You’ve heard the advice: “Just be more specific with your prompts.” Great. But specific how? What does that actually look like when you’re trying to summarise a 40-page report at 8 a.m., draft a cold email before lunch, or debug a function you didn’t write? The gap between knowing AI exists and knowing exactly what to type is where most people get stuck — and where productivity either takes off or quietly dies.
This article solves that problem. By the end, you’ll have a copy-paste AI prompt library organised by the work you actually do — writing, analysis, email, planning, research, and code. Every prompt below is field-tested, immediately usable, and built on the same underlying formula that separates prompts that produce brilliant output from prompts that produce polished nonsense. There are 20 prompts in total. Steal all of them.
What you’ll need
- An AI assistant account — ChatGPT, Claude, or any leading model will work with these prompts
- Access to your tool of choice (see the comparison table below if you’re still choosing)
- A document or notes app to build your personal prompt library as you go
- Five minutes — that’s genuinely all it takes to start seeing results
If you’re based in Algeria or North Africa and haven’t been able to subscribe to a premium AI model because of payment barriers, Click DZ lets you pay in Algerian dinar (DZD) via CIB, EDAHABIA or BaridiMob — no international card needed, activation within minutes.
The P.A.C.E. Prompt Formula — your framework for every prompt you’ll ever write
Before handing you 20 prompts, you need the skeleton behind them. Once you see it, you’ll be able to write your own prompts on the fly — and fix the ones that aren’t working. The framework is called P.A.C.E.:
| Letter | Stands for | What to include | Example fragment |
|---|---|---|---|
| P | Persona | Who the AI should act as | “Act as a senior product manager…” |
| A | Action | The precise task to complete | “…write a one-page brief…” |
| C | Context | Background, constraints, audience | “…for a B2B SaaS product targeting HR teams…” |
| E | Examples / Format | Output structure, tone example, length | “…format as: Problem / Solution / Next step. Max 300 words.” |
P.A.C.E. is deliberately short. You don’t need all four elements in every prompt — but whenever a response feels vague or off-target, ask yourself which element you skipped. It’s almost always C (context) or E (format). Now let’s use it.
Step 1 — Writing prompts (copy these now)
Writing tasks are where AI earns its keep fastest. The key is giving it a voice anchor — a word count, a tone description, or an example sentence — so it doesn’t default to “professional newsletter” mode for everything.
Blog post outline from a rough idea
Act as a content strategist with 10 years of SEO experience. I have a rough idea for a blog post: [PASTE YOUR IDEA IN ONE SENTENCE]. Create a detailed outline with: one working H1 title (keyword-first), 5–6 H2 sections, two bullet points under each section explaining what I'll cover, and a one-sentence meta description under 160 characters. Audience: [describe your reader]. Tone: conversational but authoritative. Do not write the full article yet.
First draft from bullet points
I have rough bullet notes for a section of an article. Turn them into flowing, natural prose — no bullet points in the output. Keep every idea but make it read like a human wrote it after thinking carefully. Match this tone: confident, friendly, practical (like a smart colleague explaining something over coffee). My bullet notes: [PASTE YOUR BULLETS HERE] Target length: approximately [X] words.
Rewrite to a different tone
Rewrite the following text. Keep every fact and argument intact — nothing added, nothing removed. Change only the tone to: [DESCRIBE TONE, e.g. "urgent and punchy", "warm and encouraging", "formal and precise"]. Original text: [PASTE TEXT]
Turn a LinkedIn post into a Twitter/X thread
Convert this LinkedIn post into a Twitter/X thread. Rules: tweet 1 is a hook that makes someone stop scrolling (under 280 chars, no hashtags in tweet 1). Tweets 2–8 each deliver one clear idea. Final tweet is a strong takeaway or call to action. Number each tweet. LinkedIn post: [PASTE POST]
Step 2 — Analysis prompts
Analysis prompts need the most context of any category. Vague input produces vague output. When you paste data or a document, always tell the AI what decision you’re trying to make — it changes everything about what it surfaces.
Summarise a long document with a decision lens
Read the following document and give me a structured summary. I need to make a decision about [DESCRIBE DECISION]. Structure your summary as: 1. Core argument or finding (2–3 sentences) 2. Key data points I should not miss (up to 5 bullets) 3. Risks or caveats the document raises 4. What the document recommends (if anything) 5. My decision lens: based on this, what should I pay most attention to? Document: [PASTE DOCUMENT]
Competitor analysis from public info
Act as a business analyst. Based only on the information I provide (do not invent details), compare these competitors on the following dimensions: pricing model, primary audience, key differentiator, and one visible weakness. My product: [DESCRIBE YOURS] Competitors and info I have: [PASTE INFO] Format as a table. Add a final "So what?" row summarising where the biggest opportunity lies for me.
Find holes in an argument
Steel-man then red-team the following argument. First, explain it as strongly as possible (2–3 sentences). Then list every assumption it relies on. Then give me 3–4 specific counterarguments a sharp critic would raise. Be direct — I want to stress-test this before presenting it. Argument: [PASTE ARGUMENT]
Step 3 — Email prompts
Email prompts shine when you treat the AI like a ghostwriter who knows your voice. Give it one example of an email you’ve written before, and it will match your style for everything after.
Cold outreach with a real hook
Write a cold outreach email. The hook in the first line must reference something specific and real about the recipient (I'll fill in the [SPECIFIC DETAIL] placeholder). Do not use flattery. Get to the offer by line 3. Close with one clear, low-friction call to action. Recipient: [NAME, ROLE, COMPANY] Specific detail to reference: [E.g. "their recent podcast episode on X", "the job post they published for Y"] My offer: [WHAT I'M OFFERING IN ONE LINE] My name and company: [YOUR INFO] Desired tone: direct, human, not salesy Max length: 120 words
Difficult reply (declining, pushing back, setting a boundary)
I need to reply to the email below. I want to [decline / push back on X / set a limit on Y] without damaging the relationship. Keep it respectful, clear, and direct. No hollow phrases like "I hope this email finds you well." Get straight to the point in the first sentence. Original email: [PASTE EMAIL] My reply should convey: [WHAT YOU WANT TO SAY IN PLAIN ENGLISH]
Step 4 — Planning and strategy prompts
These prompts are where AI becomes a thinking partner, not just a text generator. The goal is to get it to structure your thinking, not replace it.
Weekly plan from a messy to-do list
I have an unstructured list of tasks for the week. Help me create a focused weekly plan. First, categorise each task as: Deep work (needs focus), Quick win (under 20 min), Waiting-on (blocked), or Delete/defer. Then suggest a daily schedule Mon–Fri that front-loads deep work in the morning and batches quick wins in the afternoon. My tasks: [PASTE LIST] Constraints: - Meetings already blocked: [LIST TIMES] - My most focused hours are: [MORNING / AFTERNOON / EVENING]
Project kickoff brief
Write a project kickoff brief for: [PROJECT NAME]. Include: (1) Problem statement — one crisp paragraph, (2) Success metrics — 3 measurable outcomes, (3) Scope: in and out of scope, (4) Key risks and how to mitigate them, (5) First 3 actions to take this week. Context I have: [DESCRIBE WHAT YOU KNOW] Stakeholders: [LIST] Deadline: [DATE]
Step 5 — Research and code prompts
For research, the best prompts are the ones that push the AI to show its work — assumptions, gaps, and uncertainty — rather than just delivering confident-sounding answers. For code, the trick is always to give it the environment and the failure mode.
Research synthesis from multiple sources
I'm going to paste several excerpts from different sources on the topic of [TOPIC]. Your job is to synthesise them — find the common threads, note where sources disagree, and identify what's still genuinely uncertain or debated. Output format: - Consensus view (what most sources agree on) - Key debates (where they diverge and why) - Gaps (what the sources don't address but I should research further) - My takeaway (suggest one clear implication for [MY CONTEXT]) Sources: [PASTE EXCERPTS, LABELLED SOURCE 1, SOURCE 2, etc.]
Explain code and suggest improvements
I'm a [BEGINNER / INTERMEDIATE / ADVANCED] developer working in [LANGUAGE/FRAMEWORK]. First, explain what the following code does in plain English — as if explaining to someone on my team who hasn't seen it before. Second, identify any bugs, inefficiencies, or bad practices. Third, rewrite only the parts that need fixing, with a comment explaining each change. Do not rewrite the entire codebase — only the parts with real issues. Code: [PASTE CODE]
Best AI tools for building a prompt library
| Tool | Best for | Notes |
|---|---|---|
| ChatGPT (GPT-4o) | Writing, email, general tasks | Custom GPTs let you save system prompts permanently; best ecosystem for plugins |
| Claude (Anthropic) | Long documents, analysis, nuanced writing | 200K context window; excellent for pasting entire reports or codebases |
| Gemini Advanced | Research synthesis, Google Workspace integration | Connects to Gmail, Docs, Drive natively; good for research with live sources |
| Notion AI | Storing and reusing prompts in-context | Best for teams who want a shared prompt library inside their existing workspace |
| PromptBase / FlowGPT | Discovering and selling prompts | Marketplace for ready-made prompts; good for inspiration but quality varies |
Pro tips & power moves
- The “anti-examples” trick. Instead of just saying what you want, tell the AI one thing you don’t want. Example: “Write this in a confident tone — but avoid sounding like a LinkedIn influencer.” This single negative constraint cuts bad outputs in half.
- Chain your prompts, don’t cram them. When a task has multiple phases (outline → draft → edit), run them as separate messages rather than one giant prompt. The model holds context and the output quality at each step is dramatically higher because it’s not juggling ten instructions at once.
- Use the “teach it back” test. After getting an analysis or summary, send one more message: “Now explain the most important insight from what you just wrote to a smart 12-year-old.” If the explanation is muddled, the original output had a hidden problem. This is faster than reading 800 words twice.
- Save your best prompts as Custom Instructions. In ChatGPT, you can set persistent instructions that apply to every conversation. Put your P.A.C.E. preferences there (“I prefer bullet points over paragraphs for analysis outputs; my writing tone is X; I work in Y industry”). Every prompt you send after that starts 200 words ahead.
- Give it your own writing as a style sample. Paste two or three paragraphs you’ve written before and say “match this voice for everything you write in this session.” AI voice-matching is genuinely good now — it picks up cadence, vocabulary, and even punctuation habits. This is the fastest path from “AI-sounding” output to output you’d actually send.
Common mistakes to avoid
- Asking for opinion when you need structure. “What do you think about my marketing plan?” gets you a polite hedge. “List the 3 biggest logical gaps in this marketing plan and why each is a risk” gets you something actionable.
- Forgetting the audience. Every prompt should specify who the output is for. The same information written for a CEO and for a junior analyst should look completely different — AI won’t know which you need unless you say.
- Treating the first output as final. The best AI users iterate. Reply with “make this 20% shorter” or “the third paragraph buries the key point — move it to sentence 1” and the second or third output is almost always much better.
- Using vague tone words. “Professional,” “casual,” and “engaging” mean different things to different people — and to AI. Instead of “casual,” say “like a Slack message to a colleague you respect.” Concrete beats abstract every time.
- Not saving what works. The most expensive mistake is crafting a brilliant prompt, getting a brilliant output, and then losing the prompt. Keep a living document. Organise by job function. Your future self will thank you.
Get Premium AI Access — Paid in DZD
These prompts work best with a premium AI subscription. If you’re in Algeria or North Africa and can’t pay with an international card, Click DZ has you covered — 100% genuine licences, pay via CIB, EDAHABIA or BaridiMob, instant activation, and a 4.9/5 rating from 1,200+ customers. Save up to 60% vs official prices.
FAQ
Do these prompts work with free AI tools, or do I need a paid plan?
Most of them work on free tiers, including ChatGPT Free (GPT-4o mini) and Claude Free. However, the analysis and long-document prompts work significantly better on paid models that handle large context windows. If you’re using AI for work regularly, a paid subscription pays for itself in under a week of saved time.
How do I build a prompt library I’ll actually use?
Keep it somewhere you already live — Notion, a Google Doc, even a pinned note on your phone. Organise by job function (the same structure as this article works well). Every time a prompt produces an output you’re proud of, save the exact prompt that created it. Start with five prompts and add as you go. Don’t try to build the perfect library before using it.
What’s the fastest way to improve a prompt that isn’t working?
Run through P.A.C.E. Which element is missing? Nine times out of ten it’s C (context) or E (format). Add two sentences of context explaining the background, or add an explicit format instruction (“respond as a numbered list” or “keep this under 200 words”). If that still doesn’t work, try starting with “Act as a [specific role]” — role assignment is the single highest-leverage prompt technique.
Conclusion
A great prompt library isn’t about having hundreds of prompts — it’s about having the right 20, organised so you can find them in five seconds. Start with the prompts in this article. Save the ones that work. Adapt the ones that almost work. Over time you’ll build a system that fits the way you actually think and the work you actually do.
Two resources worth bookmarking alongside this: the ChatGPT vs Claude 2026 comparison will help you pick the right model for different categories of prompts, and our top AI models for 2026 guide covers every major tool available today with honest assessments of what each is actually best at.
Your action checklist
- ✅ Copy at least 5 prompts from this article into a personal prompt document right now
- ✅ Apply the P.A.C.E. framework to the next prompt you write — check that all 4 elements are present
- ✅ Set your writing style sample in ChatGPT’s Custom Instructions (paste 2–3 paragraphs of your own writing)
- ✅ Test the “anti-examples” trick on one prompt today — add one “do not” constraint and observe the difference
- ✅ Pick one repeating task at work (weekly update, status email, data summary) and build a dedicated prompt template for it
- ✅ If you need a premium AI subscription and you’re in Algeria, visit Click DZ to get genuine access paid in DZD

