--- title: "Claude Prompt Generator: A Reusable Prompt to Build Prompts" description: "Use Claude itself as a prompt generator. Copy a meta-prompt that interviews you, writes a structured prompt, and gets saved for reuse, with a worked example." url: https://www.promptpal.app/blog/claude-prompt-generator-reusable-prompt author: "Nwokemodo Chibuike Daniel" published: 2026-11-03T08:02+00:00 updated: 2026-11-03T13:57:31.053Z category: "General" tags: ["claude", "prompt generator", "prompt engineering", "prompt templates", "prompt management"] language: en reading_time_minutes: 7 image: https://images.ctfassets.net/xzkuxcljsk12/3iTmYkrxMPpF6Yy1iUXfPA/19ca9c48eaed67b3dbf640f245833944/claude-prompt-generator.png publisher: PromptPal --- # Claude Prompt Generator: A Reusable Prompt to Build Prompts Use Claude itself as a prompt generator. Copy a meta-prompt that interviews you, writes a structured prompt, and gets saved for reuse, with a worked example. The best Claude prompt generator is Claude. You give it a short meta-prompt that tells it how to interview you and what a good prompt looks like, and it writes the prompt you paste into a fresh chat. Below is a complete meta-prompt you can copy, a worked example, and a way to keep what you build. Standalone generator sites exist, and some are fine for a quick start. But they don't know your task, your audience or your quality bar. A meta-prompt you control does, because you can edit it. ## Why a generator beats writing from scratch Most weak prompts fail the same way: they leave out the facts Claude needs. Who is the audience? What does the finished output look like? What should it avoid? You know these things, but you forget to type them when you're in a hurry. A generator fixes that by asking. Instead of staring at a blank box, you answer four or five questions, and the structure comes for free. You also get consistency. Two people on a team using the same generator will produce prompts with the same skeleton, which makes the outputs easier to compare. There is a limit. A generator can't invent context you don't have. If you answer "make it good" to every question, you'll get a tidy prompt that produces generic text. The quality of the output still tracks the quality of your answers. ## The meta-prompt: copy and paste this This one works in a normal Claude chat. It has Claude ask questions first, which matters: if you let it write immediately, it fills gaps with guesses. > `You are a prompt writer. Your job is to write one high-quality prompt that I will paste into a new chat with Claude.` > > > > `Step 1: Ask me up to five questions, one message, numbered. Cover only what you need: the task, who the output is for, the input I will provide, the format I want back, and anything the output must avoid. Skip any question my request already answers.` > > > > `Step 2: After I answer, write the prompt. Use this structure:` > > `- Role: one sentence on who Claude should act as.` > > `- Task: what to do, in plain imperative sentences.` > > `- Context: the facts Claude needs, with [placeholders] for anything that changes each time.` > > `- Constraints: length, tone, things to avoid.` > > `- Output format: exactly what the response should look like.` > > `- If information is missing, tell Claude to ask before answering.` > > > > `Step 3: Put the finished prompt in a single code block. Under it, list in two or three bullets which placeholders I need to fill in.` > > > > `Do not run the prompt yourself. Do not add commentary before the code block.` > > > > `My request: [describe the task in one or two sentences]` Two details are doing real work. "Do not run the prompt yourself" stops Claude from answering instead of writing. And the placeholder list at the end turns a one-off prompt into a template you can reuse. ## Worked example: a weekly client update Say you're an account manager who sends a status email every Friday. Your request line is: > Write a prompt that turns my raw weekly notes into a client status email. Claude comes back with questions. Here are plausible answers: 1. The client is a marketing director who skims on mobile. 2. Input is bullet notes, usually messy, sometimes with internal jargon. 3. Output should be under 150 words with a short list of what shipped, what's blocked and what we need from them. 4. Tone is warm and direct. No apologies for delays unless I say so. 5. Never mention internal budget or staff names. The prompt it generates looks like this: > `You are an account manager writing a weekly status email to a client.` > > > > `Task: Turn the raw notes below into a client-ready status email.` > > > > `Context:` > > `- Reader: [client name and role]. They skim on mobile and want to know what changed and what is needed from them.` > > `- Project: [project name]` > > `- Raw notes: [paste notes]` > > > > `Constraints:` > > `- Under 150 words.` > > `- Warm, direct tone. Do not apologise for delays unless the notes explicitly call for it.` > > `- Remove internal jargon. Never mention internal budget or staff names.` > > > > `Output format:` > > `- One-line subject.` > > `- Three short sections with bold labels: Shipped, Blocked, Needed from you.` > > `- One-sentence sign-off.` > > > > `If the notes are unclear about whether something shipped or is still in progress, ask me before writing.` Compare this with the first thing most people type: "Turn these notes into a client email." That version gets you a friendly, vague email of about 250 words, with the internal jargon polished into longer sentences. The generated version gets a 120-word email with the three labelled sections, and the jargon gone. The last line also matters. When a note says "auth work mostly done," Claude asks whether it shipped rather than quietly picking one. ## Tuning the generator for how Claude responds A few adjustments make the output better. Whether they matter depends on the model version you're using, so test them rather than assuming. ### Ask for XML-style tags when the input is long Claude generally handles clearly separated sections well. If your prompt includes a long pasted document, tell the generator to wrap it in tags like `` and refer to them by name in the task. This stops instructions and source material from blurring together. Add this line to Step 2 of the meta-prompt: > `When the prompt includes pasted material, wrap each piece in descriptive XML-style tags and refer to those tags by name in the instructions.` ### Add an example when format matters If you need an exact output shape, one sample beats a paragraph of description. Ask the generator to include a short example output, with a note telling Claude to match the structure but not copy the content. Skip this if the task is open-ended, because an example will pull every response toward it. ### Use a writer-specific variant For drafting and editing, change the interview. A prompt writing generator for writers needs different questions: voice, a sample of your past work, what you refuse to write, and where the piece will be published. > `You are a prompt writer helping a professional writer. Ask me up to five numbered questions about: the piece type, the publication and reader, my voice (ask me to paste a 150-word sample if I have one), banned phrases and habits, and the length. Then write one reusable prompt in a single code block with [placeholders] for the topic and source material. Do not write the piece itself.` > > > > `My request: [describe the writing task]` If you write for a living, the workflow in [ChatGPT for Writing: A Practical Workflow With Prompts](https://promptpal.app/blog/chatgpt-for-writing-workflow-prompts) transfers to Claude with minor changes. ## Test the generated prompt before you trust it A prompt that reads well can still fail on real input. Run it three times with different inputs: a typical one, a messy one and an awkward edge case. For the client email, that means clean notes, notes full of jargon, and notes with nothing blocked. When something goes wrong, don't patch the output by hand. Fix the prompt. Paste the weak output back into the same chat as the generator and say: > `The prompt you wrote produced this output: [paste output]. The problem is [what went wrong]. Revise the prompt so this does not happen, and tell me which line you changed.` Asking which line changed is useful. It keeps revisions small and stops Claude from rewriting a working prompt just to fix one flaw. ## Keep the prompts you generate The generation step takes a couple of minutes. The cost shows up later, when you need the same prompt next month and can't find the chat. Anything you've run more than twice belongs somewhere you can search. In PromptPal you can save the finished prompt to your [prompt library](https://promptpal.app/features/prompt-library), keep the `[placeholders]` intact, and version it as you tune it. That matters more than it sounds: when a revision makes things worse, you can go back. The same applies to the meta-prompt itself. Save your generator, and when you improve it, you improve every prompt it produces from then on. If you want the reasoning behind this, [prompt version control](https://promptpal.app/prompt-version-control) covers it in detail. For a team, save one agreed generator in a shared workspace so everyone's prompts follow the same structure. Our post on [building a prompt library your team will use](https://promptpal.app/blog/build-prompt-library-team-will-use) covers how to keep that from turning into a junk drawer. ## When a generator is not worth it Skip it for one-off questions. If you're asking Claude to explain a concept or rephrase a sentence, a plain request is faster, and a five-question interview is overkill. Skip it too when you already have a prompt that works. Improve that one directly. And if what you need is a tweak to an existing prompt rather than a new one, an optimiser approach is better; the principles in [ChatGPT Prompt Optimizer: How to Improve Any Prompt](https://promptpal.app/blog/chatgpt-prompt-optimizer-improve-prompts) apply to Claude as well. Use a generator for recurring tasks where the structure matters: reports, emails, reviews, summaries, content briefs. Those are the places where one good template repays the setup many times. ## Next step Copy the meta-prompt above, generate a prompt for a task you repeat every week, and save it. If you want ready-made starting points instead, browse the [public prompt library](https://promptpal.app/prompts/public) and adapt a template to your own task. --- Originally published at https://www.promptpal.app/blog/claude-prompt-generator-reusable-prompt