WhatsApp’s General-Purpose Chatbot Rule (2026): What It Bars, and the Self-Hosted Alternative

Did WhatsApp ban AI chatbots in 2026?

Not a blanket ban. Per TechCrunch's 18 October 2025 report, WhatsApp added an AI-provider section to its Business Platform terms barring AI model providers from distributing general-purpose assistants through Meta's channel, effective 15 January 2026. Automation on WhatsApp is governed by older, separate rules that still apply.

What was reported, and nothing beyond it:

• Who it binds: AI model providers distributing assistants on the Business Platform. Not a business replying to its own customers — though WhatsApp's separate automation rules still cover that business.
• Named as affected: OpenAI, Perplexity, Luzia, Poke and Microsoft (TechCrunch, Ivan Mehta, 18 October 2025).
• Outcome: by the 15 January 2026 effective date, OpenAI, Perplexity and Microsoft had announced their WhatsApp chatbots would stop working, per the same reporting.
• Contested: regulators in the EU, Italy and Brazil opened antitrust probes into Meta over these rules, as reported. Probes are investigations, not findings.

The table above sets the four routes side by side on pricing model and account risk. The comparison hub holds the full set; the per-route verdicts follow the policy sections.

What counts as a general-purpose chatbot under WhatsApp's terms?

The clause targets AI model providers distributing open-ended assistants through Meta's official channel. It is not written about a business answering its own customers. WhatsApp publishes no checklist separating a permitted flow from a barred one — it publishes an enforcement hook instead.

The Business Messaging Policy, last updated 23 September 2026, opens with a section titled "Create a Quality Experience" and states: "If content or actions violate our terms or policies, we may limit or remove your access to our services."

SporeSec, a security vendor, characterised the platform in a 6 July 2026 post as still permitting structured bots for orders, FAQs and lead capture. That is a vendor's reading, labelled as one; WhatsApp's own documents do not draw the line there.

The distinction a buyer can act on is scope, not vocabulary:

• Narrow and expected: a message about an order the person placed, a booking they made, a question they asked.
• Open-ended: an assistant that answers anything, for anyone, as a general destination.

The narrow shape is also the one that survives pacing and duplicate-content checks. That alignment is a design constraint seen from two directions, not a loophole.

Does the rule apply to the WhatsApp Business app or only the Business Platform?

The restriction sits in the Business Platform terms — the API and business-solution-provider channel. The free Business app and the consumer terms are separate documents. The Business Messaging Policy, last updated 23 September 2026, nonetheless covers the Business app, the Business Platform and "other WhatsApp for Business offerings or features that we may provide to you in the future."

That breadth matters. A small business running the free Business app on one phone is not distributing an assistant to the public, so the AI-provider clause has little to bite on there. A business wiring an unofficial client to its own number falls under the consumer-facing documents and the Help Center language on unauthorised automation, not under the AI clause.

Two doors, two rulebooks. The January 2026 change governs what AI model providers may distribute through Meta's official channel. It says nothing about, and grants nothing to, a self-hosted client paired as a linked device.

Is a self-hosted WhatsApp tool exempt from WhatsApp's terms?

No. Self-hosting changes which rulebook applies, not whether one does. The Help Center article on unauthorised automation states that WhatsApp's products are not designed for bulk or automated messaging, and that using them for those two purposes has always been considered a violation of its Terms of Service. (The copy retrieved was the Turkish-language version; the translation is as above.)

The enforcement basis is written down. WhatsApp's Messaging Guidelines, last updated 23 September 2026, say it may act "based on the limited information available to WhatsApp, including basic account, group, and community profile information, as well as messages reported by other users."

Partial information, plus user reports. That is why careful operators can still get caught, and why no vendor can promise otherwise. Genuka's comparison, last updated 8 October 2026, reaches the same conclusion about unofficial APIs: they drive an account the way WhatsApp Web does, without Meta approval, and the number can be banned.

The honest framing for SocialMate is architectural, not legal. It is not an AI model provider and does not distribute an assistant on Meta's platform. It is a local tool you run on your own number, and your own LLM is the brain. That describes how it works. It is not a claim of exemption from any rule.

Route A: how much does the WhatsApp Cloud API cost per message in 2026?

Meta bills per message delivered, not per message sent, across four categories — marketing, utility, authentication and service — with rates that vary by market–category pair, per WhatsApp's platform pricing page retrieved in October 2026. Business solution providers add their own fees on top.

Genuka's 8 October 2026 comparison reports that Twilio adds $0.005 to every message, inbound or outbound, and that 360dialog charges from €49 per number per month with no markup on Meta's fees. Those are secondary figures from a vendor that discloses its own stake. Verify each against the provider's current rate card before modelling anything.

The pricing model matters more than any single rate. Metered billing scales with the work, so the busier the operation, the wider the gap against a flat licence. The same pricing page notes that service messages are not charged, and that utility messages sent in response to users are not charged. Confirm the current rule on the page rather than assuming it.

Be fair about what Route A buys: compliance posture, template approval and a vendor relationship. It is not free of cost, and the official channel does not make an account unbannable. See SocialMate vs the WhatsApp Cloud API.

Route B: Evolution API vs WAHA vs a turnkey self-hosted app — what actually differs?

Evolution API and WAHA are raw engines: free, open-source, Docker-based, handing you a protocol layer. A turnkey app ships the anti-ban logic, interface and archival storage already built. The Terms of Service exposure is broadly the same across all three, because all three are unofficial clients paired as linked devices.

The real variable is labour, and it is easy to underprice. On a raw engine you deploy the container, design the schema, build the interface, write retry logic for blocked sends, and own every upgrade. Genuka's 8 October 2026 assessment is blunt about the category: unofficial tools drive a number the way WhatsApp Web does, without Meta approval, and the number can be banned.

Our own comparison of self-hosted engines puts the operational point the same way — Evolution API is "a raw self-hosted engine, not a turnkey product. Docker, anti-ban, GUI, and storage are on you" — a claim from our Evolution API comparison, dated 26 August 2026, so weigh it as ours. The same page notes Docker is mandatory on those engines.

Two fair counters, since the source is us. If you enjoy infrastructure work, the engineering time may be the cheapest part of your build and the raw engine wins on control. If a blocked send is a lost customer, the turnkey route is cheaper than it looks. For the engine-by-engine view, see SocialMate vs self-hosted APIs and vs WAHA.

Route C: what does self-hosting with SocialMate actually change?

Self-hosting changes three things: who holds the data, how you pay, and how much safety engineering you must build. It does not change the rules, which apply to a self-hosted tool exactly as they apply elsewhere. SocialMate runs as a desktop app on Windows, macOS and Linux, or headless via Docker and systemd on a VPS — the same app, the same release line.

This section rests on product facts and one cited assessment, not on third-party research. The three changes:

• Data. Chats and contacts stay in local storage on your machine. Nothing is routed through SocialMate's servers.
• Cost. A flat licence instead of per-message metering.
• Safety engineering. On a raw engine you build it; here it ships — randomised delays and jitter, pacing profiles, per-number rate limits, session warming, real typing indicators, read receipts before replying, adaptive throttling, a duplicate-content guard and live risk scoring.

One network detail is rarely stated. On the desktop app the connection leaves from your own residential IP, ideally the same home network as the phone that owns the account. That removes a risk factor; it does not eliminate bans. A VPS deploy uses a datacenter IP with lower network reputation — identical pipeline, different path. Pro adds per-account proxy routing so a VPS account can egress through your own residential or mobile proxy. It changes the path only; it never bypasses rate limits or warming, and it never lets you send faster. Our own comparison table sits at the hub.

Will I get banned for automating WhatsApp on my own number?

It is possible on every route, including the official one, and no published ban rate exists to size that risk. Times of India reported on 7 August 2026 that WhatsApp banned 5.1 million accounts in June 2026. Genuka's 8 October 2026 comparison states plainly that no ban rate is published by WhatsApp or by the open-source projects, so treat any percentage you see with suspicion.

The documented signals, each attributed:

• Modified client signature. WhatChimp's documentation, updated 24 April 2026, describes detection of unofficial or modified clients, often within hours to days. A vendor account, not policy text.
• Bulk and automated behaviour. WhatsApp's 2018 whitepaper, Stopping Abuse, reported removing over two million accounts per month for bulk or automated behaviour, more than 75% without a recent user report. Dated — read it as method, not current scale.
• User reports. Named in the Messaging Guidelines, 23 September 2026.

Device fingerprints, IP reputation and message pacing are not documented by any of these sources as WhatsApp's detection method. SocialMate's pacing, jitter and rate limits are risk reduction on your side of the wire, not a description of Meta's enforcement.

Can a legitimate account be banned by mistake?

Yes, and Meta has conceded it once. Times of India reported on 7 August 2026 that an early-August 2026 moderation wave locked legitimate accounts out, citing an on-screen message dated 3 August 2026, and that a Meta spokesperson told TechCrunch the locks were an error and that "mistakes can occur."

That is the strongest honest evidence in the public record, and it cuts against every vendor's safety claim, ours included. Automated enforcement acts on partial signals, so a careful sender can still be caught.

Three operational conclusions follow:

1. Keep volume gradual. A sudden jump is the least defensible pattern.
2. Keep content non-identical. The same text to many contacts is the pattern the rules describe as bulk.
3. Keep a fallback channel. Email or SMS for the handful of conversations that matter most.

WhatChimp's documentation claims no appeal succeeds for technical bans. Treat that as a vendor's account rather than confirmed policy. Prevention is still the only lever you reliably control.

How many messages a day can I send without risking a ban?

WhatsApp publishes no safe daily number, and no vendor can supply one honestly. What exists is a set of ceilings, plus gating that decides when you may approach them. On SocialMate, Free allows up to 200 messages per day on one account. Pro starts at 500 per account per day and, through High-Volume Mode after 72 hours of warming, scales to 5,000 per account per day.

Read those numbers as caps, not as safe volumes. High-Volume Mode is opt-in, unlocks only after the warming protocol, and is never framed as risk-free. A brand-new number starting at its ceiling is the least defensible pattern there is; the 72-hour window exists to establish a normal rhythm first.

A better question to ask any vendor is what happens when risk rises, not what the ceiling is. Here, adaptive throttling slows sends as the live risk score climbs, cold-outreach detection flags lists that were never opt-in, and cooled accounts resume by risk level rather than a fixed timer. Times of India's 7 August 2026 report of 5.1 million removals in a single month is the reminder that enforcement does not care about your intent.

Where does the AI actually run if I self-host?

Your model does the thinking, and your machine holds the memory and the transcript. SocialMate exposes WhatsApp as tools through a native MCP server (44 tools) for Claude Desktop, Claude Code, Cursor and Cline, and through a native n8n community node, current version 2.7.0, whose Get AI Context operation returns a whole thread, role-mapped and token-windowed. AI replies, summaries and triage generated by SocialMate are coming soon, not shipped.

That distinction is the durable answer for anyone whose assistant stopped working after the AI-provider clause took effect on 15 January 2026 (TechCrunch, 18 October 2025). The question that matters is not which platform hosts the assistant, but who holds the model, the memory and the transcript. SocialMate runs no model. It sends, reads, reacts, types, marks read, feeds the transcript back, and on Pro stores what your agent learned about a contact or a piece of media. Agent Memory is storage of what your model produced; SocialMate generates none of it.

Two setup details trip people. The local API is off by default in the app's API & Integrations screen, which is the most common reason MCP tools fail to appear. And MCP has no inbound push — to react to incoming messages automatically, poll the fetch tool or drive a trigger from the n8n SocialMate Trigger node. Neither is a per-service connector; MCP is an open protocol surface, like the REST API and webhooks.

Start with /docs/mcp-server for the three-step client setup, or /docs/n8n if your AI lives in a workflow rather than a desktop client.

What does SocialMate cost, and what does Free include?

A flat licence with no per-message fee: Free at $0 forever with no card, Pro at $99 per year or $10 per month. One licence includes one device; extra seats cost $4 per month each. WhatsApp accounts are unlimited on Pro, and no WhatsApp Business or Meta approval is involved.

The comparison that matters is against metered billing. Genuka's 8 October 2026 figures — Twilio's $0.005 per message, 360dialog from €49 per number per month — show how a per-message model compounds with volume in a way a fixed licence does not.

	Free	Pro
WhatsApp accounts	1	Unlimited
Messages per day, per account	200	500, up to 5,000 after 72h warming
Local HTTP API	Read endpoints, plain text send up to 200/day, 9 webhook events, 2 endpoints	Adds media, polls, locations, contact cards, group ops, smart queue, scheduled messages and batches, unlimited webhooks, all 35 events
History and search	Live only	Full archive, full-text search, replay on reconnect
Tunnel	Quick, rotating URL	Named, stable URL

The split is worth reading closely, because it is where free tiers usually hide their teeth. Human-like pacing, real typing indicators, read receipts before replying, pacing profiles, the duplicate-content guard, session warming and live risk scoring ship on Free. They are not an upgrade.

The 7-day Pro trial is opt-in from inside the app, needs only an email, takes no card, and is one per machine. See pricing for current figures, or download the free app and check the caps yourself.

What is a safe setup protocol for a new number?

A safe setup ramps gradually, varies content, contacts only people who expect to hear from you, and watches risk signals before raising volume. There is no configuration that makes a number safe — there is a sequence that keeps exposure low while you learn how the account behaves.

A protocol operators actually run:

1. Start on the home network. Run the desktop app on the same connection as the phone. It is the network picture WhatsApp already associates with the number.
2. Warm for 72 hours. Send small, personalised, expected messages. Do not start at your ceiling.
3. Use the Safe pacing profile. Randomised delays, real typing indicators, read receipts before replying — all of this ships on every tier, including Free.
4. Never send identical text to many contacts. The duplicate-content guard blocks it, because it is the pattern the rules describe as bulk.
5. Watch the risk score, not your ambitions. Cold-outreach detection and adaptive throttling exist to slow you before something breaks.
6. Keep a fallback channel. If a number cools, risk-based auto-resume handles recovery; customers still need email or SMS meanwhile.

The consent line is not decoration. One batch is many individual, personalised messages to people who already ordered, booked or asked. A bought, scraped or guessed list is the fastest route to a ban, and it is not a use case this product supports — SocialMate is not a bulk-messaging, broadcast or spam platform. For the longer version, see /kb/warming-new-numbers and /kb/will-i-get-banned.

Sources and retrieval dates

Every factual claim above traces to one of these. Dates are the source's own publication or last-updated date; all retrieved October 2026.

• TechCrunch, Ivan Mehta, 18 October 2025 — the AI-provider clause, the named affected providers, and the reported EU, Italy and Brazil antitrust probes. https://techcrunch.com/2025/10/18/whatssapp-changes-its-terms-to-bar-general-purpose-chatbots-from-its-platform/
• WhatsApp Business Messaging Policy, last updated 23 September 2026 — scope across business offerings, the "Create a Quality Experience" section and the enforcement clause. https://whatsappbusiness.com/policy/
• WhatsApp Terms for the WhatsApp Business Platform, page states "Effective September 23, 2026" — the primary terms document. https://www.whatsapp.com/legal/WhatsApp-Terms-for-WhatsApp-Business-Platform
• WhatsApp Messaging Guidelines, last updated 23 September 2026 — the "limited information available" enforcement basis. https://www.whatsapp.com/legal/messaging-guidelines
• WhatsApp Help Center, unauthorised use of automation or bulk messaging (Turkish-locale copy retrieved) — the Terms of Service line on bulk and automated messaging. https://faq.whatsapp.com/5957850900902049/
• WhatsApp Business Platform pricing, retrieved October 2026 — per-message billing by category, charged on delivery, with service and user-initiated utility messages not charged. https://whatsappbusiness.com/products/platform-pricing/
• Times of India, Neha Saini, 7 August 2026 — 5.1 million accounts banned in June 2026, and the early-August 2026 moderation error a Meta spokesperson acknowledged to TechCrunch. https://timesofindia.indiatimes.com/technology/tech-news/whatsapp-banned-5-1-million-accounts-in-june-heres-what-can-get-your-account-banned/articleshow/133023334.cms
• WhatsApp whitepaper, Stopping Abuse: How WhatsApp Fights Bulk Messaging and Automated Behavior (2018) — over two million accounts removed per month, over 75% without a recent user report. https://internetlab.org.br/wp-content/uploads/2019/10/WA_StoppingAbuse_Whitepaper_020618-Final-1.pdf
• Genuka, last updated 8 October 2026 — the disclosure that no ban rate is published, the Twilio $0.005 and 360dialog €49 figures, and the ToS assessment of unofficial APIs. https://wa.genuka.com/en/docs/compare/unofficial-whatsapp-apis
• WhatChimp documentation, updated 24 April 2026 — modified-client detection window and the no-appeal claim. https://help.whatchimp.com/docs/account-safety/whatsapp-account-bans-why-they-happen-and-how-to-stay-protected
• SporeSec, 6 July 2026 — the structured-bot characterisation, used only as attributed commentary. https://sporesec.com/en/blog/whatsapp-unofficial-api-ban-risk
• SocialMate documentation and product pages — /pricing, /docs, /docs/mcp-server, /docs/n8n, /compare/.

Three sources above are vendor blogs from companies selling in this category: Genuka, WhatChimp and SporeSec. They carry claims no primary document does, and every such claim is labelled as theirs in the text. Verify each against a live URL before relying on it. The Times of India figures, the Twilio and 360dialog rates, and the SporeSec characterisation in particular are single-sourced here.

Route	Who it is for	Pricing model	Where conversations live	Ban / account risk	What you build yourself
Meta Cloud API	Teams that need the official channel and will absorb metered costs	Metered per message delivered, in four categories (marketing, utility, authentication, service); rates vary by market–category pair, per WhatsApp's platform pricing page retrieved October 2026	Meta systems, outside your infrastructure	Official channel, so permitted by policy — not risk-free. Enforcement still applies, and Times of India reported 5.1 million accounts removed in June 2026 (7 Aug 2026)	Business verification, template approval, integration, billing
Business solution providers (Twilio, 360dialog, WATI)	Teams that want a vendor to run the integration and support	Meta's per-message fees plus vendor charges: Genuka's 8 October 2026 comparison reports Twilio adds $0.005 to every message, inbound or outbound, and 360dialog licenses from €49 per number per month with no markup on Meta's fees	Vendor-hosted, on top of Meta's systems	Same official-channel exposure as the Cloud API, plus the vendor relationship; verification and approval still apply	Less integration work, more vendor dependency and a second invoice
Raw self-hosted engines (Evolution API, WAHA, Baileys)	Developers who want no licence fee and will build the rest	Free and open-source licence; the real cost is engineering time and maintenance	Your own database, which you design, run and back up	Unofficial client. Can violate WhatsApp's Terms of Service and the number can be banned. No safeguards ship with the engine	Docker deployment, anti-ban logic, interface, storage, upgrades — everything above the protocol layer
Turnkey self-hosted app (SocialMate)	Founders, marketers, agencies and developers who want their own number, local data and a flat fee without DevOps	Flat licence, no per-message fee. Free $0 (200 messages/day, one account); Pro $99/year or $10/month, starting at a safe 500/day per account and scaling, opt-in via High-Volume Mode, to 5,000/day after 72h of warming. One device included, extra seats $4/month	Local storage on your own machine. Nothing routed through SocialMate's servers	Unofficial client. Can violate WhatsApp's Terms of Service; bans are possible. The anti-ban pipeline — pacing, warming, adaptive throttle, duplicate-content guard, live risk scoring — reduces risk, it does not eliminate it	Nothing above setup. You supply the consent-first contact list and the content; pacing and safeguards are built in

Set up a new WhatsApp number for automated messaging

1. Install and pair Install the desktop app (Windows, macOS, Linux) or deploy the same app headless via Docker or systemd. Pair your number as a linked device by scanning the QR in the app, or in the browser at /admin on a server install.
2. Choose the Safe pacing profile Set pacing to Safe for the first 72 hours. Keep the real typing indicator and read receipts before replying switched on; both ship on Free.
3. Leave the duplicate-content guard on It is always on for Free and on for Pro. The High-Volume Mode override exists for deliberate cases only; leave it off while a number is new, because identical text to many contacts is the pattern the rules describe as bulk.
4. Send below the daily cap Free caps at 200 messages per day on one account; Pro starts at 500 per account per day. Treat both as ceilings, not targets, and stay well under them during warming.
5. Wait 72 hours before raising volume High-Volume Mode is opt-in and gated behind the warming window, scaling to 5,000 messages per day per account. Enable it only after the number has run a normal, varied rhythm for three days.
6. Check the risk dashboard before each increase Review the live risk score, cold-outreach flags and any throttle events in the app before raising a cap. If risk is climbing, hold volume steady rather than overriding the throttle.
7. Turn on the local API only when you need it Open API & Integrations and switch the local API server on; it is off by default. Create an API key (it appears once, formatted sm_…), then point n8n, your MCP client or your backend at http://127.0.0.1:3456.
8. Wire a fallback channel Keep email or SMS live for the conversations that matter most, so a cooled number costs you a delay rather than a customer.

Frequently asked questions

Does the January 2026 rule cover an AI agent I build myself with n8n or MCP?

The clause governs AI model providers distributing assistants on Meta's Business Platform. It is not written about a business running its own agent on its own number. But that agent still falls under WhatsApp's separate automation rules and the Help Center position that its products are not designed for bulk or automated messaging. SocialMate runs no model: your LLM is the brain, and SocialMate is the hands and the local memory. Bans remain possible.

Does the rule stop my business automating its own WhatsApp conversations?

Not directly. The clause targets AI model providers distributing general-purpose assistants through Meta's channel, not a business sending its own order updates or bookings. Automated messaging is still constrained by the Business Messaging Policy and Messaging Guidelines of 23 September 2026 and by the Help Center position on automation, so consent-first, expected messages remain the only defensible shape.

Are unofficial WhatsApp APIs against the Terms of Service?

They can be. Genuka's comparison, last updated 8 October 2026, states that unofficial APIs drive a number the way WhatsApp Web does without Meta approval, that this breaks the Terms of Service, and that the number can be banned. That applies to Evolution API, WAHA, Baileys and any tool pairing as a linked device, SocialMate included.

How many accounts does WhatsApp ban each month?

Times of India reported on 7 August 2026 that WhatsApp banned 5.1 million accounts in June 2026. The only other figure in the source set is from WhatsApp's 2018 whitepaper — over two million accounts per month for bulk or automated behaviour, more than 75% without a recent user report — which is too old to treat as current scale. No published ban rate exists.

Can I appeal a WhatsApp ban?

WhatChimp's documentation, updated 24 April 2026, claims no appeal succeeds for technical bans. Treat that as a vendor's account rather than confirmed policy. The practical conclusion holds either way: prevention through warming, pacing and a consent-first contact list is the only lever you reliably control.

Do I need Docker to self-host WhatsApp automation?

Not with SocialMate. It runs as a desktop app on Windows, macOS and Linux, and can also run headless on a VPS via a Docker image or a systemd installer — the same app in the same release line. For Evolution API, WAHA and similar engines, Docker is the normal deployment path.

Is SocialMate affiliated with WhatsApp or Meta?

No. SocialMate is independent and is not affiliated with, endorsed by, or partnered with WhatsApp or Meta. It is not the official WhatsApp Business API. It is an application you run yourself on your own number, without Meta Business verification.

Which parts of the anti-ban engine are paid?

None of the human-behaviour ones. Randomised delays and jitter, pacing profiles, per-number rate limits, session warming, real typing indicators, read receipts before replying, the duplicate-content guard and live risk scoring all ship on Free. Pro adds volume, media, groups, polls, scheduled messages and batches, the smart queue and per-account proxy routing.