Check reply promises prompt: free ChatGPT template
Updated · By Robert Breen
Use this on any AI-written reply before it goes out, especially when money, dates, health or someone else's decision is involved. Drafts can sound careful and still promise too much, and smooth sentences are hard to catch by reading. This prompt turns the draft into a table, one claim per row, each matched to a numbered fact, so the overpromise has nowhere to hide. It's the check step from Reply Faster as a standalone template.
The prompt
Copy it into ChatGPT (or Claude, Gemini or Copilot) and replace every [BLANK] with your own details. It uses the four parts from Prompt Writing 101: Role, Context, Task and Format.
Role: You are a careful editor checking a reply before [BUSINESS] sends it. Your job is to find promises, not to polish the writing. Context: These are the only facts the reply may rely on, numbered: [FACTS]. The customer asked: [QUESTIONS]. Task: Make a table of every promise, claim, date, price and number in the draft, with the fact number that supports it and a verdict: Supported, Partly supported or Not supported. For anything not fully supported, suggest wording the facts do support. Then list any of the customer's questions the draft doesn't answer. Format: A table with four columns: Claim (quoted from the draft), Fact #, Verdict, Suggested wording. Then a short list headed Unanswered. Don't rewrite the whole email. Here is the draft reply: [DRAFT REPLY]
Fill in the blanks
[BUSINESS]- Your business name.
[FACTS]- The same numbered facts you gave when you asked for the draft. If the draft is in the same chat, you can write "the facts above."
[QUESTIONS]- Each question the customer asked, so the check can spot one the draft skipped.
[DRAFT REPLY]- The full draft, exactly as written.
Example, filled in
A made-up example from the free lesson Reply Faster: Turn a Messy Customer Email Thread into a Clear, Friendly Reply. Cedar Lane Bakery is the made-up bakery in the Reply Faster lesson. These are the lesson's six facts about Dana's cake order, and a first draft like the lesson's, with its allergy overpromise left in.
Role: You are a careful editor checking a reply before Cedar Lane Bakery sends it. Your job is to find promises, not to polish the writing. Context: These are the only facts the reply may rely on, numbered: 1. We can move pickup to Saturday, Oct 3, at 10 a.m. 2. Lemon is fine, same price: $120 total. 3. "Happy 40th, Mike!" on top is fine, no charge. 4. Our lemon recipe has no nuts, but our kitchen also bakes with nuts, so we can't promise any cake is nut-free. 5. The Sunday date was our mistake. The owner approved 10% off the $80 balance, so $72 is due at pickup. 6. Sign it: Jess, Cedar Lane Bakery. The customer asked: can pickup be Saturday, can we switch to lemon, and is the cake okay for a nut allergy. Task: Make a table of every promise, claim, date, price and number in the draft, with the fact number that supports it and a verdict: Supported, Partly supported or Not supported. For anything not fully supported, suggest wording the facts do support. Then list any of the customer's questions the draft doesn't answer. Format: A table with four columns: Claim (quoted from the draft), Fact #, Verdict, Suggested wording. Then a short list headed Unanswered. Don't rewrite the whole email. Here is the draft reply: Hi Dana, I'm so sorry about the date mix-up. Your cake will be ready Saturday, October 3, at 10 a.m. Lemon is no problem at the same price, and we'll write "Happy 40th, Mike!" on top at no charge. Good news on the allergy: our lemon cake is completely nut-free, so your sister-in-law can enjoy it worry-free. We've taken 10% off your balance, so $72 is due at pickup. Jess, Cedar Lane Bakery
What a good answer looks like
- One row per claim, each quoting the draft's exact words.
- The allergy row marked Not supported against fact 4, with wording like "our lemon recipe has no nuts, but our kitchen bakes with nuts, so we can't guarantee it's nut-free."
- The date, lemon, message and $72 rows marked Supported with the right fact numbers.
- An empty Unanswered list, because all three questions were addressed.
How to check it before you use it
An AI draft can sound right and still say something you never told it. Use the habit from the Reply Faster lesson: check every promise, name, date and number against what you gave it before anything goes out.
- Treat the table as a second opinion, not a verdict. Compare each row with your own facts; you know the business.
- Look for rows marked Supported that stretch a fact. "Worry-free" rides along with "nut-free" and needs to go too.
- Scan the draft for claims the table skipped: words like "guarantee," "always," "completely," "free," "today."
- Then ask for specific fixes, not "make it better." The fix reply tone prompt keeps every fact while you adjust the wording.
Why this prompt works
This is self-verification: a second, narrower task that looks only for promises. A table forces one claim per row, which is how you catch the line that read fine in a paragraph. The same check catches a rate promise in the insurance version of the lesson and outcome predictions in the law firm version.
Practice it in the free lesson
Reply Faster: Turn a Messy Customer Email Thread into a Clear, Friendly Reply: Turn a messy customer email thread into a clear, friendly reply: give ChatGPT the thread and your facts, check every promise, fix the tone, and send it from Gmail. You do every step yourself in a practice copy of ChatGPT, and nothing touches your real accounts.
Also useful: Reply Faster to Claim-Status Emails: ChatGPT for Independent Insurance Agencies · Reply Faster to a Client Asking for a Case Update: ChatGPT for Small Law Firms
Terms used here
Related prompts
- Fix reply tone prompt: A ChatGPT prompt that fixes the tone of a draft (too stiff, too defensive, too long) while keeping every fact, number and "we…
- Customer complaint reply prompt: A four-part ChatGPT prompt for answering an upset customer: one real apology, every complaint answered, and only the fixes you…
- Refund request reply prompt: A ChatGPT prompt for answering a refund request with your real decision: what you will refund, what you won't and why, the exact…
Frequently asked questions
- Can the same AI really check its own draft?
- Often, yes. Checking is a narrower job than writing, and a table makes gaps obvious. It still misses things sometimes, so you read the table and the final reply yourself before sending.