WhatsApp call recording and transcription: rules, consent and how to use them

Yes, businesses can record and transcribe WhatsApp calls on the WhatsApp Business Platform. Meta’s Calling API lets you opt each inbound or outbound call into recording, transcription or both. Before anything is captured, both participants hear a spoken announcement with the purpose you provide, and after the call you receive a webhook to download the audio or a JSON transcript, which Meta keeps for 7 days.

The technical part is the easy bit. The harder questions are whether you may record in the first place, how you tell people, where the files live, who can open them and when they are deleted. This guide covers what Meta’s API does, the consent and privacy basics, storage and redaction, payment card rules, and how to turn transcripts into summaries, quality reviews and CRM updates. It is general information, not legal advice.

Can you record WhatsApp business calls?

You can, through the Cloud API Calling feature. Meta’s call recording and call transcription documentation describes two separate options:

• Recording captures the call audio.
• Transcription turns the call into text.

They are independent. You can turn on one, both or neither, and you decide per call: when you place an outbound call or when you accept an inbound one. Each has its own request settings, its own webhook and, according to Meta, its own pricing; check Meta’s documentation for current charges before you budget.

The standalone WhatsApp Business app does not offer this; it is a feature of the API. If you are new to calling, start with our WhatsApp Business Calling API guide.

How WhatsApp call recording and transcription work

	Recording	Transcription
Covers	User-initiated and business-initiated calls	User-initiated and business-initiated calls
Turned on	Per call, when you place or accept it	Per call, when you place or accept it
Notice	Spoken announcement to both sides with your purpose	Spoken announcement to both sides with your purpose
Output	Audio file (Ogg with Opus)	JSON document with text, speakers and timestamps
Delivered	Webhook after the call, usually within a minute	Webhook after the call, usually within a minute
Kept by Meta	7 days, then deleted	7 days, then deleted

The spoken announcement

Before recording or transcription starts, the API plays an announcement into both your audio and the customer’s. You supply two things:

• A purpose, up to 250 characters, written in the announcement language. It is required; a call with recording on and no purpose is rejected.
• An announcement language, from Meta’s supported list.

The customer then hears a sentence like: “The audio of this call will be recorded for the following purpose:” followed by your purpose. If both features are on, they hear one combined line saying the call will be recorded and transcribed. Capture starts only after the announcement ends, and a participant who does not agree can hang up.

Write a purpose that is short and honest:

quality assurance and to keep a record of your request

training our team and confirming the details of your booking

Check the language list before you launch. Meta lists localized announcements for English, Dutch, French, German, Hindi, Italian, Kannada, Portuguese (Brazil), Spanish, Telugu and Vietnamese at the time of writing. If your customers speak another language, plan how you will inform them clearly, for example with a greeting in their language and a note in the chat.

What the transcript contains

You do not pick a transcription language: the API detects it from the audio. Meta’s supported list is long and includes languages such as Arabic, English, German, Spanish and Turkish. The JSON transcript includes:

• the full conversation as text, with each part labelled Business or Customer
• the detected language and an overall confidence score
• time-stamped segments, with word-level timing and confidence

Speaker labels come from separate audio channels for your side and the customer’s, so they stay accurate when people talk over each other. If the call is in a language the model does not support, the transcript may be empty.

Downloading and keeping the files

The webhook includes a media ID and a download URL that is valid for 5 minutes; if it expires, request a fresh URL with the media ID. After 7 days the file is gone. Download every recording and transcript you need, check the SHA-256 hash Meta includes, and store it under your own retention policy.

Two related points:

• Voicemail. If you turn on Meta’s voicemail setting, a missed or rejected call can be answered with your announcement and the caller’s message is delivered to you as an audio message. Treat voicemails as recordings too.
• Recording on your own side. Meta’s calling FAQ notes that solution partners can add recording and transcription on top of the raw audio stream. If recording happens in your own systems rather than through Meta’s feature, Meta’s announcement is not played, so the notice is your job.

Consent and disclosure: the basics

Laws on recording calls differ by country, and sometimes by state or region. Treat this section as a list of questions to answer with your legal advisor, not as rules.

One-party or all-party consent. Some places allow a call to be recorded when one participant agrees, which can be your agent. Others require everyone on the call to agree. In the United States, for example, federal law and many states follow one-party consent, while some states require all parties. If you call across borders, plan for the stricter rule.

GDPR (EU and UK) and similar laws. Recording a call processes personal data, so you need:

• a lawful basis for each purpose (for example a contract, a legal obligation or legitimate interests, or consent where that is the right basis), documented before you start
• clear information: who records, why, how long you keep it and how to exercise rights
• data minimization: record only the calls and purposes you need
• retention limits: a deletion schedule you actually run
• access rights: customers can ask for a copy of their personal data, which can include recordings and transcripts

KVKK (Turkey). Turkey’s data protection law also requires a lawful basis and that people are informed about processing, and it has its own rules on transfers abroad. If you serve customers in Turkey, review recording with your KVKK advisor, including where recordings and transcripts are stored.

AI disclosure. If an AI voice agent answers the call, tell callers they are speaking with an automated assistant. Some laws require it, and it builds trust either way. Our WhatsApp voice AI agent guide covers how to open the call.

A practical baseline that works in most markets:

1. Say it in the greeting, even when Meta’s announcement plays.
2. Explain the purpose in plain words and keep it true.
3. Offer an alternative, such as continuing in the chat, for callers who do not want to be recorded.
4. Write it into your privacy notice and link it from your WhatsApp profile or first message.

Storing, securing and redacting recordings

Once files leave Meta, they are your responsibility. Set these rules before you turn recording on:

• Retention. Decide how long each purpose needs: quality review might need weeks, a contract dispute might need longer. Delete on schedule.
• Access. Limit who can play recordings and read transcripts by role, and keep audit logs of who opened what.
• Encryption. Encrypt files in storage and in transit, and keep them in a region that fits your data protection obligations.
• Redaction. Mask personal data in transcripts before they go to wider teams, analytics or AI tools: ID numbers, health details, addresses, account numbers.
• Separation. Keep raw audio in a restricted store; let most people work from redacted transcripts and summaries.
• Subject requests. Make sure you can find and export or delete a customer’s recordings when they ask.

If you send transcripts to a language model for summaries, the same privacy questions apply as for any AI tool. Our AI chatbot data privacy checklist walks through providers, keys and retention.

Payment cards: keep card data off recorded calls

Call centers that take card payments by phone have a specific rule to follow. The PCI Security Standards Council states that storing card verification codes (the three- or four-digit security code) in any audio recording after authorization violates PCI DSS, and that every effort should be made to keep them from being recorded at all.

On WhatsApp, there is a simpler path: never take card details by voice.

Agent: I'll send you a secure payment link in this chat now. You can pay there while we're on the call, and I'll confirm as soon as it goes through.

Train your team and configure your AI agent to refuse spoken card numbers and switch to a link. See accepting payments in WhatsApp chats for how payment links fit into a conversation.

How to use call transcripts

A recording sitting in storage helps nobody. Transcripts make calls searchable and useful:

Use	What you get
Call summaries	Reason, outcome and next step in three lines, attached to the customer
Handover	The next agent reads the summary instead of asking the customer to repeat
CRM updates	Lead details, booking or case status written back automatically
Quality review	Score a sample of calls against a checklist without listening end to end
Coaching	Find good and bad examples by topic and use them in training
Trends	Count reasons for calling and spot new problems early
Knowledge gaps	Questions the team or AI could not answer become new knowledge base entries

A simple quality checklist to score from transcripts:

• Greeting with name and business, and the recording notice given
• Need identified within the first minute
• Correct information, matching your policies
• No sensitive data requested by voice
• Clear next step confirmed with the customer
• Follow-up sent in the chat

Tie these to the metrics you already track; our guide to chat support metrics and SLAs explains how to report them.

Call recording and transcripts with MonoChat

MonoChat keeps calls and chats in one inbox, so what was said on a call stays next to the customer’s messages.

WhatsApp calls. With WhatsApp calls in MonoChat, your team or an AI voice agent answers calls with the customer’s chat history in view. After each call, MonoChat can summarize the outcome, tag intent and sentiment, and push structured notes to your CRM, helpdesk or internal systems.

Phone calls over SIP. On your SIP line, call recording and transcripts follow four steps:

1. Connect your SIP provider: trunk credentials, inbound numbers and outbound caller ID.
2. Set the recording policy: which calls are recorded, a recording notice in the call flow, and who can access recordings.
3. Turn on transcripts and summaries, so recorded calls become text with a short AI summary.
4. Review, search and automate: search transcripts, review quality in analytics and trigger workflows from call outcomes.

For privacy, MonoChat provides role-based permissions, audit logs, recording policy controls, PII-safe workflows and optional redaction; see our security page for more. AI call summaries extract intent, action items and CRM-ready notes, and workflow automation can create a ticket, book an appointment or send a follow-up on WhatsApp or SMS from the call outcome.

You can start on the free Basic plan (one agent, one channel) at app.monochat.ai, or see the pricing page for other plans.

Sources

• Meta for Developers: Call recording
• Meta for Developers: Call transcription
• Meta for Developers: Configure call settings (voicemail)
• Meta for Developers: Calling FAQ
• Meta for Developers: Calling API pricing
• PCI Security Standards Council: FAQ on card verification codes in audio recordings