Log inStart Free Trial
Meta Tech Provider

AI first response on WhatsApp, then a person takes over

An AI first response answers a WhatsApp customer within seconds, collects what a person needs, and then leaves the chat to that person. Build it as an automation with a Reply with AI step on the First message trigger, for one AI reply per new chat, or on all incoming messages with a low hourly AI cap and a long pause once a person replies from the app.

easyCoexistence lets your own AI send the first reply on the WhatsApp Business number you keep in the app, through Meta's Cloud API and Coexistence, and steps back when you answer from the phone.

First message

Fires when a chat has no stored state: a new contact, or a chat quiet for about 30 days.

1 to 60

AI replies per chat per hour an automation allows; 1 limits the AI to one reply an hour on a chat.

0 to 168 hours

How long an automation stays out of a chat after a reply from the phone; 4 by default.

Escalation paths

WhatsApp's Business Messaging Policy asks automated chats to offer prompt, clear and direct ways to reach a person.

What is an AI first response?

It is the first reply a customer gets, written by your AI, with a person handling the rest of the conversation. The customer is not left waiting: within seconds they get an answer to a simple question, or a request for the one detail the person will need, such as an order number or a date. The person then picks up a chat that is already moving. On a number connected through Coexistence, both sides use the same number: the AI through Meta's Cloud API, the person from the WhatsApp Business app on the phone, in one thread the customer sees as one conversation.

1Customer writes"Hi, do you fix cracked screens?"
2AI answers firstYes, and asks for the phone model
3You see it in the appSame chat, same number, on your phone
4You replyWith the price and a time slot
5AI steps backPaused on that chat for the hours you set

Which trigger should the first response use?

Two builds work, and they differ in how many AI replies a customer gets before you arrive. First message gives exactly one AI reply on a new chat and none after it, whoever answers next. The catch-all, Every message, keeps the AI answering until you reply from the phone, so pair it with a low "AI replies per chat per hour, at most" and a long "Pause on a chat after I answer it (hours)". Pick First message when you answer quickly and want the AI as a greeter; pick the capped catch-all when customers often wait an hour or more for you.

First message + Reply with AIEvery message + Reply with AI, capped
AI replies before youOne, on the chat's first messageUntil you reply, at most the hourly cap
Returning customersOnly after about 30 quiet daysYes, on any message outside the pause
After you replyNothing more from this automationQuiet for the pause hours, then the AI again
Suggested settingsDefaults are fineAI cap 1 to 3 per hour, pause 24 to 72 hours

What should the first reply say?

Three parts, in two or three lines: an answer if the Instructions contain it, the one or two details a person needs next, and who answers and when. The built-in rules already tell the model to offer a person when the customer asks or when it cannot help, and not to invent prices, stock, dates or policies. The first AI reply to a person in a day also opens by saying it is the business's automated assistant, so you do not need to write that line. Make the hand-off specific in your Instructions, with a name or team and real hours, because "someone will be in touch" reads as a dead end.

Instructions for a first-response Reply with AI step
You send the first reply for Lumen Phone Repair on WhatsApp.
A technician, Rafael, takes over the chat himself.

1. If the answer is below, give it in one line.
2. Ask for what Rafael needs: the phone model and what is broken.
3. End with: "Rafael will reply here, usually within an hour
   (Mon to Sat, 9:00 to 18:00)."

Do not quote repair prices or promise same-day repairs.
Services: screens, batteries, charging ports, water damage.
Address: [here]. Parking: [here].

How does the hand-off to a person happen?

By replying, or with a step. The customer's message and the AI's answer are already in the chat on your phone, and the app notifies you about the message as it did before the number was connected. When you, or a teammate on a linked device, send a reply from the app, Meta reports it to easyCoexistence as an echo and the automation pauses on that chat for the hours set under Advanced. On the capped catch-all build, set that pause to 24 hours or more, so the AI does not come back the next morning in the middle of your conversation. To make the AI stop on its own, add a Hand to a person step after it: it pauses the bot on that chat, 24 hours by default, and emails you. To alert a team that does not watch the phone, add Notify me or Call my webhook after the AI step, pointing at n8n or your system, which can post to a team channel and returns no reply.

How do you keep the AI brief until a person arrives?

With four settings on the automation, all in the editor and in create_automation:

  • AI replies per chat per hour, at most (settings.ai_max_per_hour): 1 to 3. Past it, runs are Skipped with "AI reply limit for this chat reached this hour".
  • Longest reply (tokens) (max_tokens): 200 to 300 keeps answers to a few lines; the default is 600.
  • A Wait of 1 minute before Reply with AI, so three lines sent in a row get one answer.
  • Earlier messages it reads (history): the default 10 is plenty for a first response; 0 makes the AI read only the newest message.

Can your AI build the first response for you?

Yes. With the easyCoexistence connector in ChatGPT or Claude, describe who takes over, when, and what the AI should ask. The assistant calls list_numbers and create_automation, then test_automation on a sample message. A test sends nothing and reads only the sample, so it shows the exact first reply a new customer would get. The capped catch-all build, as the assistant would send it:

create_automation: AI first response, then a person
{
  "connection_id": "…from list_numbers…",
  "name": "AI first response",
  "trigger": { "type": "message_received", "match": "any" },
  "steps": [
    { "type": "wait", "minutes": 1 },
    {
      "type": "ai_reply",
      "provider": "openai",
      "model": "<a model id from your key's list>",
      "prompt": "You send the first reply for Lumen Phone Repair. …",
      "history": 10,
      "max_tokens": 250
    }
  ],
  "settings": { "takeover_hours": 48, "ai_max_per_hour": 2 }
}

How to set it up

  1. Save your AI keyOn the Automations page, click AI keys, choose the provider, paste the key and click Test and save.
  2. Start from AI assistantClick New automation, then Use this on Blank. Set Which messages to Every message or First message, and add a Reply with AI step from the step menu.
  3. Write the first-response instructionsIn Reply with AI, pick the Model and write what to answer, what to ask and who takes over when.
  4. Set the hand-offOpen Advanced. On the catch-all build, set the pause to 24 hours or more and AI replies per chat per hour to 1, 2 or 3.
  5. Test, then turn it onSwitch it Off, click Save and Run test with a first question. Read the reply, then switch it On and click Save.

Common mistakes

  • Expecting First message on a chat you opened from the phone. Your message created the chat, so the customer's reply is not a first message and no AI first response goes out.
  • Letting the AI promise "a person will reply in 5 minutes" when nobody watches the phone. Put the real response time and hours in the Instructions.
  • Leaving the pause at 4 hours on the catch-all build. The AI returns that afternoon in a chat you are still handling.
  • Adding keyword automations without checking the first message. Keywords rank first, so a new customer who writes "price" gets the fixed text instead of the AI first response.
Doing this with EasyCoexistence

The AI runs on your key with Anthropic, OpenAI or OpenRouter, with no charge for AI or automations. US$ 9 down to US$ 0.75 per number per month, falling as you add numbers, with the first 3 days free.

easyCoexistence is a multichannel API and MCP server that connects AI agents such as Claude and ChatGPT to WhatsApp, Instagram, TikTok, Facebook and more. On WhatsApp it uses Coexistence: the business keeps using its number in the WhatsApp Business app while the AI works through Meta's official API.

Frequently asked questions

Can the first response go out only during opening hours?

Run the first response all day and give nights an After hours automation for returning customers: a trigger takes one match, and First message does not check the clock. A new customer at night still gets the first response, since First message ranks higher.

Will customers notice when a person takes over?

Only if you tell them. Your reply comes from the same number and business name as the AI's. A first line such as "Hi Ana, Rafael here" makes the change clear and matches the hand-off the AI announced.

Can the first response be a fixed text instead of AI?

Yes. Use Reply with text with {{first_name}} for a greeting that never changes, or put Reply with text before Reply with AI in the same automation so the greeting is fixed and the answer comes from your model.

What if my AI key stops working?

The automation pauses with "Paused: your {provider} key was refused" and you get one email. Customers get no AI reply until you save a key that works, after which it starts again by itself; answer waiting chats from the phone meanwhile.

Keep reading

Ready to get started?

Set up WhatsApp Coexistence in minutes, not months. The app keeps working on the phone.

Start Free Trial

Verified on