Why WhatsApp bans numbers, and how AI agents stay online
WhatsApp bans numbers for spam-like behavior: cold messages, bursts, automation, unofficial apps and reports. What its own rules say and how AI agents stay online.
Key takeaways
- WhatsApp's own pages tie bans to spam, bulk or automated messaging, messaging people who never asked, unofficial apps and scraping, and it publishes no numeric thresholds.
- The most detailed official description of detection is a 6 February 2019 white paper: WhatsApp checks accounts at registration, while they message, and after reports and blocks.
- A restriction (no new chats, ends by itself) is not a ban, and a ban comes with an in-app review request unless it is final.
- Builders report bans after cold first messages and right after linking a third-party client, but nobody outside WhatsApp can confirm the cause of any single ban.
- No setup can promise that a number is never banned. The practical goal is to remove the avoidable triggers and have a recovery plan.
WhatsApp bans a number when its activity looks like spam or automation, or when people report or block it. Its own pages name bulk and automated messages, texting people who never asked, bursts from brand-new accounts, unofficial apps and scraping. Builders add cold first messages and the act of linking a third-party client. No setup makes a ban impossible, but most triggers can be avoided.
Two kinds of source appear below. Official means WhatsApp’s Help Center and Terms of Service, read on 19 October 2026. Reported means what builders describe in GitHub issues for Baileys, WAHA, Evolution API and whatsapp-web.js. Reports are anecdotes: nobody outside WhatsApp can check the setup or the cause.
What does WhatsApp say gets a number banned?
WhatsApp’s Help Center says it bans an account when it believes the activity breaks the Terms of Service, for example spam, scams or risk to other users’ safety (About account bans). The Help Center and the Terms (effective 4 January 2021) name these behaviors:
- Bulk, automated or auto-dialed messaging. The Terms rule out “bulk messaging, auto-messaging, auto-dialing” and creating accounts through unauthorized or automated means (Terms of Service).
- Messaging people who did not ask. The responsible-use page says to message only people who contacted you first or asked you to, to stop when a contact asks, and warns that accounts reported multiple times are banned (How to use WhatsApp responsibly).
- Unofficial apps. Linking an account to unofficial versions of WhatsApp violates the Terms, and the account can be temporarily or permanently banned or lose the ability to link devices (About unofficial apps). It names apps such as GB WhatsApp, not Baileys, WAHA, whatsapp-web.js or Evolution API, so whether a library counts is WhatsApp’s call. Plan as if it can.
- Scraping. Automated extraction of phone numbers, profile pictures or statuses (About temporarily banned accounts).
- Non-personal use. The Terms also rule out non-personal use unless WhatsApp has authorized it.
WhatsApp publishes no thresholds, neither messages per day nor number of reports. The Terms also let it suspend access at any time, for example if you break their “letter or spirit”.
How does WhatsApp detect abusive accounts?
WhatsApp’s most detailed public description is its white paper “Stopping Abuse: How WhatsApp Fights Bulk Messaging and Automated Behavior”, dated 6 February 2019 and still linked from the responsible-use page. It is old: read the numbers as history. It describes three stages.
| Stage | What the white paper says WhatsApp looks at |
|---|---|
| At registration | Basic account data, IP address and carrier. Whether a similar phone number was recently abused, and whether the network used to register has a history of suspicious behavior. About 20% of bans in its three-month sample happened at registration. |
| While messaging | Speed and volume. Its example: an account that registered five minutes earlier and tries to send 100 messages in 15 seconds is almost certainly abusive. A new account that messages dozens of people who do not have it in their contacts is judged against its history, since that could be spam or a person announcing a new number. It also looks for signs of automation, such as sending without ever showing a typing indicator. |
| After negative feedback | Reports and blocks. Anyone who gets a first message from an unknown number is offered “report” and “block” at once. A report of that first message is trusted more. A report against a number that did not start the chat gets extra checks for coordinated false reports. |
Two points matter for agents. WhatsApp says it weighs hundreds of factors together and bans obvious cases automatically. And in the three months before publication it banned over two million accounts a month for bulk or automated behavior, more than 75% of them with no recent user report. “Nobody complained” does not mean “safe”.
Restriction, temporary ban or ban: what is the difference?
A restriction limits new contacts and ends by itself, a temporary ban is tied to unofficial clients or scraping, and a ban locks the number out until a review says otherwise.
| State | What you see | What still works | How it ends |
|---|---|---|---|
| Restricted | “Your account is restricted right now” or “may be restricted soon” | Contacts and groups you already chatted or called with. No new chats or calls, no new groups, no adding group members, no linked devices. | By itself, when the time runs out. Tap New Chat and pick a contact you never chatted with to see the time left. |
| Temporarily banned | A “Temporarily banned” message, which WhatsApp ties to unofficial versions of WhatsApp or scraping | Not stated | Switch to the official app or stop scraping. If not, the account can be banned for good. The length is not published. |
| Banned | “This phone number can’t use WhatsApp” | Nothing, and no access to chat history or backups | Tap Request review in the app. Per the message, most reviews finish within 24 hours. No button means the ban is final. |
Libraries that connect as a linked device can be logged out by a restriction, since the page lists linked devices among the blocked features.
Builders often meet a restriction as an error code. Error 463, the “reachout timelock”, is the one most reported in 2026: sends to new contacts fail while the session stays connected. A Baileys contributor who read WhatsApp Web’s client code concluded in March 2026 that it is an account restriction on new chats (Baileys #2441). That matches WhatsApp’s restricted-account page, which never mentions the code. Details: error 463 explained.
What do builders report gets numbers banned?
Builder reports on GitHub are not WhatsApp statements.
- Cold first messages. Since March 2026, Baileys users report that sends to people who never wrote fail with error 463, while chats with people who did write keep working (Baileys #2441). The thread describes a time-based limit on messaging unknown people; what triggers it is still a hypothesis.
- The act of linking. In May 2026 a WAHA user reported that every fresh number was banned one to two minutes after linking, with nothing sent or received (WAHA #2068). The maintainer answered with default settings and links to the project’s blocking and proxy guides. The thread ends without a cause.
- Shared or datacenter networks. Unconfirmed. The white paper says network reputation counts: a new account on a network recently used by abusers is more likely to be judged abusive. WAHA’s maintainer pointed to a proxy guide in that thread. Nothing shows that a given proxy setup prevents bans.
Few reports give the number’s age, volume or network, so treat them as leads, not rules.
A checklist to remove the avoidable triggers
Each item names the kind of source behind it.
- Let people write first. Official: message only people who contacted you first or asked you to. Give users a wa.me link instead of sending cold.
- Honor stop requests at once. Official: stop when a contact asks. Our reading: a typed STOP costs a user less than a report or a block, so offer it.
- Do not burst from a new number. Official: sudden volume from a new account is the white paper’s clearest example of abuse. Ramp up slowly: see how to warm up a new number.
- Choose unofficial clients knowingly. Official: linking to unofficial versions violates the Terms and can end in a ban. No setting removes that. See unofficial WhatsApp APIs compared.
- Give each number its own phone, IP and history. Official: registration data and network history feed detection. Reported: builders add proxies. Our reading: do not run many accounts from one device or network.
- Plan for losing a number. Keep each user’s history and a second way to reach them on your side, so a ban costs you a number, not your users (moving users to a new number).
How do whasp’s rules map to these triggers?
whasp builds most of this checklist into every number: it runs real personal WhatsApp accounts on cloud phones, so it can enforce the rules for you.
| Trigger | What whasp does |
|---|---|
| Cold first messages | Users always write first. A send to someone who never wrote returns 409 chat_not_open, on replacement numbers too. |
| Unwanted messages | STOP, UNSUBSCRIBE, OPTOUT, CANCEL, END, QUIT or “opt out” opt the chat out automatically. It lifts when the person writes again. |
| Bursts and volume | Sends from one number go out one at a time, about a second apart. Per number: 100 active chats in 30 days, 40 new chats in 24 hours, 1,500 sent in 24 hours. |
| Brand-new numbers | New numbers are warmed up so they start like a person. |
| Shared networks | Each number has its own ISP IP, never shared with another number. |
| Trouble you do not notice | Health is healthy, at_risk or critical, with a webhook event on every change. at_risk covers a WhatsApp warning, a temporary ban or a limit above 80%. |
| A ban | The number is retired and you get a ready one at no cost, same label, profile and paid period. A number.replaced event tells you. |
What whasp does not change: a number can still be restricted or banned. whasp is not the WhatsApp Business Platform and is not affiliated with Meta or WhatsApp. Its numbers are personal accounts, so WhatsApp can restrict or ban them like any other. What changes is who runs the phone and carries the loss. Chats do not move: the new number starts empty, and your users have to write to it.
If your bot serves your own customers with support, orders or bookings, the official Business Platform is built for that. How to put an AI agent on WhatsApp compares every option.
What to do next
If a number was banned today, start with what to do next. Before a launch, read how to warm up a number and error 463. Limits and prices are on /pricing, the API in /docs.
Questions
Why did WhatsApp ban my number when I only run a bot?
WhatsApp says it bans accounts whose activity it believes breaks its Terms, for example spam, scams, bulk or automated messaging and use of unofficial apps. The Terms also rule out non-personal use unless WhatsApp authorizes it. It does not tell an account which signal caused a ban.
How many messages per day gets a WhatsApp number banned?
WhatsApp publishes no number. Its 2019 white paper gives only an extreme example: an account that registered five minutes earlier and tries to send 100 messages in 15 seconds is almost certainly abusive. Builders report restrictions at far lower volumes, so there is no safe count to rely on.
What is the difference between a restricted and a banned WhatsApp number?
A restricted account can keep chatting with people it already talked to but cannot start chats with new contacts, and the restriction ends by itself. A banned account cannot use WhatsApp at all, and the owner can request a review in the app unless the ban is final.
Can WhatsApp ban a number that never sends a message?
Builders report numbers banned minutes after linking a third-party client, with nothing sent. WhatsApp has not explained why. Its help pages say that linking an account to unofficial versions of WhatsApp can lead to temporary or permanent bans.
Keep reading
All articles
Rules and bansWhatsApp's AI chatbot ban 2026: the rule and where it stands
Since 15 January 2026 Meta bars general-purpose AI assistants from WhatsApp's Business API. The rule, who left, what Italy and the EU did, and what is still open.
Rules and bansChatGPT, Copilot and AI assistants on WhatsApp: what happened
ChatGPT, Copilot and Perplexity left WhatsApp on 15 January 2026. Only ChatGPT is back, in the EEA since 13 July 2026. The dates, the EU order and what it means.
Rules and bansIs your AI bot allowed on the WhatsApp Business API?
Under Meta's 2026 terms a support bot is fine and a general-purpose assistant is not. A test with examples, the gray areas, and what happens if Meta rules you out.