What an AI Agent Should Never Do on Your WhatsApp Number
The limits an AI agent should never cross on your WhatsApp number: strangers, cold messages, ignored stop requests, and a takeover that keeps you in charge of every chat you step into.
Daniel RothUpdated 9 min read

On this page
- What should an AI agent never do on your WhatsApp number?
- Who is responsible for consent and opt-out?
- Can WhatsApp still limit your number when an agent answers for you?
- Does the agent reply to strangers by default?
- Can the agent start a conversation with someone who never wrote to you?
- What happens when you write in a chat the agent is handling?
- How do you stop the agent, in one chat or everywhere?
- What if a customer types STOP?
- What should still wait for your yes?
- How do you set these limits before you switch an agent on?
- FAQ
AI agent WhatsApp safety comes down to limits you set before the first message. One reply to a stranger, one first message to someone who never opted in, or one ignored "stop", and you are the one explaining it. WhatsApp itself can limit a number that people report.
Below: the limits, who owns consent, what WhatsApp can do to the number, how a takeover works, how to stop the agent, and the second-phone test to run before the first customer sees it.
What should an AI agent never do on your WhatsApp number?
An AI agent should never message someone who has not asked to hear from you, answer chats you did not choose, fight you for a chat you have stepped into, ignore a request to stop, or run bulk or irreversible actions without your yes. Five limits cover most of the damage a badly set up whatsapp ai agent can do.
- Never message someone first unless you set it up to.
- Never reply in a chat you did not name.
- Never fight you for a chat you have stepped into.
- Never treat a stop request as noise.
- Never send bulk or irreversible actions without your yes.
Who is responsible for consent and opt-out?
You are. Consent and opt-out are the sender's responsibility, and an AI agent on WhatsApp sends in your name. WhatsApp's business policy requires opt-in permission before you contact someone and respect for any request to block, discontinue or opt out. An agent changes neither rule.
The WhatsApp Business policy states the opt-in rule: "You may only contact people on WhatsApp if: (a) they have given you their mobile phone number or username; and (b) you have received opt-in permission from the recipient confirming that they wish to receive subsequent messages or calls from you."
The same policy covers the way out: "You must respect all requests (either on or off WhatsApp) by a person to block, discontinue, or otherwise opt out of communications from you via WhatsApp, including removing that person from your contacts list."
In practice: collect permission first, keep a record of each whatsapp opt-out request, and point the agent only at chats where both are true. How to collect permission: the opt-in collection guide.
Can WhatsApp still limit your number when an agent answers for you?
"No tool can promise that. WhatsApp decides, and it can limit numbers that people report or that message contacts who did not expect it. Blueticks sends from your own WhatsApp account. Send to people who expect to hear from you." That is the Blueticks pricing FAQ, word for word.
WhatsApp's own business policy describes the mechanism: "Users can block or report businesses. Our systems may limit the amount of messages a business can send or calls a business can initiate, based on this user feedback."
Blueticks runs on your own number, "through WhatsApp Web and the Blueticks extension" (pricing page), and the AI Agents guide states that Blueticks is not affiliated with WhatsApp or Meta. The variable is who you point it at: people who already wrote to you, or people who never asked.
Does the agent reply to strangers by default?
No. Selected chats (allow list) is the default: the agent answers only the chats you add. People writing for the first time get a reply only if you tick Respond to new incoming chats. With the default, the agent ignores any chat you have not added. Check both under Triggers before you turn anything on.
An unknown number writes to you, the agent answers, and now a conversation exists that you never agreed to. The other option, All chats except (block list), answers everyone except the chats you block; avoid it on day one.
Gotcha: a default you never read is still a default you accepted.
Can the agent start a conversation with someone who never wrote to you?
Only if you set it up to. On a schedule, it speaks first through a time trigger: "A time trigger wakes the agent on a schedule, for example to send you the day’s list every morning." You describe the task and pick the Chat where the result goes, so a first message is your decision. Its sending scope is the other guard: under Allow sending messages to, the default is the same chats it is triggered in.
The AI Agents guide adds Send result to (optional) for sending the result somewhere else.
Gotcha: a schedule pointed at people who never asked is a cold message with a timer on it. Check the Chat before you save.
What happens when you write in a chat the agent is handling?
Your message takes over the chat. Under Human takeover, Allow human takeover and Take over automatically when I write in the chat are both on from the start. Write in a chat yourself and the agent goes quiet in that chat. You speak, the agent steps back.
The guide's own words: "When you take over, the agent stays quiet in that chat until you hand it back: type continue in your operator chat, or open Conversations and release the chat. By default the agent also picks the chat up again by itself after 2 hours without a message from you."
Also during the build: "Pick an Operator chat to get the agent’s notices and questions."

Failure mode: the agent went quiet unexpectedly. You wrote in that chat. Type continue in your operator chat, release it on Conversations, or wait. By default the agent also picks the chat up again by itself after 2 hours without a message from you.
Worked handoffs: a lead-qualifying agent and a clinic or salon receptionist.
How do you stop the agent, in one chat or everywhere?
One chat: write in it yourself and the agent goes quiet there. Everywhere: a paused agent sends nothing. New agents are created paused and the agents list shows them as Paused; nothing goes out until you click Turn it on, and the card then reads Live on WhatsApp.
For one chat, write in it; the agent stays quiet there until you type continue in your operator chat or release the chat on Conversations. By default the agent also picks the chat up again by itself after 2 hours without a message from you. For longer, remove the chat from Selected chats too.
For everything, use the two levers the guide describes: a paused agent sends nothing, so keep a new agent paused until its limits are set; and the agent ignores any chat you have not added, so removing the chats from Selected chats leaves it nothing to answer.

Gotcha: set your operator chat before you go live, then run the takeover test once while you watch.
What if a customer types STOP?
Do not rely on the agent to read a customer's STOP as an opt-out. The AI Agents guide describes no STOP handling, so the request is yours to honour, and the WhatsApp Business policy says to respect it. Three steps, durable one first.
- Remove that chat from Selected chats. The agent ignores any chat you have not added.
- Write in the chat yourself so the agent goes quiet there now. A takeover alone is not an opt-out. By default the agent also picks the chat up again by itself after 2 hours without a message from you.
- Reply yourself and confirm the request is done.
Then record the opt-out. The policy says "including removing that person from your contacts list", so remove them from every list the agent could reach.
What should still wait for your yes?
Anything you cannot undo, anything that reaches more than one person, and anything that spends money should wait for your yes. Automate the routine replies in the chats you named and hold the rest for approval. That is human in the loop WhatsApp work: the agent asks, you answer, and nothing irreversible runs on a default.
The one live mechanism sits under Integrations (MCP): connect an app, then tick the tools the agent may use. Every tool starts off, so the agent can only use the tools you tick.
Three buckets: routine replies in named chats run unattended; irreversible or wide actions wait for approval; a new contact or chat gets a human look first.
Gotcha: an approval nobody answers is a delay, so the default is nothing. The approval loop covers how an approval ping and a yes work.
How do you set these limits before you switch an agent on?
Set five things first: one test chat under Selected chats (allow list), Respond to new incoming chats left off, sending at its default, Knowledge with Answer only from the knowledge above ticked, and an Operator chat. New agents are created paused, so nothing goes out while you set them.
In order, per the AI Agents guide:
- In the web app, click AI Agents in the left menu (Agents 2.0 if your account also has flow bots), then New agent.
- On "Build a new agent in 1 minute", describe in plain words what it answers, for whom, and what it must never do. Send it; Blueticks writes the instructions and the settings.
- Under Knowledge, paste the answers it may give, or click Upload to add a file. Tick Answer only from the knowledge above.
- Under Triggers, keep the default Selected chats (allow list) and add one chat you control: your own second phone. Leave Respond to new incoming chats unticked and sending at its default.
- Pick an Operator chat to get the agent's notices and questions.
- Click Turn it on; the card changes to Live on WhatsApp.
- Test 1, the named chat: from the second phone, send the question you hear most. Check the reply stays inside your Knowledge.
- Test 2, the stranger: from a number you did not add, send a message. Silence is the pass.
- Test 3, the takeover: write in the test chat from the business side. The agent goes quiet there. Hand it back by typing continue in your operator chat, or release the chat on Conversations. By default the agent also picks the chat up again by itself after 2 hours without a message from you.
- Only then tick Respond to new incoming chats if first-time writers should get a reply, and add the customer chats you want answered.
Run all three before a customer sees the agent. To run them on your own number, build the agent in AI Agents, keep it paused while you add one test chat, and switch it on only when the limits are set.

What breaks, and what to check first:
- A paused agent looks broken because it is doing exactly what you told it: nothing.
- You wrote in a chat yourself: continue in the operator chat, Conversations, or the 2-hour default.
- The app shows Low Credits and then Out of Credits.
- With WhatsApp Web Only, WhatsApp Web must be open in Chrome.
FAQ
Does an AI agent need my customers' consent before it messages them?
Yes. The WhatsApp Business policy allows contact only when the person gave you their number and you "have received opt-in permission from the recipient".
Can an AI agent on WhatsApp message people first?
On a schedule, through a time trigger you set, to the Chat you picked; by default it sends only to the same chats it is triggered in. An agent that replies waits for a message.
What happens if I write in a chat my agent is handling?
Your message takes over the chat and the agent goes quiet there. By default the agent also picks the chat up again by itself after 2 hours without a message from you.
Does a customer's STOP switch off the agent?
Do not count on it. Remove the chat from Selected chats, write in it yourself, and honour the request.
Will an AI agent get my WhatsApp number banned?
The Blueticks pricing FAQ answers this: "No tool can promise that. WhatsApp can limit numbers that get reported or that message people who do not expect it, whether you use Blueticks or send by hand."
Does the agent keep answering when my computer is off?
The pricing page states: "On Basic and Standard, Blueticks runs in your browser. On Pro, your agent, schedules and campaigns are cloud-hosted and keep working even when your computer is off." More in whether the agent answers when the computer is off.


