AI Prompts for WhatsApp Business: 15 Replies That Sound Human (2026)

Nick Timms
Nick Timms, Co-founder
September 3, 2026·7 min read·verifiedReviewed by Duda Bardavid

15 AI prompts for WhatsApp Business replies: short, human, and fast, with the 24-hour window respected. Three run a connected WhatsApp inbox themselves.

Key takeaways

  • WhatsApp prompts are not email prompts cut short. The channel has its own physics: replies read on a phone in seconds, tone runs personal, and the 24-hour customer service window decides whether you can even send a free-form message.
  • The fastest quality gain is a length ceiling. Every prompt here caps the reply, because the AI failure mode on WhatsApp is writing email-shaped paragraphs into a chat bubble.
  • Prompts 1 to 12 work by copy-paste in any assistant. The last three run against a WhatsApp inbox connected through MCP: reading the real conversation, drafting in thread, and using approved templates when the window has closed.
  • The rule that holds everywhere in this library holds here: the AI drafts and organizes, a human sends. On WhatsApp, where messages land instantly and personally, that review step matters more, not less.
Table of contents

The short version: WhatsApp replies are their own writing problem, and email prompts make them worse. A good WhatsApp reply is two or three sentences, sounds like a person, answers on the first screen of a phone, and respects the channel's one hard rule: outside the 24-hour customer service window, WhatsApp Business only lets you open with a pre-approved template. The fifteen prompts below are built for those constraints. Twelve work by copy-paste in ChatGPT, Claude, or Gemini; the last three run against a WhatsApp inbox connected through MCP.

Where this page sits in the library: this is the WhatsApp page. For the setting-up of automatic replies inside the WhatsApp Business app itself, the WhatsApp auto-reply guide covers that ground with ready-made message samples; for what the API actually costs, the WhatsApp Business API pricing breakdown has the receipted numbers.

What makes WhatsApp replies different

Three things, and every prompt on this page encodes them. Length: a chat bubble is not a letter, and anything past four sentences reads as a wall on a phone. Register: customers write to businesses on WhatsApp the way they write to friends, and a formal reply lands colder there than the same words would in email. And the window: once 24 hours pass after the customer's last message, a business account cannot send free-form text at all, only an approved template, which changes what "follow up tomorrow" means operationally.

The prompts are grouped the way a conversation actually runs: opening a reply, holding the middle, and closing or reopening the thread.

Opening: the first reply (prompts 1-5)

1. The first response. Speed matters more on WhatsApp than anywhere; this makes fast also good.

Claude logoChatGPT logoGemini logoWorks in Claude, ChatGPT, or Gemini
A customer sent this on WhatsApp: [paste]. Draft a reply under 3 sentences: acknowledge the specific thing they asked, give the answer or the honest next step with a time, and sound like a helpful person, not a support system. No greeting longer than two words.

The kind of message it produces:

WhatsApp logoWhatsApp
Hi Ana! Yes, the blue one is back in stock. Want me to hold one for you until tomorrow evening?

2. The order-status answer. The most common question on any commerce WhatsApp, answered without the tracking-number runaround.

Claude logoChatGPT logoGemini logoWorks in Claude, ChatGPT, or Gemini
Customer asks where their order is: [paste their message and the order status]. Reply in 2 sentences with where it actually is and when it arrives. Offer the tracking link only after the answer, not instead of it.

3. The delay message. Bad news, phone-sized.

Claude logoChatGPT logoGemini logoWorks in Claude, ChatGPT, or Gemini
We are late on this customer's order: [paste context]. Draft a WhatsApp reply under 4 sentences: say it is late plainly, give the new date, offer one concrete make-good, no excuses about warehouses.

4. The refusal that keeps the customer. Saying no in a chat thread without the chill of policy language.

Claude logoChatGPT logoGemini logoWorks in Claude, ChatGPT, or Gemini
I have to say no to this request: [paste]. Draft a WhatsApp reply under 4 sentences that says no in the first sentence, explains why in plain words, and offers the nearest thing I can do. No "unfortunately, as per our policy."

5. The bilingual switch. WhatsApp customers write in whatever language they live in.

Claude logoChatGPT logoGemini logoWorks in Claude, ChatGPT, or Gemini
Reply to this message in the language the customer used: [paste]. Keep my meaning exactly: [state what you want to say]. Match the formality of how they wrote, and keep it under 3 sentences.

Holding the middle: tone and length repair (prompts 6-9)

6. The de-corporate pass. For the draft you wrote as an email before remembering the channel.

Claude logoChatGPT logoGemini logoWorks in Claude, ChatGPT, or Gemini
Rewrite this for WhatsApp: [paste draft]. Cut it to a third of the length, drop the letter structure, keep every fact and commitment. It should read like a competent person typing on their phone, not a template.

7. The energy match. A customer using emoji and exclamation marks should not get a reply that reads like a bank.

Claude logoChatGPT logoGemini logoWorks in Claude, ChatGPT, or Gemini
Match this customer's energy without copying their style: [paste their messages]. Rewrite my draft [paste] so the warmth level fits theirs. Do not add emoji unless they used them first, and never more than one.

8. The de-escalation bubble. Anger arrives faster and rawer on WhatsApp than email.

Claude logoChatGPT logoGemini logoWorks in Claude, ChatGPT, or Gemini
This customer is angry, in a fast back-and-forth: [paste the thread]. Draft the next message only, under 3 sentences: name their actual complaint so they know they were heard, give one concrete step happening today, and do not match their tone.

9. The long-thread summary. For the teammate joining a 60-message thread mid-conversation.

Claude logoChatGPT logoGemini logoWorks in Claude, ChatGPT, or Gemini
Summarize this WhatsApp thread in 5 lines for a teammate taking over: [paste]. What the customer wants, what we promised and when, what is still open, their current mood, and the one thing not to bring up again.

Drag AI

The inbox your team and your AI work in together

Shared inbox, live chat, and AI in Gmail, with an MCP server your AI tools can drive.

Start free trial
ChromeWebDesktopMobileAPIMCP

Closing and reopening: the window (prompts 10-12)

10. The clean close. Ending the thread so it stays ended.

Claude logoChatGPT logoGemini logoWorks in Claude, ChatGPT, or Gemini
This issue is resolved: [paste last messages]. Draft a closing message under 3 sentences: confirm what was fixed, tell them exactly how to reach us if it recurs, and leave the door open without inviting a reply for its own sake.

11. The 23rd-hour follow-up. When the window is about to close and the answer is not ready.

Claude logoChatGPT logoGemini logoWorks in Claude, ChatGPT, or Gemini
The customer's 24-hour window closes soon and I do not have their answer yet: [paste context]. Draft a short holding message I can send now, inside the window: honest about the wait, specific about when we will have it, and asking one question that keeps the conversation open if they reply.

12. The template draft. For reopening after the window, WhatsApp requires a pre-approved template; write it once, well.

Claude logoChatGPT logoGemini logoWorks in Claude, ChatGPT, or Gemini
Draft a WhatsApp message template for [use case, e.g. "order update after the service window"], under 300 characters, with placeholder variables in {{curly braces}} for the personalized parts. It must make sense to someone who last heard from us days ago, and read as useful information, not marketing.

The prompts that work the WhatsApp inbox themselves (prompts 13-15)

With WhatsApp connected to an assistant through MCP, the pasting disappears: the assistant reads the real conversation, drafts in thread, and uses the approved template list when the window has closed. These are written against Drag's MCP server, whose WhatsApp tools are the same ones the copy-paste prompts approximate. The connection guide covers setup in about ten minutes.

13. The live thread catch-up and draft.

Claude logoChatGPT logoGemini logoWorks in Claude, ChatGPT, or Gemini
Read Maria's WhatsApp conversation, summarize where it stands in 3 lines, and draft the next reply under 3 sentences for my review. Do not send anything.

What happens: get_whatsapp_conversation pulls the real thread, and the draft lands for review with the summary above it. The 60-message catch-up becomes one sentence of asking.

The result shape, from a connected inbox:

Drag logoReply, using Drag's MCP tools
Maria ordered the ceramic set on Aug 28, reported one broken cup Monday, and we promised a replacement by Friday. Still open: she asked yesterday whether shipping is on us (it is, per our returns policy). She is friendly but nudging. Draft ready in the thread: confirms free shipping, replacement ships Thursday, tracking to follow.

14. The window-aware nudge. The follow-up that knows which rule applies before it writes a word.

Claude logoChatGPT logoGemini logoWorks in Claude, ChatGPT, or Gemini
We owe replies in three WhatsApp conversations. For each: check whether the 24-hour window is still open. Draft a free-form reply where it is, and pick the best approved template with filled variables where it is not. Show me all three for review.

What happens: the assistant checks each conversation's timing, drafts free-form where allowed, and calls list_whatsapp_templates to propose the compliant option where not. The channel's one hard rule gets enforced by the tool rather than remembered by a human.

15. The morning WhatsApp sweep. The triage habit, on the channel where customers expect the fastest answer.

Claude logoChatGPT logoGemini logoWorks in Claude, ChatGPT, or Gemini
Look at our open WhatsApp conversations. Which are waiting on us, how long, and which sound frustrated? Give me the reply order with one line of reasoning each. Do not reply to anything yet.

What happens: search_threads and get_whatsapp_conversation read the real queue, and the ranking arrives with quoted evidence. The channel that punishes slow answers hardest gets answered first, in order.

Where the prompts stop and the system starts

Everything above works with any assistant and a clipboard, except the last three, which need a WhatsApp inbox an assistant can reach. For disclosure, that is our product: Drag runs WhatsApp and email in one shared inbox for teams, from $12 a seat with Drag AI from $18, and its MCP server (Pro plan) is what makes prompts 13 to 15 run. It is also, as far as our own market checks can tell, the only Gmail shared inbox with both WhatsApp and an MCP server, which is why the connected prompts on this page have no equivalent elsewhere.

FAQ

Can I use ChatGPT to write WhatsApp Business replies?

Yes, and it is good at the two things WhatsApp punishes: length and tone. The essential constraint to give it is a sentence cap, because the default failure is an email-shaped paragraph in a chat bubble. Paste the customer's actual messages so the register matches, and always review before sending; WhatsApp lands instantly and personally.

What is the WhatsApp Business 24-hour window?

After a customer's last message, a business has 24 hours in which it can reply with free-form text. Once the window closes, WhatsApp only allows pre-approved message templates until the customer writes again. Practically, it means "we will get back to you tomorrow" needs either a holding message sent inside the window or an approved template ready for outside it.

How is this different from WhatsApp auto-replies?

Auto-replies are canned messages the WhatsApp Business app sends on triggers, like greetings and away messages; our auto-reply guide covers those with samples. The prompts on this page are for the replies in between: real conversations where an AI drafts something specific to the thread and a human reviews and sends it.

The rest of the prompt library

This page is one of nine in the Drag prompt library. The customer service prompts handle the hard tickets in any channel, the support QA prompts review replies before they send, and the shared inbox prompts run the team queue these conversations live in.

Nick Timms

Nick Timms

Co-founder

Building Drag for nearly ten years: shared inboxes, boards, and now the AI and agent layer, all on Gmail, plus HeyHelp for the personal inbox. Writes the honest versions of the comparisons.

Drag AI

The inbox your team and your AI work in together

Shared inbox, boards, live chat, and WhatsApp with AI included, in Gmail and beyond, plus an MCP server your AI tools can drive.

7-day trial, no card required4.7 · 1,200+ reviews
Gmail extension
Web app
Desktop
iOS + Android
API
MCP server: works with Claude, ChatGPT, Gemini, Copilot, Cursor