Twilio Call Recording: How to Set It Up and Get Structured Data Out

Twilio gives you three different ways to record a call, and picking the wrong one is the most common reason recordings come out one-sided, cut short, or missing entirely. This guide covers all three, the webhook that tells you the recording is ready, and what to do with the audio afterwards — because a raw MP3 is rarely the thing you actually need.

Option 1: TwiML <Record> — voicemail-style capture

The <Record> verb records whatever the caller says after it executes. It is the right tool for voicemails, message lines, and single-sided capture. It is the wrong tool for conversations — it records one leg only.

<?xml version="1.0" encoding="UTF-8"?>
<Response>
  <Say>This call may be recorded for quality and service purposes.</Say>
  <Record maxLength="120"
          playBeep="true"
          recordingStatusCallback="https://example.com/recording-done"/>
</Response>

Note the consent line before the recording starts. If you operate across US states, treat every call as all-party consent — announce it, always.

Option 2: recording a real two-sided conversation

For an actual phone conversation — an inbound customer call bridged to your team, or an outbound call — put the record attribute on <Dial>. Use the dual variant so each side lands on its own channel; every transcription engine downstream will thank you for it.

<Response>
  <Say>This call may be recorded for quality and service purposes.</Say>
  <Dial record="record-from-answer-dual"
        recordingStatusCallback="https://example.com/recording-done">
    <Number>+15551234567</Number>
  </Dial>
</Response>

record-from-answer starts when the callee picks up (no ringing in your audio); the -dual suffix gives you a two-channel file with the caller on one channel and your team on the other. Dual-channel audio noticeably improves transcription accuracy and makes speaker attribution trivial.

Option 3: the REST API

When you create calls programmatically, turn recording on in the same request:

const client = require('twilio')(accountSid, authToken);

const call = await client.calls.create({
  to: '+15551234567',
  from: '+15557654321',
  url: 'https://example.com/twiml',
  record: true,
  recordingChannels: 'dual',
  recordingStatusCallback: 'https://example.com/recording-done',
  recordingStatusCallbackEvent: ['completed'],
});

You can also start a recording on a live call by POSTing to the call's Recordings resource — useful when recording should begin only after a consent confirmation.

Getting the file: recordingStatusCallback

Do not poll for recordings. Twilio calls your webhook when the file is ready:

app.post('/recording-done', express.urlencoded({ extended: false }), async (req, res) => {
  const { RecordingSid, RecordingUrl, CallSid, RecordingDuration } = req.body;

  // The media file lives at RecordingUrl + '.mp3' (or '.wav')
  const audio = await fetch(RecordingUrl + '.mp3', {
    headers: {
      Authorization: 'Basic '
        + Buffer.from(accountSid + ':' + authToken).toString('base64'),
    },
  });

  // ...store it, or forward it to transcription/extraction
  res.sendStatus(200);
});

Two production notes. First, validate the X-Twilio-Signature header on this endpoint so only Twilio can call it. Second, decide your retention policy up front: recordings billed and stored in Twilio forever are a cost and a liability; many teams download, process, and delete.

The part nobody tells you: the MP3 is not the goal

Almost nobody wants an audio file. They want what was said on the call — and usually one specific thing: the appointment, the phone number, the address. You have three routes from audio to structured data. Build it yourself with a speech-to-text API plus an LLM extraction step plus validation — flexible, but you own accuracy, edge cases, and maintenance. Use Twilio Conversational Intelligence — good built-in operators for sentiment and entities, though generic entities are not validated, dispatch-ready data. Or send the recording to a purpose-built extraction service.

For the address case specifically: HearLoc consumes Twilio recordings directly — you point recordingStatusCallback at your HearLoc ingest URL (generated in the dashboard, signature-validated), and each recording comes back as a verified, dispatch-ready address: extracted, role-labeled, validated against Google, with gate codes and unit numbers captured and only the uncertain results flagged for review. Supported today; no code beyond the webhook URL.

# In your Twilio console or TwiML, set:
recordingStatusCallback = https://hearloc.com/api/v1/integrations/twilio/ingest/<your-token>

# Every completed recording is then processed automatically;
# results arrive in your dashboard, by signed webhook, or via the REST API.

Consent, in one paragraph

Recording laws vary by state; several require all-party consent. The safe default for any multi-state operation: play a recording notice at the start of every call and keep that notice inside the recording itself. We wrote a separate practical guide on US call-recording consent — linked below.