← Back to Resources
Prompt Engineering·9 min read·September 7, 2026·By Kelly Stratton

9 Secret Prompts for Better Claude Answers

Nine lines you paste before or after your request that turn a generic Claude answer into one built on your voice, your facts, and your customer. Each one ready to copy, with when to use it and what comes back.

Get the full guide as a PDFFree

Save it, print it, read it offline. It's yours.

Download PDF

Every generic answer you’ve ever gotten from Claude was one line away from a good one. You didn’t need a longer prompt or a course. You needed to hand Claude the thing it was guessing at: what you want, how you sound, who it’s for, what to leave out. These are the nine lines that do that. Paste them exactly as written, or keep them in a note on your phone and grab the one you need.

Number one does the most work, so it’s first and it gets the full walkthrough. The other eight each get the prompt, when to use it, and what you’ll see come back.

Use these to chat better with Claude
1. Ask me firstClaude asks you questions before it does anything. Use it for anything that depends on details about you.
2. Match this voicePaste something you wrote. Everything after sounds like you.
3. Three versionsShort, warm, direct. Pick one instead of editing one.
4. Find the weak spotClaude grades its own answer and fixes it.
5. Be the expertA job title and a standard to hold the work to.
6. Leave this outRules about what not to do. They land harder than rules about what to do.
7. Only these factsStops Claude filling gaps with things you never said.
8. Say not foundNo invented names, numbers, dates, or sources.
9. Write it for themWho's reading, and what state they're in.
Prompt 1

Ask Me First

Paste this at the top of your message, before the request itself:

Before you answer or do any work, ask me questions, one at a time. Keep going until you are 95% confident that you can complete the task the way that I ask.

Then write what you want under it. Claude comes back with a question instead of an answer. You answer it. It asks the next one. When it has what it needs, it does the work.

ClaudeExample output
Before you answer or do any work, ask me questions, one at a time. Keep going until you are 95% confident that you can complete the task the way that I ask. Write a welcome email for new clients of my bookkeeping business.

First question: what does a new client get from you in their first month? For example, a setup call, a cleanup of their books, monthly reports, or something else. I'll shape the email around what they should expect first.

From there it keeps going: what the client needs to send you to get started, whether the price is already settled, how you want to sound, what you sign off as. Four or five questions, then it writes.

ClaudeExample output
Setup call in week one, then I clean up the last three months. They send me bank statements, last year's return, and their invoicing login. $350 a month, first month billed after the call. Warm, short, sign off Kelly.

Hi Dana, welcome aboard. Here's what happens next. We'll start with a setup call in your first week, and after that I'll go back through the last three months and get everything tidy. Before the call, send me three things: your bank statements for those months, last year's tax return, and a login for whatever you invoice with. Your first month is $350, billed after the call. Any questions before then, reply here. Looking forward to it. Kelly

That’s the email the script talks about: your real steps, your real price, your sign-off. The old way gives you “We’re thrilled to have you on board!” and a paragraph about “our commitment to your success.” Same Claude. It stopped guessing.

When to use it. Anything Claude can’t know without you: an email to a real person, a plan for your week, a quote, a job description, anything with your prices or your customers in it. Skip it for quick factual questions, where it only slows you down.

Why it works. Claude has to fill every gap you leave. If you don’t say who the client is or what happens next, it fills those gaps with the most common version of a welcome email, which is the generic one you’ve seen a hundred times. The questions move the facts from your head into the chat before it commits. “One at a time” matters too. Hand someone six questions at once and they answer three of them halfway.

Make it permanent

Click your initials in the bottom-left corner of Claude, then Settings, then Profile, and find the box for your personal preferences. Paste this and save: "When I ask you to write something I'll send to another person, or to make something that depends on details about me, my business, my customers, or my prices, ask me questions first, one at a time, until you're 95% confident. For quick factual questions, answer directly." From then on it's on for the work that needs it and out of the way for everything else. If a task makes it ask too much, add: "No more than five questions, then make your best assumption and mark it in brackets."

Prompts 2 to 9

The Other Eight

2. Match this voice

Here's something I wrote that sounds like me: [paste an email, a post, or a paragraph]. Match this voice in everything you write for me in this chat.

When: the first message of any chat where Claude is writing something you’ll send. What comes back: your sentence length, your warmth, your sign-off, without you describing any of it. Claude matches a sample far better than it follows “friendly but professional.”

ClaudeExample output
Here's something I wrote that sounds like me: "Hey Tom, quick one. Same order as last time works, 250 of the large. Can you get them here by the 20th? Thanks a bunch. Kelly." Match this voice in everything you write for me in this chat. Now write a reminder to a client that her invoice is a week late.

Hey Dana, quick one. Invoice 118 came due last Friday and I haven't seen it land yet. Could you take a look when you get a sec? If it's already on its way, ignore me. Thanks a bunch. Kelly

3. Three versions

Give me three versions: one short, one warm, one direct. Label them.

When: any time you’d otherwise get one draft and start editing. What comes back: three labeled options, and one of them is usually right. Picking takes ten seconds. Editing one that’s almost right takes ten minutes. I use this one on every subject line.

ClaudeExample output
Give me three versions: one short, one warm, one direct. Label them. It's a subject line for an email telling clients my rates go up in October.

Short: Rate update for October
Warm: A small change coming in October, and a thank you
Direct: My rates are changing on October 1. Here's what that means for you.

4. Find the weak spot

What's the weakest part of that answer? Fix it and tell me what you changed.

When: after any answer that’s fine but not quite there. What comes back: Claude names the soft sentence, the missing detail, or the claim it can’t back up, and hands you the fixed version. It’s better at finding holes in a draft than at being perfect the first time.

ClaudeExample output
What's the weakest part of that answer? Fix it and tell me what you changed.

The weakest part is the second paragraph. It says the price change "reflects the value I provide," which is the kind of line people skim past and quietly resent. I replaced it with the one concrete reason you gave me: your software costs went up in July and you held the rate for two years. Everything else stayed.

5. Be the expert

You're an experienced [job title]. Review this the way you would for a paying client, and hold it to that standard.

When: contracts, pricing, a plan, anything where you’d pay a professional to look it over. What comes back: the questions a bookkeeper, an editor, or a lawyer would ask, and the things they’d flag. A role tells Claude which knowledge to lead with.

ClaudeExample output
You're an experienced bookkeeper. Review this the way you would for a paying client, and hold it to that standard. [pasted: a simple monthly budget]

Three things I'd flag before I signed off on this. One, there's no line for taxes set aside, and at your income that's roughly a quarter of what comes in. Two, software subscriptions are listed once but you named four tools, so I'd expect that number to be higher. Three, nothing here is marked as a one-time cost versus monthly, so the total looks steadier than it is. Want me to rebuild it with those three fixed?

6. Leave this out

No bullet points. No exclamation marks. Under 120 words. Don't open with a greeting.

When: anything that’s coming back too long, too chirpy, or too formatted. Change the numbers and the rules to fit. What comes back: the same content in the shape you asked for. Rules about what not to do land harder than rules about what to do.

ClaudeExample output
No bullet points. No exclamation marks. Under 120 words. Don't open with a greeting. Rewrite the appointment reminder.

Your appointment is tomorrow, Thursday, at 10am. If you need to move it, reply to this message and I'll find you another slot this week. Otherwise I'll see you then. Kelly

7. Only these facts

Here are all the facts: [paste them]. Use only these. If you need something that isn't here, ask me instead of filling it in.

When: you have the details and you want them used exactly: dates, prices, names, steps. What comes back: work built on what you gave it and nothing else, plus a question when a gap shows up. Seven is for when you have the facts. Eight, below, is for when you don’t.

ClaudeExample output
Here are all the facts: the workshop is Saturday October 4, 9am to noon, at the library on Main Street, $45, limit 12 people, bring a laptop. Use only these. If you need something that isn't here, ask me instead of filling it in. Write the announcement.

One question before I write it: is there a link or an email for signing up? You didn't include one and I don't want to invent it. If there isn't one yet, I'll write "reply to this email to grab a spot."

8. Say not found

If you're not sure something is true, say "not found" instead of guessing. Never make up a name, a number, a date, or a source.

When: research, summaries, anything you’ll repeat to someone else and can’t check yourself. What comes back: honest gaps. “Not found” three times is worth more than three plausible answers you’d have to go verify.

ClaudeExample output
If you're not sure something is true, say "not found" instead of guessing. Never make up a name, a number, a date, or a source. What are the rules for collecting sales tax on services in my state?

Which state? That changes the answer completely. And a heads-up before you rely on me for this one: I can explain the general pattern (most states don't tax services, some tax specific ones), but for the current rule in your state I'd say not found unless you let me look it up, and even then check it against your state's revenue site before you change how you invoice.

9. Write it for them

This is going to [who they are], and right now they're [their situation]. Write it for them, not for me.

When: every reply to a real person. “A customer who’s already annoyed” changes every word Claude picks. “A new client who’s nervous about handing over her books” changes them again. What comes back: a message that reads like it was written to that person on that day, because it was.

ClaudeExample output
This is going to a customer who paid for a session I had to cancel this morning, and right now she's annoyed and thinks I'm unreliable. Write it for her, not for me.

Hi Dana, I'm sorry about this morning. You cleared time for it and I let you down. I've moved you to Thursday at 10, and that session is on me. If Thursday doesn't work, tell me what does and I'll make it fit. Kelly

Three of them in one message

"You're an experienced bookkeeper. This is going to a new client who signed up yesterday and is nervous about handing over her books. Here's an email I sent last month that sounds like me: [paste]. Match that voice, under 150 words, no bullet points. Before you write, ask me anything you need to know, one question at a time." That's prompts 1, 2, 5, 6, and 9 in four sentences. What comes back needs almost no editing.

Fixes

When It Doesn't Work

  • Claude answers right away instead of asking. Prompt 1 is buried under a big paste. Move it to the top of the message. Or you’re deep in a long chat where earlier instructions are steering. Start a new one.
  • It asks questions you already answered. Your answers were spread over several messages. Reply with all the facts in one message: “Here’s everything: [facts]. That’s all you need, go ahead.”
  • The voice still isn’t yours. One sample isn’t enough. Paste three, and pick the ones that sound most like you talking, not your most formal writing.
  • It ignored a rule from prompt 6. Rules fade in a long chat. Repeat the line, or put it in your personal preferences so it’s on every time.
  • The saved preferences aren’t kicking in. They apply to chats started after you saved them. Open a fresh one.
  • The answers are better but Claude still doesn’t know your business. These nine fix how you ask. They don’t give Claude your files, your prices, or your history for good. That’s what a Project is for: a folder in Claude where your documents live and every chat starts already knowing them. Different guide, and the two work together.
Do this today

Take one thing you were going to write today anyway. Put prompt 1 above it, answer the questions, and read what comes back next to what you would have written. Then paste the permanent version into your settings so it's on for good. Tomorrow, add prompt 2 with one email you're proud of. Two prompts, two days, and every answer from here on starts from your details instead of everyone else's.

FAQ

Questions People Ask About Claude Prompts

Straight answers

How do I get Claude to ask me questions before answering?

Paste this at the top of your message, before the request: "Before you answer or do any work, ask me questions, one at a time. Keep going until you are 95% confident that you can complete the task the way that I ask." Claude asks, you answer, and after four or five questions it does the work.

What is the best prompt to make AI understand what I want?

The one that makes it ask instead of guess: tell Claude to ask you questions, one at a time, until it's 95% confident it can do the task the way you want. The questions move the facts from your head into the chat before Claude commits, and one at a time matters because six questions at once get half-answered.

How do you get better answers from Claude?

Hand Claude the thing it's guessing at: what you want, how you sound, who it's for, and what to leave out. Nine short lines do that, like pasting something you wrote and saying match this voice, asking for three versions labeled short, warm, and direct, or giving it a job title and a standard to hold the work to.

Why does AI give generic answers?

Because Claude has to fill every gap you leave, so if you don't say who the client is or what happens next, it fills those gaps with the most common version, the generic one you've seen a hundred times. Get the facts into the chat first, by answering its questions or pasting them in, and the same Claude stops guessing.

What are the best Claude prompting tips for beginners?

Start with two: put the ask-me-first line above anything that depends on details about you, and paste one email you're proud of with the words match this voice at the start of any chat where Claude is writing something you'll send. Then save the ask-me-first line in Settings, then Profile, so it's on for good.

How do I make Claude sound like me?

Paste something you wrote, an email, a post, or a paragraph, and say: match this voice in everything you write for me in this chat. Claude picks up your sentence length, your warmth, and your sign-off without you describing any of it, and if the voice still isn't yours, paste three samples that sound like you talking.

How do I stop Claude from making things up?

Add this line: "If you're not sure something is true, say 'not found' instead of guessing. Never make up a name, a number, a date, or a source." Use it for research, summaries, and anything you'll repeat to someone else, because "not found" three times is worth more than three plausible answers you'd have to verify.
> newsletter

One simple AI win in your inbox every Sunday. No jargon, no overwhelm, unsubscribe whenever you like.