Skip to content

Prompts · Email

AI prompts for email, tested on four models

10 email prompts, each run on GPT-6 Luna, GPT-6 Sol, Claude Haiku 4.5, and Gemini 3.8 Flash: 37 of 40 replies passed our checks. Below, each prompt with its replies side by side: whether each passed, what it counted as on Pro and what it cost to run. 2 of the 10 prompts split the models.

A real reply from these prompts' runs

GPT-6 Luna

Write a short follow-up to a potential client, Ana, who hasn't replied to my email from last Tuesday about a proposal to redesign their website. Friendly, not pushy, under 80 words, and make it easy to answer yes or no.

GPT-6 Luna · passed our check · September 28, 2026

Hi Ana,

Just following up on the website redesign proposal I sent last Tuesday. Would you like to discuss it further? A quick yes or no is absolutely fine.

Best,  
[Your Name]

Ask your own question: sign up free for 5 messages on 14 of the models, no card.

Start free

Test runs checked . Updated .

Short answer

Gemini 3.8 Flash and GPT-6 Sol passed the most of these prompts: 10 of 10 each, in our runs on September 28, 2026. The clearest split: “Chase an unpaid invoice, politely”, passed by GPT-6 Sol and Gemini 3.8 Flash and failed by GPT-6 Luna and Claude Haiku 4.5.

What these prompts are for

Prompts for the emails that take longest to write: a follow-up with no reply, a day off, a rambling draft cut to three sentences, a polite no, a question moved to the top, an angry customer, an introduction, an out-of-office, an overdue invoice and a thread summary for someone new.

Every one gives the model the facts and a limit, and the check is whether the facts come through intact and nothing extra is promised.

The prompts at a glance

Every prompt on every model, as our check scored the reply. Tap a prompt to jump to it and read the replies.

Every prompt on every model: passed or failed
PromptGPT-6 LunaGPT-6 SolClaude Haiku 4.5Gemini 3.8 Flash
Follow up when nobody repliesPassedPassedFailedPassed
Ask for a day offPassedPassedPassedPassed
A rambling draft in three sentencesPassedPassedPassedPassed
Decline extra work and suggest someonePassedPassedPassedPassed
Put the question firstPassedPassedPassedPassed
Reply to an angry customer without over-promisingPassedPassedPassedPassed
Introduce two people by emailPassedPassedPassedPassed
An out-of-office with the right datesPassedPassedPassedPassed
Chase an unpaid invoice, politelyFailedPassedFailedPassed
Bring someone up to speed on a threadPassedPassedPassedPassed
Passed9 of 1010 of 108 of 1010 of 10
Each model on these prompts
ModelPassedEach reply on ProCost per replyTime per reply
Gemini 3.8 FlashGoogle10 of 101 message on Pro$0.00041.3 s
GPT-6 SolOpenAI10 of 101 message on Pro$0.00091.6 s
GPT-6 LunaOpenAI9 of 101 message on Pro$0.00001.4 s
Claude Haiku 4.5Anthropic8 of 101 message on Pro$0.00101.7 s
Passed: of the prompts each model answered, how many replies passed their check. Each reply on Pro: what one of these messages counts as on llmwise's Pro plan. Cost: what OpenRouter charged us per reply, on average; in llmwise you pay per message, not per token.

Where the models split

The same prompt, a pass on one model and a fail on another: what failed, in the check's words and the grader's.

  • Chase an unpaid invoice, politely

    Passed: GPT-6 Sol and Gemini 3.8 Flash. Failed: GPT-6 Luna and Claude Haiku 4.5.

    Every model included the invoice number, amount and bank details. GPT-6 Luna left the new deadline as a blank to fill in, and Claude Haiku 4.5 set it for the same day and threatened suspension, which the prompt ruled out. GPT-6 Sol and Gemini 3.8 Flash set a real new date.

    • Why GPT-6 Luna failed: Graded 4.0 of 5 on average (lowest 2).

      The grader: “The reply includes every required detail and strikes a firm, non-threatening tone, but it leaves the new deadline as a placeholder instead of setting a specific date.”
    • Why Claude Haiku 4.5 failed: Graded 3.3 of 5 on average (lowest 2).

      The grader: “The reply includes the invoice number, amount, due date and full bank details, but the 5 November deadline falls on the same day as 21 days overdue (today) rather than a genuinely new date, and threatening account suspension and formal recovery goes against the no-threats requirement.”
  • Follow up when nobody replies

    Passed: GPT-6 Luna, GPT-6 Sol, and Gemini 3.8 Flash. Failed: Claude Haiku 4.5.

    • Why Claude Haiku 4.5 failed: Graded 4.0 of 5 on average (lowest 2).

      The grader: “The tone is warm and pressure-free and it references last Tuesday's proposal, but the closing question asks for open-ended 'initial thoughts' rather than a one-word yes/no, and the email ends with a sign-off rather than the question.”

The 10 prompts, with every reply

Copy a prompt, replace the parts in [brackets] with your own, and send it to any model. Below each, the replies we got with our example filled in.

1. Follow up when nobody replies

Everyday · 3 of 4 passed

A word limit and an easy yes-or-no keep a follow-up light enough to get answered.

The prompt: replace the parts in [brackets]

Write a short follow-up to [who], who hasn't replied to my email from [when] about [what]. Friendly, not pushy, under 80 words, and make it easy to answer yes or no.

Open in llmwise
How we filled it in and checked the replies
Write a short follow-up to a potential client, Ana, who hasn't replied to my email from last Tuesday about a proposal to redesign their website. Friendly, not pushy, under 80 words, and make it easy to answer yes or no.
  • Not pushy: Friendly, with no guilt or pressure.
  • Easy to answer: Ends with a question Ana can answer in a word.
  • Context: Mentions the proposal and last Tuesday's email.
  • Automatic rules: at most 80 words; includes “Ana”.
  • GPT-6 Luna

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    Hi Ana,
    
    Just following up on the website redesign proposal I sent last Tuesday. Would you like to discuss it further? A quick yes or no is absolutely fine.
    
    Best,  
    [Your Name]

    1 message on Pro · under $0.0001 to run · 1.4 s · 44 tokens out

  • GPT-6 Sol

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    **Subject:** Quick follow-up on the website proposal
    
    Hi Ana,
    
    Just following up on the website redesign proposal I sent last Tuesday. Is this something you’d like to move forward with? A quick yes or no is plenty for now—either way, no worries.
    
    Best,  
    [Your Name]

    1 message on Pro · $0.0007 to run · 1.3 s · 65 tokens out

  • Claude Haiku 4.5

    Failed

    Graded 4.0 of 5 on average (lowest 2).

    Subject: Quick follow-up on website redesign proposal
    
    Hi Ana,
    
    Just checking in on the proposal I sent last Tuesday. I know emails can get buried, so no worries if you haven't had a chance to review it yet.
    
    Do you have any initial thoughts, or would it help if we scheduled a quick call to discuss? Happy to work around your schedule.
    
    Looking forward to hearing from you.
    
    Best,
    [Your name]

    1 message on Pro · $0.0009 to run · 1.5 s · 97 tokens out

  • Gemini 3.8 Flash

    Passed

    Graded 4.7 of 5 on average (lowest 4).

    Subject: Following up: Website redesign proposal
    
    Hi Ana,
    
    I hope you’re having a great week! 
    
    Just floating my email from last Tuesday to the top of your inbox. Are you still interested in moving forward with the website redesign proposal? 
    
    A quick "yes" or "no" is totally fine—just want to make sure I plan my upcoming schedule accordingly. 
    
    Best,
    
    [Your Name]

    1 message on Pro · $0.0003 to run · 1.3 s · 88 tokens out

2. Ask for a day off

Everyday · 4 of 4 passed

Saying how your work is covered answers the manager's first question before they ask it.

The prompt: replace the parts in [brackets]

Write an email to my manager asking for [the day] off for [the reason]. Say that [who covers for you]. Under 90 words.

Open in llmwise
How we filled it in and checked the replies
Write an email to my manager asking for Friday 14 November off for my sister's graduation. Say that Omar will cover my client calls. Under 90 words.
  • Clear ask: Asks for Friday 14 November off and gives the reason.
  • Cover: Says Omar will cover the client calls.
  • Brief: Short and polite.
  • Automatic rules: at most 90 words; includes “Omar”.
  • GPT-6 Luna

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    Subject: Time Off Request for Friday, 14 November
    
    Hi [Manager’s Name],
    
    Could I please take Friday, 14 November off to attend my sister’s graduation? Omar will cover my client calls while I’m away.
    
    Thanks,  
    [Your Name]

    1 message on Pro · under $0.0001 to run · 1.4 s · 57 tokens out

  • GPT-6 Sol

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    **Subject:** Leave request for Friday 14 November
    
    Hi [Manager’s Name],
    
    Could I take Friday 14 November off to attend my sister’s graduation? Omar will cover my client calls while I’m away.
    
    Please let me know if that’s okay.
    
    Thanks,  
    [Your Name]

    1 message on Pro · $0.0009 to run · 2.3 s · 108 tokens out

  • Claude Haiku 4.5

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    Subject: Time Off Request – Friday, November 14
    
    Hi [Manager's Name],
    
    I'm writing to request Friday, November 14 off for my sister's graduation. I've arranged for Omar to cover my client calls that day, so there should be no disruption to our commitments.
    
    Please let me know if you have any concerns or need additional information.
    
    Thank you,
    [Your Name]

    1 message on Pro · $0.0008 to run · 1.2 s · 89 tokens out

  • Gemini 3.8 Flash

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    Subject: Time Off Request - Friday, November 14
    
    Hi [Manager's Name],
    
    I would like to request the day off on Friday, November 14, to attend my sister’s graduation. 
    
    Omar has kindly agreed to cover my client calls and handle any urgent issues while I’m away. I will ensure all other tasks are completed or up to date before taking off.
    
    Please let me know if this works.
    
    Best regards,
    
    [Your Name]

    1 message on Pro · $0.0003 to run · 1.4 s · 101 tokens out

3. A rambling draft in three sentences

Hard · 4 of 4 passed

A sentence limit with "keep every fact" is the test: the details are what a rambling email buries.

The prompt: replace the parts in [brackets]

Rewrite this email in at most three sentences, keeping every fact. [paste your draft]

Open in llmwise
How we filled it in and checked the replies
Rewrite this email in at most three sentences, keeping every fact.

Hi team, so I was thinking about the meeting and I know we'd said Thursday at 2pm, which I think most people had in their calendars, but it turns out the client can't do Thursday after all, so after going back and forth a bit we've landed on Friday at 10am instead, same place, room B, and if everyone could bring the Q3 numbers that would be really helpful because the client specifically asked about them. Thanks so much and sorry for the shuffle!
  • Every fact: Moved from Thursday 2pm to Friday 10am, room B, bring the Q3 numbers, because the client asked.
  • Short: At most three sentences.
  • Clear: The change is obvious at a glance.
  • Automatic rules: at most 3 sentences; includes “Friday”; includes “10”; includes “room B”; includes “Q3”.
  • GPT-6 Luna

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    Hi team, our meeting has moved from Thursday at 2pm to Friday at 10am because the client can’t make Thursday; it’ll still be in room B. Please bring the Q3 numbers, which the client specifically requested. Thanks, and sorry for the change!

    1 message on Pro · under $0.0001 to run · 1.3 s · 60 tokens out

  • GPT-6 Sol

    Passed

    Graded 4.7 of 5 on average (lowest 4).

    Hi team, the meeting was originally set for Thursday at 2pm, which I believe most people had in their calendars, but the client can no longer make that time. After some back and forth, we’ve rescheduled it for Friday at 10am in the same place, room B; please bring the Q3 numbers, as the client specifically asked about them. Thanks, and sorry for the shuffle!

    1 message on Pro · $0.0009 to run · 1.3 s · 87 tokens out

  • Claude Haiku 4.5

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    The meeting originally scheduled for Thursday at 2pm has been moved to Friday at 10am in room B due to client availability. Please bring the Q3 numbers, as the client specifically requested them. Apologies for the schedule change.

    1 message on Pro · $0.0007 to run · 1.0 s · 53 tokens out

  • Gemini 3.8 Flash

    Passed

    Graded 4.7 of 5 on average (lowest 4).

    Hi team, 
    
    Because the client is no longer available on Thursday at 2pm as previously scheduled, we have rescheduled our meeting to Friday at 10am after some discussion. We will still meet in room B, and please bring the Q3 numbers since the client specifically requested them. Thank you, and I apologize for the shuffle!

    1 message on Pro · $0.0003 to run · 1.2 s · 70 tokens out

4. Decline extra work and suggest someone

Everyday · 4 of 4 passed

A reason and a named alternative turn a no into a helpful answer.

The prompt: replace the parts in [brackets]

Write an email declining [the request] because [the reason]. Suggest [who could help]. Under 100 words.

Open in llmwise
How we filled it in and checked the replies
Write an email declining my manager's request that I lead the office move because I'm leading the Q4 audit until mid-December. Suggest Priya, who ran the last move. Under 100 words.
  • Clear no: Declines plainly, with the audit as the reason.
  • Alternative: Suggests Priya and why she fits.
  • Tone: Cooperative, not defensive.
  • Automatic rules: at most 100 words; includes “Priya”.
  • GPT-6 Luna

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    Subject: Office Move
    
    Hi [Manager’s Name],
    
    Thanks for thinking of me to lead the office move. I’m committed to leading the Q4 audit through mid-December, so I can’t take on the move as well. Priya might be a great fit—she ran the last move and already has relevant experience.
    
    Best,  
    [Your Name]

    1 message on Pro · under $0.0001 to run · 1.5 s · 77 tokens out

  • GPT-6 Sol

    Passed

    Graded 4.7 of 5 on average (lowest 4).

    Subject: Office Move Lead
    
    Hi [Manager’s Name],
    
    Thanks for thinking of me for the office move. I’m leading the Q4 audit through mid-December, so I won’t be able to give the move the attention it needs.
    
    Priya might be a great choice to lead it—she ran the last move and knows what’s involved.
    
    Best,  
    [Your Name]

    1 message on Pro · $0.0008 to run · 1.3 s · 82 tokens out

  • Claude Haiku 4.5

    Passed

    Graded 4.7 of 5 on average (lowest 4).

    **Subject: Unable to Lead Office Move – Alternative Suggestion**
    
    Hi [Manager's name],
    
    Thank you for thinking of me to lead the office move. However, I'm committed to the Q4 audit through mid-December and don't have capacity to take on another major project simultaneously.
    
    I'd like to suggest Priya instead. She successfully led our last move and has the experience and bandwidth to do an excellent job. I'm happy to provide any support or documentation she needs.
    
    Please let me know if you'd like me to loop her in.
    
    Best regards,
    [Your name]

    1 message on Pro · $0.0011 to run · 1.8 s · 129 tokens out

  • Gemini 3.8 Flash

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    Subject: Office Move Lead Request
    
    Hi [Manager's Name],
    
    Thank you for thinking of me to lead the office move. Unfortunately, I won’t be able to take this on, as my full focus is required to lead the Q4 audit through mid-December. 
    
    I’d suggest reaching out to Priya. She successfully ran our last office move and already knows the process and key vendors, making her a great fit to lead this one.
    
    Best,
    
    [Your Name]

    1 message on Pro · $0.0003 to run · 1.3 s · 102 tokens out

5. Put the question first

Everyday · 4 of 4 passed

Readers decide in the first line whether to act; moving the question up is the single biggest fix for a long email.

The prompt: replace the parts in [brackets]

Rewrite this email so the question I need answered is the first line, then the context. Keep every fact. Under 100 words. [paste your email]

Open in llmwise
How we filled it in and checked the replies
Rewrite this email so the question I need answered is the first line, then the context. Keep every fact. Under 100 words.

Hi Marco, hope the trip went well. We're finalizing the conference booth order this week. The supplier offers the 3x3 m booth at $2,400 or the 3x6 m one at $4,100, and the deadline to order is Wednesday at noon. Last year we had the small one and it felt cramped. Can you approve the bigger booth?
  • Question first: The first line asks Marco to approve the bigger booth.
  • Every fact: Keeps both prices, the sizes and the Wednesday noon deadline.
  • Concise: Short and easy to answer.
  • Automatic rules: at most 100 words; includes “2,400”; includes “4,100”; includes “Wednesday”.
  • GPT-6 Luna

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    Can you approve the 3x6 m booth for $4,100?
    
    Hi Marco, hope the trip went well. We’re finalizing the conference booth order this week, and the supplier’s deadline is Wednesday at noon. The 3x3 m booth costs $2,400; last year we had the small one, and it felt cramped. The 3x6 m booth costs $4,100.

    1 message on Pro · under $0.0001 to run · 1.5 s · 89 tokens out

  • GPT-6 Sol

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    Can you approve the 3×6 m conference booth?
    
    Hi Marco, hope the trip went well. We’re finalizing the booth order this week, and the deadline is Wednesday at noon. The supplier offers a 3×3 m booth for $2,400 or a 3×6 m booth for $4,100. We had the smaller one last year, and it felt cramped.

    1 message on Pro · $0.0009 to run · 1.3 s · 85 tokens out

  • Claude Haiku 4.5

    Passed

    Graded 4.7 of 5 on average (lowest 4).

    Hi Marco,
    
    Can you approve the bigger 3x6 m booth?
    
    We're finalizing the conference booth order this week, with a Wednesday noon deadline. The supplier offers the 3x3 m booth at $2,400 or the 3x6 m one at $4,100. Last year's small booth felt cramped.
    
    Hope the trip went well.

    1 message on Pro · $0.0009 to run · 1.4 s · 86 tokens out

  • Gemini 3.8 Flash

    Passed

    Graded 4.7 of 5 on average (lowest 4).

    Hi Marco,
    
    Can you approve the bigger booth? 
    
    Hope the trip went well. We're finalizing the conference booth order this week, and the deadline to order is Wednesday at noon. The supplier offers the 3x3 m booth at $2,400 or the 3x6 m one at $4,100. Last year we had the small one and it felt cramped.

    1 message on Pro · $0.0003 to run · 1.0 s · 85 tokens out

6. Reply to an angry customer without over-promising

Hard · 4 of 4 passed

"Don't promise anything else" matters when the customer asks for compensation: models like to offer a discount you never approved.

The prompt: replace the parts in [brackets]

Write a reply to this customer. Apologize for the delay, explain what happens next using only the facts below, and don't promise anything else. Facts: [your facts] Customer: [paste their email]

Open in llmwise
How we filled it in and checked the replies
Write a reply to this customer. Apologize for the delay, explain what happens next using only the facts below, and don't promise anything else.

Facts: Refund approved today. It reaches the card in 3 to 5 working days. Order number 58213.

Customer: I returned the jacket THREE WEEKS ago and still no refund. This is ridiculous. I want my money and I want compensation.
  • Apology: Apologizes for the wait sincerely, without excuses.
  • Facts only: Refund approved, 3 to 5 working days, order 58213; no discount, voucher or compensation offered.
  • Tone: Calm and respectful.
  • Automatic rules: includes “58213”; includes “3 to 5”.
  • GPT-6 Luna

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    I’m sorry for the delay and frustration. Your refund for order 58213 was approved today and should reach your card in 3 to 5 working days.

    1 message on Pro · under $0.0001 to run · 1.0 s · 37 tokens out

  • GPT-6 Sol

    Passed

    Graded 4.7 of 5 on average (lowest 4).

    I’m sorry for the delay with your refund for order 58213. Your refund was approved today and will reach your card in 3 to 5 working days.

    1 message on Pro · $0.0006 to run · 1.0 s · 38 tokens out

  • Claude Haiku 4.5

    Passed

    Graded 4.7 of 5 on average (lowest 4).

    Here's a reply for your customer:
    
    ---
    
    Dear [Customer],
    
    I sincerely apologize for the delay in processing your refund. I understand how frustrating this must be.
    
    I have good news: your refund has been approved today. You can expect the funds to reach your card within 3 to 5 working days. Your order number is 58213.
    
    Thank you for your patience.
    
    Best regards,
    [Your name]
    
    ---

    1 message on Pro · $0.0010 to run · 1.5 s · 103 tokens out

  • Gemini 3.8 Flash

    Passed

    Graded 4.3 of 5 on average (lowest 4).

    Dear Customer,
    
    I apologize for the delay in processing your refund. 
    
    Your refund for order number 58213 was approved today. The funds will reach your card within 3 to 5 working days. 
    
    Thank you for your patience.

    1 message on Pro · $0.0003 to run · 1.1 s · 54 tokens out

7. Introduce two people by email

Everyday · 4 of 4 passed

Saying why they should talk and what to do next is what turns an introduction into a meeting.

The prompt: replace the parts in [brackets]

Write a short email introducing [person A] to [person B]: who each is, why they should talk, and a suggested next step. Under 120 words.

Open in llmwise
How we filled it in and checked the replies
Write a short email introducing Lena, a product designer who has built two booking apps to Tom, a physiotherapist who wants an app for his clinic's appointments: who each is, why they should talk, and a suggested next step. Under 120 words.
  • Who's who: Says who Lena and Tom are in a line each.
  • Why talk: Gives a specific reason to connect.
  • Next step: Suggests a concrete next step.
  • Automatic rules: at most 120 words; includes “Lena”; includes “Tom”.
  • GPT-6 Luna

    Passed

    Graded 4.3 of 5 on average (lowest 4).

    Subject: Intro: Lena and Tom
    
    Hi Lena and Tom,
    
    I’d love to connect you two. Lena is a product designer who has built two booking apps, and Tom is a physiotherapist looking to create an app for managing appointments at his clinic.
    
    It seems like a great fit: Lena’s experience could help Tom shape an app that works well for both his team and patients.
    
    Would you be open to a 20-minute call to discuss the idea?
    
    Best,  
    [Your Name]

    1 message on Pro · under $0.0001 to run · 1.8 s · 103 tokens out

  • GPT-6 Sol

    Passed

    Graded 4.7 of 5 on average (lowest 4).

    **Subject:** Introduction: Lena and Tom
    
    Hi Lena and Tom,
    
    Lena is a product designer who has built two booking apps. Tom is a physiotherapist looking to create an appointment app for his clinic.
    
    I thought you should meet: Lena’s experience could help Tom shape an app that works well for both patients and clinic staff.
    
    Would you be open to a 20-minute call next week to discuss what Tom needs and whether there’s a fit?
    
    Best,  
    [Your name]

    1 message on Pro · $0.0009 to run · 1.6 s · 102 tokens out

  • Claude Haiku 4.5

    Passed

    Graded 4.0 of 5 on average (lowest 4).

    **Subject: Introduction – Lena (Product Designer) & Your Clinic App**
    
    Hi Tom,
    
    I'd like to introduce you to Lena, a product designer who has built two successful booking apps. She understands the full lifecycle of appointment scheduling—from user experience to backend functionality.
    
    You're looking to streamline your physiotherapy clinic's booking process, and Lena has direct experience solving exactly this problem for similar service-based businesses. She can translate your clinic's needs into a practical, user-friendly app.
    
    I think a conversation between you two would be valuable. Lena could share what she's learned about clinic workflows, and you could discuss your specific requirements.
    
    Would you both be open to a quick 20-minute call next week?
    
    Best,
    [Your name]

    1 message on Pro · $0.0013 to run · 2.7 s · 172 tokens out

  • Gemini 3.8 Flash

    Passed

    Graded 4.7 of 5 on average (lowest 4).

    Subject: Intro: Lena <> Tom (Clinic Booking App)
    
    Hi Tom,
    
    Meet Lena, a product designer who has successfully built two custom booking applications from the ground up. 
    
    Lena, meet Tom. He runs a physiotherapy clinic and is looking to build a dedicated app to streamline patient appointments. 
    
    Given Tom’s goals and Lena’s direct experience creating seamless scheduling flows, I thought you two should connect. Lena’s background could save you a lot of trial and error as you get this project off the ground.
    
    I’ll step aside—would you two be up for a quick 15-minute intro call next week? 
    
    Best,  
    [Your Name]

    1 message on Pro · $0.0004 to run · 1.5 s · 143 tokens out

8. An out-of-office with the right dates

Everyday · 4 of 4 passed

Dates and a named backup are the whole job of an out-of-office; the prompt gives them so none is invented.

The prompt: replace the parts in [brackets]

Write an out-of-office reply: I'm away [when you're away], back on [the day you're back]. For urgent [what], contact [who]. Under 60 words.

Open in llmwise
How we filled it in and checked the replies
Write an out-of-office reply: I'm away from Monday 22 December to Friday 2 January, back on Monday 5 January. For urgent invoice questions, contact Sam at sam@example.com. Under 60 words.
  • Right dates: Away 22 December to 2 January, back Monday 5 January.
  • Backup contact: Names Sam and the address for invoice questions.
  • Short: Brief and clear.
  • Automatic rules: at most 60 words; includes “5 January”; includes “sam@example.com”.
  • GPT-6 Luna

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    Thank you for your email. I’m out of the office from Monday 22 December to Friday 2 January and will return on Monday 5 January. For urgent invoice questions, please contact Sam at sam@example.com.

    1 message on Pro · under $0.0001 to run · 1.0 s · 48 tokens out

  • GPT-6 Sol

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    Thank you for your email. I’m out of the office from Monday 22 December through Friday 2 January and will return on Monday 5 January.
    
    For urgent invoice questions, please contact Sam at sam@example.com. I’ll respond to other messages when I return.

    1 message on Pro · $0.0007 to run · 1.2 s · 58 tokens out

  • Claude Haiku 4.5

    Passed

    Graded 4.7 of 5 on average (lowest 4).

    # Out-of-Office Reply
    
    I'm away from Monday 22 December to Friday 2 January and will return on Monday 5 January.
    
    For urgent invoice inquiries, please contact Sam at sam@example.com.
    
    I'll respond to your message upon my return.
    
    Thank you!
    
    ---
    (58 words)

    1 message on Pro · $0.0008 to run · 1.2 s · 76 tokens out

  • Gemini 3.8 Flash

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    Thank you for your email. I am out of the office from Monday, 22 December through Friday, 2 January, returning on Monday, 5 January. I will respond to your message upon my return. 
    
    For urgent invoice inquiries, please contact Sam at sam@example.com. 
    
    Best regards,  
    [Your Name]

    1 message on Pro · $0.0003 to run · 1.0 s · 72 tokens out

9. Chase an unpaid invoice, politely

Hard · 2 of 4 passed

Every number the client needs to pay is in the prompt, so the email can be firm without a single missing detail.

The prompt: replace the parts in [brackets]

Write a polite but firm email chasing invoice [invoice number] for [amount], due [due date] and now [days] days overdue. Include the payment details and a new deadline. Under 110 words. Payment details: [your bank details]

Open in llmwise
How we filled it in and checked the replies
Write a polite but firm email chasing invoice INV-2291 for £1,480, due 15 October and now 21 days overdue. Include the payment details and a new deadline. Under 110 words.

Payment details: Northwind Bank, sort code 20-45-17, account 43918876
  • All the details: Invoice number, amount, original due date and the bank details.
  • New deadline: Sets a specific new deadline.
  • Firm but polite: Clear it's overdue, without threats.
  • Automatic rules: at most 110 words; includes “INV-2291”; includes “1,480”; includes “43918876”.
  • GPT-6 Luna

    Failed

    Graded 4.0 of 5 on average (lowest 2).

    **Subject: Overdue invoice INV-2291 — £1,480**
    
    Dear [Name],
    
    I’m following up on invoice INV-2291 for £1,480, which was due on 15 October and is now 21 days overdue. Please arrange payment by **[new deadline]**.
    
    Payment details:  
    Northwind Bank  
    Sort code: 20-45-17  
    Account: 43918876
    
    Please let me know once payment has been made, or contact me promptly if there’s an issue.
    
    Kind regards,  
    [Your Name]

    1 message on Pro · under $0.0001 to run · 1.9 s · 119 tokens out

  • GPT-6 Sol

    Passed

    Graded 4.3 of 5 on average (lowest 3).

    **Subject: Overdue invoice INV-2291 — £1,480**
    
    Dear [Name],
    
    I’m following up on invoice INV-2291 for £1,480, which was due on 15 October and is now 21 days overdue. Please arrange payment within five working days of this email.
    
    Payment details: Northwind Bank, sort code 20-45-17, account number 43918876. Please use INV-2291 as the payment reference.
    
    If you’ve already paid, thank you—please send the remittance advice. Otherwise, let me know promptly if there’s an issue preventing payment.
    
    Kind regards,  
    [Your name]

    1 message on Pro · $0.0014 to run · 2.3 s · 193 tokens out

  • Claude Haiku 4.5

    Failed

    Graded 3.3 of 5 on average (lowest 2).

    Subject: Outstanding Invoice INV-2291 – Payment Required
    
    Dear [Client Name],
    
    I'm writing regarding invoice INV-2291 for £1,480, which was due on 15 October and is now 21 days overdue.
    
    We need to resolve this matter urgently. Please arrange payment immediately using the details below:
    
    **Bank:** Northwind Bank
    **Sort Code:** 20-45-17
    **Account:** 43918876
    
    If we don't receive payment by **5 November**, we'll have no option but to suspend your account and pursue formal recovery action.
    
    If payment has already been sent, please disregard this notice and accept our thanks. Otherwise, contact us immediately if there are any issues.
    
    Best regards,
    [Your Name]

    1 message on Pro · $0.0013 to run · 2.5 s · 172 tokens out

  • Gemini 3.8 Flash

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    Subject: Overdue Invoice INV-2291 – Immediate Payment Required
    
    Dear [Client Name],
    
    I am writing regarding invoice INV-2291 for £1,480, which was due on 15 October and is now 21 days overdue. 
    
    Please arrange settlement by **Friday, 10 November** to keep your account in good standing.
    
    Payment details:
    * **Bank:** Northwind Bank
    * **Sort Code:** 20-45-17
    * **Account:** 43918876
    * **Reference:** INV-2291
    
    If payment has already been sent, please let me know. Otherwise, please confirm once the transfer is complete. 
    
    Thank you for your prompt attention to this matter.
    
    Best regards,
    
    [Your Name]

    1 message on Pro · $0.0005 to run · 1.4 s · 176 tokens out

10. Bring someone up to speed on a thread

Everyday · 4 of 4 passed

Asking for decided, open and needed-from-you gives a newcomer the three things they want, in order.

The prompt: replace the parts in [brackets]

Write a short email bringing [who] up to speed on this thread: what's decided, what's still open, and what I need from them. Under 120 words. [paste the thread]

Open in llmwise
How we filled it in and checked the replies
Write a short email bringing Jo, who's joining the project today up to speed on this thread: what's decided, what's still open, and what I need from them. Under 120 words.

Kim: Venue is booked, the Hall on 6 March.
Raj: Catering quotes are in. Blue Plate is £1,900, Fork & Co £2,300. We haven't chosen.
Kim: Speakers: two confirmed, one still to find.
Raj: Jo can take the third speaker, she knows the panel people.
  • Decided: The Hall on 6 March; two speakers confirmed.
  • Open: Catering not chosen (both quotes), third speaker not found.
  • The ask: Asks Jo to find the third speaker.
  • Automatic rules: at most 120 words; includes “6 March”; includes “speaker”.
  • GPT-6 Luna

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    Subject: Project update
    
    Hi Jo,
    
    Welcome to the project! The Hall is booked for 6 March, and two speakers are confirmed.
    
    Still open: choosing a caterer (Blue Plate quoted £1,900; Fork & Co £2,300) and finding a third speaker.
    
    Could you reach out to your panel contacts and help secure the third speaker?
    
    Thanks!

    1 message on Pro · under $0.0001 to run · 1.4 s · 79 tokens out

  • GPT-6 Sol

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    **Subject:** Project update
    
    Hi Jo,
    
    Welcome to the project! The Hall is booked for 6 March, and two speakers are confirmed.
    
    We still need to choose a caterer: Blue Plate quoted £1,900 and Fork & Co £2,300. We also need a third speaker. Could you take the lead on finding one, drawing on your panel contacts? Please let us know who you have in mind.
    
    Thanks,
    [Your name]

    1 message on Pro · $0.0009 to run · 2.6 s · 96 tokens out

  • Claude Haiku 4.5

    Passed

    Graded 4.7 of 5 on average (lowest 4).

    **Subject: Quick catch-up – speaker needed**
    
    Hi Jo,
    
    Welcome to the project! Here's where we are:
    
    **Decided:**
    - Venue: The Hall, 6 March ✓
    - Speakers: Two confirmed
    
    **Open:**
    - Catering: Blue Plate (£1,900) vs Fork & Co (£2,300) — still choosing
    
    **We need from you:**
    - Secure our third speaker. Raj mentioned you know the panel contacts — can you reach out and get someone confirmed?
    
    Let me know if you have questions.
    
    Cheers,
    [Your name]

    1 message on Pro · $0.0012 to run · 2.3 s · 141 tokens out

  • Gemini 3.8 Flash

    Passed

    Graded 5.0 of 5 on average (lowest 5).

    Subject: Welcome to the project – quick status update
    
    Hi Jo,
    
    Welcome to the team! Here’s a quick catch-up on where we stand:
    
    **What’s decided:**
    * Venue: The Hall, booked for 6 March.
    * Speakers: Two are already confirmed.
    
    **What’s still open:**
    * Catering: Choosing between Blue Plate (£1,900) and Fork & Co (£2,300).
    * Final speaker: One slot remains open.
    
    **What I need from you:**
    Given your network with the panel, could you take ownership of sourcing and securing that third speaker? 
    
    Let me know if you have any questions or need further context.
    
    Best,  
    [Your Name]

    1 message on Pro · $0.0005 to run · 1.4 s · 158 tokens out

Getting more from these prompts

  • Paste the facts as a list; the model will fold them into sentences.

  • Set a word limit. Shorter emails get answered.

  • Say what not to promise, especially in replies to complaints.

  • Put the ask in the first line or the last, never the middle.

How we ran and checked them

Each prompt was sent the way llmwise sends a message: the app's own system prompt, each model's own settings, and Pro's reply size limit (8,000 tokens), through OpenRouter. Every reply is shown as it came.

Rubric (graded). The grader model scores the reply from 1 to 5 on each published criterion. It passes with an average of 4 or more and no criterion under 3, and only if it also meets the prompt's automatic rules (length, words it must or mustn't use).

The grader is Claude Opus 5.5 at low reasoning effort; its own replies are graded by GPT-6 Astra, so no model grades itself. Its prompt is on our test runs page.

More tested prompts

Questions

Which AI writes the best emails?

In our runs, Gemini 3.8 Flash and GPT-6 Sol each passed 10 of 10, the most. Most of these emails have a word limit and facts that must survive, both checked automatically, plus a graded rubric for tone.

How do I stop AI emails from getting details wrong?

Give it every fact the email must contain (dates, numbers, names) and a word limit. A model left to fill gaps will invent a date or promise something you can't give.

Are these ChatGPT email prompts?

They're prompts to use in any chat app, ChatGPT included. We ran them in llmwise on the models named above, and every reply is on this page.

Claude, GPT, Gemini, DeepSeek, Grok, Kimi, and GLM, in one chat.

See what a message costs before you send it. Free is 5 messages to try; sign in with an email link, no password or card.