Log inStart Free Trial
Meta Tech Provider

WhatsApp automation best practices: 12 rules

The core WhatsApp automation best practices: test before going live, let only one automation answer a given message, keep a clear route to a person, send non-template replies only within 24 hours of the customer's last message, use opt-in and approved templates after that, give the AI written facts, cap AI replies, and read the run log weekly.

easyCoexistence connects your WhatsApp Business number to Meta's Cloud API through Coexistence and builds these rules in: one automation per incoming message, a pause when you reply from the phone, the 24-hour check, caps on AI replies, failure emails and 24/7 monitoring of the number.

Escalation paths

WhatsApp's Business Messaging Policy asks businesses using automation for prompt, clear and direct routes to a person.

24 hours

Meta allows non-template messages only within 24 hours of the customer's last message; after that, approved templates.

2 conditions

Meta's opt-in rule: the person gave you their number and agreed to receive your messages.

1 automation

At most one automation answers an incoming message: keyword first, then first message, after hours, and the catch-all.

14 days

Meta disconnects a Coexistence number whose WhatsApp Business app is not opened on the phone for about this long.

20 seconds

How long a Call my webhook step waits for your system to answer with a reply.

What are the 12 rules at a glance?

Three come straight from Meta: offer a person, respect the 24-hour window, and get opt-in before you message first. The other nine come from how automations behave on a number you keep in the app, where you, your AI and your systems share the same chats. Each rule below names the exact setting or label that applies it, so you can check your own setup against the list in a few minutes.

1. Test before going liveSave it Off, use Test it, then switch On.
2. One owner per messageKnow the priority order; one catch-all only.
3. No duplicate greetingTurn off the app's own greeting or away message.
4. A person one message awayOffer a human, and act when asked.
5. Step back when you step inSame pause hours on every automation.
6. Respect the 24-hour windowText and AI inside it, templates after it.
7. Opt-in before you write firstAnd honor STOP at once.
8. Facts, not freedom, for the AIPrices and policies written in Instructions.
9. Cap length and volumeLongest reply and AI replies per hour.
10. Fast, verified webhooksReply in 20 s, check X-EC-Signature.
11. Read the runsSkip reasons and failure emails, weekly.
12. Keep the number healthyOpen the app; watch quality.

What should you do before an automation goes live?

Rule 1, test first. New automations start On, so to test without risk, save with the switch Off, type a customer-like message in Test it and read every step. A test sends nothing to anyone, but the AI and your webhook really run, which is the point: you see the real reply and the real token count.

Rule 2, decide which automation owns which messages. Only one automation answers a given message, chosen by kind: keyword automations first, then First message, then After hours, then the catch-all, oldest first within a kind. Two catch-alls never both reply; the older one wins and the newer never runs, so merge them.

Rule 3, retire the duplicate. The WhatsApp Business app's own greeting and away messages keep working after you connect, and software cannot see them. If a First message or After hours automation does the same job, turn the app's version off in its Business tools, or customers get two greetings.

How should automations treat customers and your team?

Rule 4, keep a person one message away. Meta's messaging policy asks for prompt, clear and direct routes to a human beside automated replies. Every Reply with AI step already gets a line telling the model to say a person will answer when asked; back it with a keyword automation on words like "human" or "attendant" that confirms with a fixed reply and alerts your team through Call my webhook.

Rule 5, let the automation step back when you step in. A reply from the app pauses an automation on that chat for "Pause on a chat after I answer it (hours)", 4 by default. The setting belongs to each automation, so give them all the same value, or one of them keeps talking over you.

Rule 6, respect the 24-hour window. Replies that are not templates, fixed or written by AI, go out only within 24 hours of the customer's last message; past that, the run is skipped. Wait goes up to 1,380 minutes, 23 hours, which keeps a follow-up to the customer's message inside the window.

Which rules keep messages welcome rather than spammy?

Rule 7, get opt-in before you write first. Meta's rule has two parts: the person gave you their number, and agreed to receive your messages. Templates you send to people who did not expect them draw blocks and reports, which lower the number's quality. Handle opt-outs in the same place: a keyword automation set to Is exactly, on STOP and its local variants, confirms in one line and calls your webhook to mark the contact.

Rule 12, keep the number itself healthy, because no automation answers on a disconnected number. Open the WhatsApp Business app on the phone at least every 10 days, since Meta disconnects a Coexistence number left unopened for about 14 days and API traffic does not count. Watch the quality rating, and pause whatever starts drawing blocks.

How do you keep AI replies accurate and affordable?

Rule 8, give the AI facts, not freedom. The Instructions field, up to 20,000 characters, is all the model knows about the business: write prices, opening hours, delivery areas and policies there, or say where the customer can find them. A line added under every prompt tells the model never to invent prices, stock, dates or policies, and it works best when the facts are actually in the Instructions. Edits apply from the next run, so update the text the day a price changes.

Rule 9, cap length and volume. "Longest reply (tokens)" defaults to 600. On a model that answers directly, 200 to 300 keeps answers phone-sized and cheaper; on a model that thinks first, the cap covers thinking too, so set 1,000 to 2,000 and ask for short answers in the Instructions. "AI replies per chat per hour, at most" defaults to 10 and stops two automated systems from replying to each other on your key. "Earlier messages it reads" defaults to 10: raise it for conversations that need context, lower it to save tokens.

What makes webhooks and monitoring dependable?

Rule 10, keep webhook steps fast and verified. A Call my webhook step waits 20 seconds; answer {"reply": "..."} within that time and the reply is sent, or answer with no reply when your system only records the message. Check X-EC-Signature before trusting the request, and do nothing irreversible when the body says dry_run is true, because Test it calls your webhook for real.

Rule 11, read the runs. Failed runs email you, at most once per automation every 6 hours, so a person can answer the customer. Skipped runs send no email: open the Runs tab once a week and read the skip reasons, such as "More than 24 hours since the customer's last message" or "AI reply limit for this chat reached this hour". An automation marked Paused is waiting for an AI key that works; saving one restarts it by itself.

Common mistakes

  • Creating a second catch-all to try a new prompt. Only the older one ever answers; test the new prompt with Test it, or edit the existing automation.
  • Running the app's away message next to an After hours automation. Customers get two replies, and only one of them shows up in your run log.
  • Planning a next-day follow-up with Wait. The step stops at 23 hours so the reply stays inside the window; later reminders need an approved template from your system or your AI.
  • Writing a welcome text that assumes a new customer. First message also fires for regulars the first time they write after automations are on, and again after about 30 quiet days.
Doing this with EasyCoexistence

Your AI runs on your own key, with no extra charge for 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

How many automations should one number have?

There is no limit on automations or runs, but most numbers start with three: a welcome on First message, keyword replies for the top questions, and an AI reply on the catch-all or After hours. Add more only when the Runs tab shows a gap.

Should automated replies say that an AI wrote them?

Meta's policy centres on a clear route to a person, and local consumer rules may ask for more. A short line in the welcome message, such as "Our assistant answers first; write human for a person", covers both and sets expectations.

Do these rules apply to several numbers on one account?

Yes, per number. Automations, priorities and pause hours belong to one number, while the AI key is shared per provider across the account. Usage of that key adds up on one provider bill, so caps matter more as numbers grow.

How often should I review my automations?

Read the Runs tab weekly, and review the Instructions whenever prices, hours or policies change. Ask your AI "Did any automation fail or skip this week, and why?" and it reads the runs through the MCP connector.

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