Prompt lines to copy
The lines from the videos, ready to paste into any AI chat. A new line is added with each new video.
1. Tell the AI who is reading
When the AI writes for nobody, "simpler" and "shorter" do not fix it. One line that names the reader does.
The reader is my CEO. Two minutes. No jargon.
2. When you give the AI more than one job, number the steps
Write each ask as its own numbered line, so the AI works down the list and you can check every number against the answer.
Do these steps in order: 1. <first job> 2. <second job> 3. <third job> 4. List every change you made, one line per step. If you could not do a step, say which one and why.
3. For a big piece of work, ask the AI for the outline first
Ask for the outline, read it, fix it, and only then let the AI write each part, so one wrong line at the top does not spread through the whole result.
Give me the outline first: the goal in one line, then one line per section. Wait for my OK. Then write one section at a time.
4. Ask by the page number your PDF viewer shows
When you ask an AI about one page of a PDF, use the page number in your PDF viewer, not the number printed on the page. Covers and contents pages often have no printed number, so the two can differ.
Look at PDF page [number from my PDF viewer]. Quote the lines you use, then answer: [question]
5. Paste the long document first and ask your question last
When you paste a long document into an AI chat, put the document at the top and your question at the very bottom, then check the page the answer quotes.
<paste the full document here> Using only the document above, answer this question and name the page or section your answer comes from: <your question here>
6. The AI explained a loop that does not exist. One line stops that
An AI coding assistant can describe code in a file it never opened, and the guess sounds as sure as a real answer. Anthropic's prompting docs give a line that makes it read the files first.
Never speculate about code you have not opened. If I reference a specific file, read the file before answering. Start with the file I name, then follow what it calls.
7. Read your message like a stranger before you send it
The AI only knows what is in your message. Read it as someone with no context would, and add the facts they would ask for.
Before you answer: what would someone with no context need to know to do this well? Ask me those questions first.
8. Say "edit", not "suggest"
If you want the AI to change your text, ask it to edit and make the changes. "Can you suggest some changes?" can get you a list of ideas instead of a fixed draft.
Edit this and make the changes. Keep it under [length]. Return only the new version.
9. Say it plainly, not in capital letters
Writing IMPORTANT or NEVER in capital letters can push the AI too far. Ask the normal way and say what you want.
Write three warm sentences for my coworker's goodbye card. Mention the Friday donuts.
10. Give the AI numbers and a list, not "easy"
Words like easy, quick or cheap are a guess for the AI. Say the time, the budget and the list of what you have, and the answer fits your real limits.
Limits: [time, e.g. 20 minutes]. [budget or size, e.g. one pan]. Use only: [your list]. If something does not fit these limits, leave it out and tell me.
11. Claude follows a rule better when you say why
A rule in capital letters with no reason is the weak version. Add the reason after "so" or "because", and Claude can handle the cases you did not list.
Your response will be read aloud by a text-to-speech engine, so never use ellipses since the text-to-speech engine will not know how to pronounce them.
12. Show the AI what you want instead of describing it
Words like nice, modern and premium give an AI almost nothing to go on, so it guesses the average. Paste a screenshot, a link or a few examples you like, and say "like this, but for my product." Anthropic's prompting guide calls examples one of the most reliable ways to steer the output's format, tone and structure.
Here are examples of what I like. Make something like this, but for [my product / my case]. Keep the style; change the content.
13. Make the AI show which parts of its answer are guesses
An AI sounds just as sure when it is guessing. Ask it to tag each claim by where it came from, then list the claims you should check. Anthropic's guide on reducing hallucinations recommends letting the model say it does not know and checking each claim against a source.
For each claim, tell me if it comes from what I gave you or from your memory. If you are not sure, say so. Then list the claims I should check before I use this.
14. Write your rules once in the settings, not in every chat
If you type the same rules at the start of every new chat, put them in Claude's Instructions setting once. They apply to all your chats.
I'm vegetarian. I cook for four. Keep recipes under 30 minutes.
15. Make the AI argue against a purchase before you buy
Ask "is this a good buy?" and the AI tends to say yes. Ask it to argue against the purchase first, then for it, then pick one.
Argue against this purchase first. Then argue for it. Then pick one.
16. Tell the AI to ask before it sends anything
When an AI assistant is connected to your email, let it draft the replies, but make it ask you before it sends anything to another person.
Draft the replies, but ask me before you send anything to another person.
17. Ask for the exact quotes first, then the answer
When you paste a long document into an AI chat, ask it to quote the exact lines first and answer only from those quotes. The answer comes from your document, and you can check it.
First, quote the exact lines about [your question], with page numbers. Then answer only from those quotes. If no line covers it, say so.
18. Ask what it changed, in two lines
When the AI tidies up your plan or draft, ask it to end with two lines on what it changed or removed. A clean new version does not show you what went missing.
When you finish, tell me in two lines what you changed or removed.
19. If you want more than the basics, ask for it
A one-line ask gets a minimal answer. If you want extras, like a timeline, a shopping list or a backup plan, name them in the ask.
Go beyond the basics. Give me [extra 1], [extra 2] and [extra 3].
20. Ask the AI to check its own answer against a rule
Before the AI finishes, tell it what a right answer must satisfy and ask it to check its answer against that rule. Anthropic's prompting guide says this catches errors reliably, especially for math.
Before you finish, check your answer against this rule: [what must be true]. If it fails, fix it and show the corrected answer.
21. Ask the AI to check two sources and say where each answer came from
An AI can answer a fact question from memory and sound sure. Ask it to check two sources, name each one and give its date, so you can see where the answer came from.
Check two sources, and tell me where each answer came from, with the date.
22. Put a big plan in one project, so every new chat starts with the facts
Tired of pasting the same paragraph into every new chat? Put the facts in a project's instructions once, and every chat in that project starts with them.
Facts for every chat in this project: - What: <the plan, e.g. my sister's wedding> - Number of people: <e.g. 60 guests> - Budget: <e.g. $8,000 total> - Date: <e.g. June 14> - Already decided: <add each choice here> Use these facts in every answer. If a suggestion breaks one of them, say so.
23. Crop the photo to the part you are asking about
A photo of a whole sign, board or menu makes the AI hunt for your line in small print. Crop to the part you are asking about, keep the column names, and the answer gets easier to check.
This photo is cropped to the row I care about. Read only this row and tell me the value under each column name.
24. Give the AI a role in one sentence
Ask a plain question and you get advice that fits anyone. Start with one sentence that says who the AI is, and the answer goes to your exact case.
You're a car mechanic helping a friend. What would you check on this exact car?
25. This 10-word prompt keeps finding bugs in people's apps
Steve Ruiz posted a one-line prompt on X: find all of the implicit state machines in my app. People who ran it on their own code found real bugs. Here is the prompt, a longer version that gives you a ranked report, and what to do with the results.
find all of the implicit state machines in my app For each one: list the states, show the files and lines, find invalid states and race conditions, rank by bug risk. Do not change code.
26. When a long chat forgets your rules, ask for a handoff brief
In a long chat, an AI can start ignoring a rule you gave in the first message. The rule is fine; the chat is too long. Anthropic's docs say that as the token count grows, accuracy and recall degrade, and they call it context rot. Ask the AI for a short brief of your rules and decisions, then paste it into a new chat.
Write a handoff brief for a new chat: every rule I gave you, and what we decided so far. Keep it short and use a list.
27. Fix my grammar, keep my words
Ask an AI to fix the grammar and it may rewrite your voice too. Say exactly what to change, and ask for a list of every change.
Fix only spelling and grammar. Keep my words. List every change you made.
28. Keep an AI bug fix small: change only what the bug needs
Ask an AI coding tool to fix one bug and it may also rename functions, add files and tidy code around it. Two lines keep the fix small and still show you what else it noticed.
Only change what this bug needs. No cleanup or refactoring around it. List anything else you noticed, but do not change it.
29. Label each part when you paste two messages into the AI
When you paste someone else's message and your own notes into one chat, give each part a label and put the task on its own line, so the AI does not mix them up.
Their message: <paste it here> My plan: <paste your notes here> Task: write the reply from my plan.
30. Flip the roles: let Claude interview you before it writes anything
When you don't know exactly what you want yet, ask Claude to ask you questions first. The draft then starts from your answers, not a guess.
Interview me until you could write this. One question at a time.