Answer: Browse, search and copy a curated collection of ready-to-use AI prompts organised by task and model.
60+ ready-to-use prompts for ChatGPT, Claude, and any AI assistant
A prompt is a work order for a language model: the instructions, plus whatever context the model needs to fill them. Same model, same day, same topic — a vague prompt gets you a generic answer, and a precise one gets you something you can ship. Everything here applies to ChatGPT, Claude, and Gemini alike, because the discipline transfers even when the models differ.
The habits that matter most:
• Replace bracketed placeholders like [topic] with your actual content — leftover brackets are the number one cause of generic output
• Add your brand voice, target audience, and tone preferences
• Specify the output format (bullet points, table, essay, code) before you send
• Iterate — the first answer is a draft; follow-up turns are where prompts get good
• Chain prompts: use the output of one prompt as input to the next
Fifty-three prompts across ten categories, from SEO content briefs to customer-service de-escalation. Filter by category chip, or search by task — the search matches titles, descriptions, and the full prompt text, so typing "meta description" surfaces every prompt that touches one.
| Category | Prompts | Typical use |
|---|---|---|
| SEO | 6 | Keyword clustering, title tags, meta descriptions, content briefs, competitor gaps, FAQ schema |
| Marketing | 6 | Subject lines, landing pages, onboarding sequences, social calendars, ad variants, brand voice |
| Content Writing | 6 | Outlines, rewrites, long-form drafts, repurposing, thought leadership, proofreading |
| Development | 6 | Code explanations, reviews, regex, API docs, architecture, unit tests |
| Business | 6 | SWOT, business plans, meeting notes, pricing, cold email, pitch decks |
| Design | 3 | Color palettes, wireframe descriptions, design-system docs |
| Productivity | 5 | Task breakdowns, weekly planning, decision frameworks, email replies, SOPs |
| Education | 5 | Concept explainers, quizzes, study guides, flashcards, syllabi |
| Sales | 5 | Objection handling, discovery calls, proposals, follow-ups, call scripts |
| Customer Service | 5 | Response templates, de-escalation, knowledge bases, feedback asks, win-backs |
Every prompt is static text on this page — clicking a card copies it to your clipboard, nothing is submitted to a server, and there's no account to create.
Compare two requests for the same job. Weak: "write email subject lines for a sale." Strong: the library's Email Subject Line Tester — "Write 10 email subject lines for a [type] email about [topic]. Use different psychological triggers: curiosity, urgency, personalization, numbers, questions. Score each 1-10 on likely open rate. Target audience: [audience]." Same model, same task; the second one comes back shaped like something you'd actually send. The difference is that every strong prompt carries the same five load-bearing parts:
| Element | What it does | Example fragment |
|---|---|---|
| Role | Sets expertise and voice | "You are a personal-finance editor for a credit-union newsletter" |
| Task | Names the single action | "Write a 700-word article on tracking expenses" |
| Context | Supplies facts the model can't guess | "Audience: first-job 22-year-olds with no investing background" |
| Constraints | Draws checkable boundaries | "No jargon; 8th-grade reading level; US examples" |
| Format | Defines the output shape | "Markdown with H2 sections and a five-question FAQ" |
You rarely need all five — but when an answer disappoints, the missing element is almost always one of them. No audience means generic tone; no format means a wall of text; no constraints means maximum length and minimum specificity.
Long isn't the same as specific. A three-paragraph prompt that never states the audience or format produces mush, while two tight sentences with the right constraints produce usable output. Specificity lives in constraints you can count: ten items, under 60 characters, five questions, US spelling, flag unverifiable statistics as [statistic needed]. A constraint you can check is a constraint the model can follow.
Chaining beats enlarging, too. Rather than one mega-prompt asking for outline, draft, and social posts at once, run outline → draft → edit → repurpose as separate turns, each doing one job. The library's Content Repurposer is a ready-made chain for the last link: one blog post in, a thread, a LinkedIn post, a newsletter, and captions out.
Yes, without modification — they're plain instructions, not ChatGPT-specific syntax. Models differ a little in default length and tone, so expect to iterate once rather than zero times. If a prompt underperforms on a particular assistant, the fix is the same as within one model: add the missing constraint rather than rewriting from scratch. The beginner and intermediate prompts run comfortably on smaller, faster models; the advanced ones — FAQ schema generation, system architecture reviews — are where a stronger model visibly earns its keep.
Yes. Every prompt is plain-language instructions with no ChatGPT-specific syntax, so they paste into Claude, Gemini, Copilot, or any chat assistant. Beginner and intermediate prompts run well on smaller models; the advanced ones (schema markup, system architecture reviews) benefit from a stronger model.
Yes. Copy them, adapt the placeholders, and fold them into your own or your clients' workflows with no attribution required. The library is a static page — nothing is submitted anywhere and no account is involved.
Add constraints you can actually check: a length, a format, an audience, a tone, a number of items. If you can't verify whether the output met the instruction, the model couldn't either. Concrete boundaries beat extra paragraphs of vague direction every time.
A system prompt sets standing behavior — role, rules, format — that applies across a whole conversation or workspace. A user prompt is each individual request. In ChatGPT the system layer lives in custom instructions; in the Claude API it's the system field. Most everyday work needs only well-built user prompts.
No. Language models generate probabilistically and their underlying versions change over time, so identical prompts produce different outputs on different runs. For work that matters, generate two or three candidates and pick, rather than trusting the first draft.