Ninety percent of prompt advice is theory. Here are ten prompts I paste into real work every week. Each one has a job: kill a task you do by hand, or fix an output that keeps coming back wrong.
No framework lectures. No "role, task, context" homework. Copy, paste, fill the brackets, get the result. If a prompt needs tuning, the notes after each one tell you which knob to turn.
1. Summarize for Busy Me
Summarize the following in 3 bullet points under 15 words each.
Then give me 1 actionable takeaway: [paste text]
Use this on articles, meeting transcripts, long emails. The word cap forces the model to compress instead of rephrase. If you get 4 bullets anyway, add "exactly 3" to the prompt. That constraint phrase fixes 90 percent of summary drift.
2. Explain Like I'm 5
Explain [topic] like I'm 5 years old. Use simple analogies,
avoid jargon, and give a real-world example I'd understand.
The best tool for learning a codebase you inherited. Feed it a function, a concept, or a system design. The analogy constraint is the secret: models explain best when forced to map the idea onto something physical.
3. Debug My Thinking
I'm trying to solve [problem]. Here's my approach: [describe].
What am I missing? What would you try instead?
This beats "review my plan" every time. The two-part structure forces the model to find gaps and propose alternatives, not agree with you. I use it before every design decision that touches money or infrastructure.
4. Code Review
Review this [language] code for: bugs, performance issues,
security vulnerabilities, and style improvements. Be specific:
[code block]
Run this on every pull request before a human looks at it. The category list matters: "be specific" stops the generic "looks good" response. Expect real findings. I catch race conditions and unhandled errors with this weekly.
5. Debug This Error
I'm getting [error message]. Here's the relevant code and what I've tried:
[code + context]. What's causing this and how do I fix it?
Include the stack trace and what you already tried. The "what I've tried" part is the difference maker. Without it you get the first Stack Overflow answer; with it you get a diagnosis that rules out your dead ends.
6. Architecture Decision
I'm designing [system]. The trade-off is [A] vs [B]. Help me decide by
analyzing: scalability, maintainability, cost, and time to implement.
Forces a weighted comparison instead of a gut take. The four fixed criteria keep the answer honest. When the model recommends one, ask it to reverse the decision and argue the other side. That second pass surfaces the real risks.
7. Automate This Process
I spend X hours/week on [task]. Design an automation workflow
using AI tools to reduce this by 80%. Include specific tools and steps.
This prompt pays for itself. It turns vague "I should automate that" feelings into a named toolchain and concrete steps. I estimate half the automations on this site started as a response to this exact prompt.
8. Email Triage
I have [X] unread emails. Help me draft a system to triage them:
urgent replies, read-and-file, newsletters to unsubscribe, delegate.
Output is a workflow, not an email. It gives you the rules for sorting before you open the inbox, which is the whole game. Pair it with a mail filter and you cut inbox time by an hour a day.
9. Generate Flashcards
Create 20 flashcards from this text. Question on one side,
answer on the other. Cover the key concepts:
[paste study material]
Study prep for anything with a syllabus. The 20-card cap keeps it focused on the load-bearing concepts. Import the output into Anki and you have a review deck in five minutes flat.
10. Socratic Method
Teach me [topic] using the Socratic method. Ask me questions
that lead me to discover the answers myself.
The most underrated prompt on this list. It turns the model from an answer machine into a tutor. Great for concepts you have to explain to other people later, because you learn them instead of recognizing them.
I used it on HTTP caching and it took twelve questions before I hit the answer. Twelve questions I would not have asked myself. That is the point. The model forces the gaps into the open.
What each prompt replaced
A prompt earns its place when it kills a recurring task. Here is what each one replaced for me:
- Summarize for Busy Me replaced reading long threads twice
- Explain Like I'm 5 replaced staring at docs until they made sense
- Debug My Thinking replaced asking a senior to review my plan
- Code Review replaced the first pass of manual PR review
- Debug This Error replaced tabbing between five Stack Overflow tabs
- Architecture Decision replaced a two-hour debate with a coworker
- Automate This Process replaced a weekend of manual data entry
- Email Triage replaced an hour of inbox anxiety every morning
- Generate Flashcards replaced typing study cards by hand
- Socratic Method replaced rereading chapters hoping they stick
Total recovered: several hours a week, mostly the boring kind.
Why copy-paste beats fresh-writing
You can write a prompt from scratch every time. It feels productive and it wastes your time. Fresh prompts come out vague because you write them at speed. The copy-paste versions win because they were tuned on real failures.
Every prompt above started as a rough draft. Each one broke in production, got a fix, and the fix got folded back into the text. That is why the phrasing looks specific: it is scar tissue. When you write a fresh prompt you are skipping the scar tissue and repeating the failure.
Treat your prompt file the way you treat code. Add to it, test it, and when a variant works better, update the original. Your future self will thank you, and future you is the one doing the actual work.
When to break the rules
These prompts are defaults, not laws. Three situations where you ignore them:
- The task is brand new and you have no sense of the output shape. Write a throwaway prompt and learn from the first response.
- You need a creative output, not a reliable one. The constraint-heavy style above optimizes for consistency, which is the enemy of surprise.
- The stakes are low and the question is quick. Do not load the summarize prompt for a yes-or-no question.
The rules exist because most tasks are recurring, shaped, and mildly important. For that majority, copy-paste wins. The exceptions are rare enough that you will feel them coming.
The rules that make these work
Three habits separate these prompts from the junk versions.
First, fill in the brackets completely. A prompt with a half-written context returns a half-useful answer. The extra thirty seconds you spend pasting real information is the entire difference.
Second, when the output misses, tell the model what to change. "Too long, half the length." "Too vague, add examples." One corrective round beats ten fresh prompts.
Third, keep your best prompts somewhere searchable. I keep mine in a markdown file that gets versioned. When a prompt works three times, it graduates to the file. Six months from now you will not remember the phrasing that fixed your exact problem.
That is the whole system. Ten prompts, three rules, zero theory. The models are good enough now that the bottleneck is your prompt library, not the model.
🌐 Free AI guides + tools: apexnexus.site - the free AI Nexus learning hub
☕ Support the free hub: buy us a coffee ☕
💬 Join the Discord (free community for AI automation learners): https://discord.gg/E5vuXxRtu9
