10 prompts, sorted by the job they do. Sorted by the job they do. Each one comes from real work, cleaned up so you can use it today. They work in any AI tool. Copy, paste, run.
Free. No email required. Last updated August 15, 2026.
Content
Take the robot out of the draft
When to use it. Run every AI draft through this before a human reads it.
Prompt
Rewrite the draft below. Rules:
- Short sentences. One idea per sentence.
- Plain words. Aim for a reading level a smart 5th grader handles.
- Cut every "it is not X, it is Y" sentence. Say what it IS.
- No dashes. Rewrite the sentence instead.
- Cut these words: leverage, synergy, cutting-edge, revolutionary, game-changer, supercharge, seamless, elevate, "in today's landscape", "more than just", "it is important to note", "when it comes to", "in conclusion".
- Keep my facts exactly as written. Add no new claims.
Then list the 3 weakest sentences that survived, and why they are weak.
[PASTE YOUR DRAFT]
Why it works. Long sentences, borrowed business words, and the not-X-but-Y pattern are the loudest tells that a machine wrote it. This kills all three in one pass.
Review
Find the claims you cannot back
When to use it. Before any page, ad, deck, or email goes to legal or to a client.
Prompt
Read the copy below. Pull out every sentence that makes a factual claim. That includes performance numbers, capability statements, comparisons, and superlatives.
For each one, put it in exactly one bucket:
1. BACKED. Name the source I would open to prove it.
2. CONTRADICTED. Say what the source actually says instead.
3. NO SOURCE. There is nothing behind it.
Do not soften bucket 3. List those separately at the end under the heading "Delete or prove before this ships."
Add nothing. Do not rewrite anything yet.
[PASTE YOUR COPY]
Why it works. Rewording a claim you cannot back gives you the same problem in new words. Sorting first makes the empty ones impossible to miss.
Planning
Name the result before you start
When to use it. Use at the top of any new AI job, before you write a single instruction.
Prompt
I want to hand a repeating job to AI. Interview me before you do anything.
Ask me one question at a time, and wait for my answer:
1. What exact finished thing should this produce?
2. What must it know about my business, my customer, and our voice before it starts?
3. Who reviews the result, and what are they checking for?
4. What should make us stop and go back to doing it by hand?
5. What gets saved so the next run is better than this one?
When I have answered all five, write them back to me as a one page brief. If any answer was vague, say so plainly instead of filling the gap yourself.
Why it works. Most bad AI output traces back to a job nobody defined. Answering these five out loud usually fixes the output without touching the prompt.
Planning
Build the context pack
When to use it. Use once per team or per brand, then reuse it in every job.
Prompt
Help me write the briefing an AI needs before it touches our marketing.
Interview me one question at a time and keep your questions short:
- Who buys from us, in their words and not our marketing words?
- What are they actually trying to get done?
- What do they worry about right before they buy?
- What do we sound like, and what would we never say?
- What are we not allowed to claim?
- What does a good result look like, and what does a bad one look like?
Then write it up as a briefing document under those headings. Mark anything I was unsure about as OPEN so I can go find the answer.
Why it works. When you leave a gap, the model fills it with the average of the internet. That average is where generic comes from.
Review
Check the real thing, not the preview
When to use it. Before you approve anything you have only seen in a preview pane.
Prompt
I am about to approve this work: [DESCRIBE IT]
I have checked it by: [HOW YOU CHECKED]
Do two things.
1. Name the property a real person actually experiences. Write it as a sentence a non-marketer would understand, like "the words do not overlap on a phone" or "the link goes to the right page".
2. Tell me whether my check could pass while that property is false. If it could, give me a check that can only pass if the property is true.
Be blunt. If my check proves nothing, say so.
Why it works. A preview, a passing report, and a screenshot are all stand-ins for the experience. Stand-ins are where approved work goes wrong.
Planning
One-way door or two-way door
When to use it. Use when you are stuck deciding, or deciding too slowly.
Prompt
Here is the decision in front of me: [DESCRIBE IT]
Sort it for me.
1. Is this reversible or not? If we are wrong, what does it cost to undo, in time, money, and trust?
2. If it is reversible and cheap to undo, say so and tell me to just decide today.
3. If it is hard to undo, list what I should find out first, ranked, and name the one fact that would change the answer.
Do not recommend the safe option by default. Say which one you would pick and why.
Why it works. Most decisions are reversible and get treated like they are not. The slow ones should be the doors that only open one way.
Research
Find the question everyone skips
When to use it. Before writing anything you want people to actually read and share.
Prompt
Topic: [TOPIC]
Reader: [WHO IS READING AND WHAT THEY DO]
Before writing anything, research.
1. What do the top pieces on this topic all say? That is the baseline. I do not want to repeat it.
2. What question do they all skip or answer badly?
3. Give me 3 angles that answer the skipped question. One sentence each.
Stop there and wait for me to pick one. Do not draft yet.
Why it works. Most AI content is a remix of the top results, which is why it reads the same. The skipped question is where a piece worth reading lives.
Review
The outsider test
When to use it. Before you launch anything you have been staring at for a week.
Prompt
You have no context about my company, my product, or this project. Do not ask for any.
Here is the thing a stranger would see: [PASTE THE COPY, OR DESCRIBE THE PAGE]
Answer only these, in order:
1. What do you think this is, in one sentence?
2. Who is it for?
3. What am I being asked to do?
4. What confused you, in the order you hit it?
5. What would you guess this costs, and why?
Do not be generous. If something did not land, say it did not land.
Why it works. Everyone inside the project knows what it means, which is exactly why they cannot see what it says. A reader with no context is the cheapest positioning test you can run.
Review
Did they actually agree?
When to use it. When two or three AI tools give you the same answer and you feel reassured.
Prompt
I asked several AI tools the same question and they agreed. Before I trust that, check the agreement itself.
My question was: [PASTE IT]
The answers were: [PASTE THEM]
1. Did my question contain the answer? Quote the words that led them there.
2. Could these tools have disagreed, given how I framed it? If not, their agreement is worth nothing.
3. What did all of them skip?
4. Argue the strongest case against the shared answer.
Why it works. Agreement only counts as evidence when disagreement was possible. Ask three tools a leading question and you get your own assumption back three times.
Operations
The handoff note
When to use it. At the end of any long AI session, before you close the window.
Prompt
Write a handoff note for whoever picks this up next, including me tomorrow.
Cover, in this order:
1. What is finished and how someone could check it themselves.
2. What is started but not finished, and where it stopped.
3. What is blocked, and who it is waiting on by name.
4. The exact next step, written so someone could do it without asking me anything.
5. Anything we learned that should change how we do this next time.
Keep it short. No summary of the conversation.
Why it works. Work that only exists in a chat window disappears when the window closes. The note is what makes it a system instead of a session.
Go deeper
A prompt does one job once. A protocol does it every time.
A prompt helps when you remember to use it. A protocol is the rule your team follows without you in the room. That is where trust comes from.