026-11-04T08:00+00:00 category: "General" tags: ["claude", "prompt templates", "system prompts", "claude code", "resume prompts", "claude code prompts", "claude prompts"] language: en reading_time_minutes: 11 publisher: PromptPal --- # Prompts for Claude: 7 Templates You Can Paste Today Seven copy-ready prompts for Claude, plus notes on system prompts, Claude Code and resumes, and how to fix a prompt that keeps giving you bad output. The best prompts for Claude are specific about the task, give Claude the material to work from, and say what the output should look like. Claude responds well to clear structure, so wrapping inputs in tags and stating the format you want does more than any clever phrasing. Below are seven templates you can paste as written, followed by notes on system prompts, Claude Code, resumes and prompt generators. Replace anything in `[square brackets]` and run it. ## What makes a prompt work well in Claude Four habits cover most of the gain. None of them is exotic. - **Separate instructions from material.** Put pasted text inside XML-style tags such as `` or ``. Claude handles this cleanly, and it stops your instructions blurring into the content. - **Say who the output is for.** "A skeptical CFO who has two minutes" beats "a professional tone". - **Describe the format.** Word limit, number of bullets, headings or no headings. If you don't say, you get Claude's default, which is often longer than you wanted. - **Ask for what to leave out.** "No preamble, no closing offer to help" removes a lot of padding. Two more habits help once those are routine. First, give the reason behind a rule. "Keep it under 90 words because people read this on their phone between meetings" gives Claude something to reason from when it meets a case your rule didn't cover. A bare limit gets followed mechanically, and sometimes badly. Second, show one example when the format is unusual. A single good sample beats a paragraph describing it. Claude may copy the wording too closely, so add "match the structure, not the content" to stop your sample from leaking into every answer. How strictly Claude follows all this depends on the model version you're using. Newer models tend to follow detailed instructions closely, so if the output feels off, check whether your prompt contradicts itself before blaming the model. ### A before-and-after Here is a weak prompt: > `Summarise this meeting.` You get a generic paragraph that retells the meeting in order. Nobody can act on it, and the one decision that mattered is buried in the third sentence. Now the improved version: > `You are summarising a meeting for people who did not attend and need to know what to do next.` > > > > `` > > `[paste transcript or notes]` > > `` > > > > `Write:` > > `1. Decisions made (one line each)` > > `2. Action items, formatted as "Owner: task, due date" (write "no date given" if missing)` > > `3. Open questions nobody answered` > > > > `Use only what is in the transcript. If an owner is unclear, write "unassigned". No introduction and no closing remarks.` Instead of a narrative you get three short lists, and the "unassigned" rule stops Claude inventing owners to make the list look complete. That one rule is the difference between a summary you can forward and one you have to check line by line. ## Seven prompts for Claude you can paste today ### 1. Meeting summary The prompt above is the template. Keep it as is. It works for call notes, interview debriefs and project check-ins, as long as you paste the actual transcript or notes rather than describing the meeting from memory. ### 2. Cold email rewrite Use this when you have a draft that reads like every other cold email. Claude is good at cutting; the rules below tell it where to cut. > `Rewrite the cold email below so a [job title] at a [company type] would reply to it.` > > > > `` > > `[paste draft]` > > `` > > > > `Rules:` > > `- Under 90 words` > > `- Open with something specific to them, not about us` > > `- One clear ask, phrased as a question that can be answered with yes or no` > > `- No phrases like "I hope this finds you well" or "quick call"` > > > > `Give me two versions with different opening lines, then one sentence on which you would send and why.` Asking for two versions matters more than it looks. The opening line is the part that decides whether the email gets read, and comparing two gives you a sense of what to test. ### 3. Bug report from rough notes This one saves the most time for non-engineers who have to file tickets. > `Turn these rough notes into a bug report for an engineering team.` > > > > `` > > `[paste notes, error messages, screenshots described in words]` > > `` > > > > `Use these headings: Summary, Steps to reproduce, Expected result, Actual result, Environment, Severity (with a one-line reason).` > > > > `If something needed for reproduction is missing, list it under a final heading "Questions for the reporter" rather than guessing.` Here's what it does with messy input. Say your notes read: "checkout broke after the coupon thing, Sam saw a spinner forever on mobile, maybe Safari, started Tuesday?" The output gives you a one-line summary, steps that say "apply a coupon code at checkout" and nothing invented beyond that, and a final section asking which coupon, which iOS version, and whether it happens when no coupon is used. The "Questions for the reporter" heading is what turns a guess into a follow-up. ### 4. Document analysis with quotes Use this for contracts, reports, research papers or any long document where you need an answer you can trust. > `I'm going to give you a document. Answer my question using only that document.` > > > > `` > > `[paste document]` > > `` > > > > `Question: [your question]` > > > > `First, pull out the exact quotes that are relevant and put them in tags. Then answer the question based on those quotes. If the document does not contain the answer, say so plainly.` Extracting quotes first is the single most useful trick for long documents. It keeps the answer anchored to the text, and you can check the quotes against the source in seconds. ### 5. Editing without rewriting > `Edit the text below for clarity. Keep my voice and my structure.` > > > > `` > > `[paste text]` > > `` > > > > `Return:` > > `1. The edited text` > > `2. A list of the five most significant changes you made and why` > > > > `Do not add new claims or examples. Cut anything that repeats an earlier sentence.` The change list is there to teach you something. After a few rounds you'll notice you make the same mistakes repeatedly, and you can move those fixes into your own first draft. ### 6. Decision memo > `Help me decide between these options: [option A], [option B], [option C].` > > > > `Context: [what we're trying to achieve, constraints, budget, timeline]` > > > > `For each option give the strongest argument for it, the biggest risk, and what would have to be true for it to be the right choice. Then tell me which you would pick and what single piece of information would change your mind.` Asking what would change its mind pulls out the real assumptions behind the recommendation. Often that one line tells you what to go and find out before deciding. ### 7. Prompt critique > `Here is a prompt I use regularly:` > > > > `` > > `[paste your prompt]` > > `` > > > > `The task it is meant to do: [describe task]` > > `What goes wrong: [describe typical bad output]` > > > > `Diagnose why it produces that output. Then rewrite it to fix the problem, keeping any parts that work. Explain each change in one sentence.` The "what goes wrong" line is the part people skip, and it's the part that makes the critique useful. Without it Claude polishes the prompt in general instead of fixing your actual problem. If you do this often, the [Refine agent](https://promptpal.app/features/refine-agent) does a similar job on a saved prompt without you retyping the critique request. ## Claude system prompts that stay useful A system prompt is the standing brief Claude reads before every message. Use it for things that never change: role, audience, tone, format and hard rules. Keep task-specific detail in the normal message. Where you set it depends on how you use Claude: project or custom instructions in the apps, a system parameter if you call it through the API. Here is a general one for a work assistant: > `You are a writing and analysis assistant for [team or role] at [company type].` > > > > `Audience: colleagues who are busy and read on mobile. Lead with the answer, then the support.` > > > > `Style: plain English, short paragraphs, no filler openings, no closing offers of further help.` > > > > `Rules:` > > `- If a request is ambiguous, ask one clarifying question before answering.` > > `- If you are not sure of a fact, say so instead of guessing.` > > `- Use bullet points only for lists of three or more parallel items.` > > `- Keep answers under [number] words unless I ask for more.` A narrower system prompt usually beats a broad one. Here is a version for a single job, drafting replies to customer emails: > `You draft replies to customer support emails for [company name], which sells [product].` > > > > `Tone: warm, direct, no corporate phrases. Never apologise more than once per reply.` > > > > `Every reply must: name the customer's actual problem in the first sentence, give the next step or the fix, and say when they will hear from us again.` > > > > `Never promise refunds, discounts or delivery dates. If the customer asks for one, write: "I've passed this to the team who can confirm it" and flag the request to me in a line before the draft.` > > > > `If the email is angry, keep the reply shorter, not longer.` Keep system prompts short. Every rule you add is one more thing that can clash with a later instruction. When a rule keeps getting ignored, the usual cause is that another line quietly contradicts it. Test any system prompt with three awkward messages before you rely on it: one vague, one off-topic, and one that pushes against a hard rule. ## Claude Code prompts Claude Code works best when you describe the outcome, the boundaries and how to check the work. Vague requests like "fix the auth bug" send it exploring your whole repo. > `Task: [describe the bug or feature in two or three sentences]` > > > > `Where to look: [files or directories]` > > `Do not touch: [files, APIs or behaviours that must stay the same]` > > > > `Before changing code, read the relevant files and tell me your plan in a few bullet points.` > > > > `When done:` > > `- Run [test command] and show me the result` > > `- List every file you changed and why` > > `- Flag anything you were unsure about` The plan step catches wrong assumptions cheaply, before any code changes. If the plan is off, correct it in one message and let it continue. For reviewing code rather than writing it, use a read-only version: > `Review the changes in [branch or file list] as a reviewer who has to maintain this code.` > > > > `Do not edit anything. Report:` > > `1. Bugs or edge cases likely to fail, with file and line` > > `2. Anything that breaks the conventions used elsewhere in this repo` > > `3. Missing tests, listed by behaviour, not by function name` > > > > `Skip style nitpicks that a formatter would catch.` For persistent project rules such as coding conventions and commands to run, a context file works better than repeating them in every prompt. Our post on [AGENTS.md, Skills or MCP](https://promptpal.app/blog/agents-md-skills-or-mcp) covers when to use which. ## Claude resume prompts The most common mistake is asking Claude to "improve" a resume. It rewrites everything and sometimes adds claims you never made. Narrow the job instead. > `I'm applying for the role below. Tailor my resume bullets to it.` > > > > `` > > `[paste job description]` > > `` > > > > `` > > `[paste resume]` > > `` > > > > `Rules:` > > `- Do not invent experience, numbers or tools I haven't listed` > > `- Rewrite each bullet to start with a strong verb and show the result where my resume already states one` > > `- Move the three most relevant bullets for this role to the top of each position` > > `- After the rewrite, list any requirements in the job description my resume doesn't address, so I can decide whether to add real examples` The last rule is the valuable one. It gives you a gap list you can fix honestly, instead of a resume that sounds polished and says nothing new. If a gap is real, go back to your own history and find a true example, then paste it in and run the prompt again. Read the output before you send it. Claude can keep a number you gave it but attach it to the wrong job, and a recruiter will spot that faster than you will. ## Claude prompt generators: when they help A generator is useful when you know the task but not the structure. You don't need a separate tool for the first draft, because Claude can interview you: > `I want a reusable prompt for this task: [describe task].` > > > > `Ask me up to five questions, one at a time, about the audience, the input I'll provide, the output format and what bad output looks like. Wait for my answer each time.` > > > > `Then write the prompt with [placeholders] for anything that changes between uses, and put the input in tags.` The questions do the real work, because they force you to state things you had only assumed. Once you have a prompt that works, stop generating and start reusing. The goal is one tested prompt per recurring task, not a new one every Monday. ## Fixing a Claude prompt that keeps missing Most bad output has one of four causes, and each has a direct fix. - **Too long or padded.** Add a hard word limit and the line "No preamble, no closing remarks." State the limit as a number, not as "brief". - **Invents facts.** Add "Use only the material in the tags. If it isn't there, say so." Then ask for quotes, as in template 4. - **Ignores your format.** Check for a second instruction that conflicts with it, then show a short example of the format you want. - **Inconsistent between runs.** Your prompt is leaving a decision open. Find the choice Claude is making differently each time, and make it for Claude. Change one thing at a time. If you rewrite the whole prompt, you won't know which change helped. ## Keep the prompts that work Most people write a good prompt once, lose it in a chat history, and write a worse one next week. Save the ones from this post somewhere you can find them. In PromptPal you can store a prompt with its placeholders, edit it over time and keep earlier versions, so you can see which wording actually improved the output. [Prompt version control](https://promptpal.app/prompt-version-control) explains how that works. One habit pays off quickly: when a prompt gives a bad result, change one thing, rerun it on the same input, and keep the change only if the answer improved. ## Next step Pick the two prompts above you'd use this week and save them with their placeholders filled in for your own work. The [PromptPal prompt library](https://promptpal.app/features/prompt-library) is where you can keep them, edit them and reuse them in Claude or any other model. --- Originally published at https://www.promptpal.app/blog/prompts-for-claude-templates