Inject the last voice transcription

Overview

A lot of WhatsApp traffic arrives as voice notes. Audio is great for the customer but terrible for automation: AI assistants need text, dashboards need text, and most workflow conditions need text. The !TRANSCRIPTION! command fills that gap by injecting the transcription of the last incoming audio into any outgoing message or internal note.

Command syntax

!TRANSCRIPTION!

The command takes no arguments. Drop it anywhere in the message body and Wazzap replaces it with the text of the last incoming voice note.

How it works

1. The contact sends a voice note over WhatsApp.
2. Wazzap transcribes the audio automatically and stores the text against the conversation.
3. You compose a message that contains !TRANSCRIPTION!.
4. Wazzap replaces the command with the latest transcription and dispatches the resulting message (or posts the internal note).

Examples

Internal team log of voice notes

!INTERNAL! Customer's last voice note: "!TRANSCRIPTION!"

Posts the latest audio in plain text into your team's feed. Your reps can scan it on the desktop without having to play the audio.

Feed AI workflow with the customer's words

!INTERNAL! Pass to assistant: !TRANSCRIPTION!

Use this as the trigger payload for a downstream workflow step that calls an AI model. The model gets the customer's actual words, not a generic system prompt.

Confirm understanding of an audio request

Just to confirm I understood: "!TRANSCRIPTION!". Is that right?

A polite read-back that closes the loop on audio requests, especially useful for booking and logistics conversations where mishearing has real consequences.

Use cases

• AI assistants over audio. Feed the transcription into your AI prompt so the model can reply to voice notes with text answers.
• Internal scanning. Let agents read voice notes in the team feed instead of playing every audio one by one.
• CRM audit trails. Stamp every conversation with the text version of what was said, so search and reporting actually catch audio content.
• Confirmation messages. Read the audio back to the customer in text form to confirm you understood. Great for logistics, bookings, support tickets.

Important notes

• The command resolves to the most recent incoming voice note. If the contact never sent an audio, expect an empty value or the literal command, depending on sub-account settings.
• Transcription quality depends on audio clarity and language. Wazzap handles Spanish, Portuguese, and English well. Heavy regional slang, background noise, or short bursts can degrade accuracy.
• The command is case-sensitive. Use !TRANSCRIPTION!.
• Pair freely with Delay, Spintax, Internal, Agent and Source. The parser handles all combinations.

Troubleshooting

The output is empty

The contact hasn't sent a voice note yet, or the last incoming message wasn't audio. The command only pulls from voice notes, not text. If your workflow can run before any audio exists, gate the step with a condition that checks the last message type.

The transcription is garbled or wrong

Audio quality matters. Background noise, very short clips, or mixed languages in a single note can confuse the transcription engine. If the contact's audio is consistently low-fidelity, consider a human review step before any AI workflow acts on the text.

The customer received the literal !TRANSCRIPTION! string

The message was sent from the phone (the WhatsApp app doesn't parse Wazzap commands), or the syntax was typed wrong. Send from the CRM's Conversations tab or a workflow step and confirm the command is in uppercase with exclamation marks on both sides.