Voice Replies Tutorial: Send an AI Draft as a WhatsApp Voice Note
Some customers talk instead of type. Send an AI draft as a voice note — tap the Voice button, listen to the preview all the way through, send. Plus how to pick your AI voice, and when the app suggests going voice.
Plenty of customers run their whole business by voice note — they talk, they expect you to talk back. Until now every AI reply went out as text. Now a draft waiting for your approval can go out as a real WhatsApp voice note instead: same words, spoken in a natural voice, delivered as the audio bubble your customer already lives in.
One thing stays exactly the same: you hear every voice note before the customer does. A voice message can't be unsent and can't be skimmed, so the app won't let one leave until you've listened to it end to end.
Prerequisite: you're already reviewing AI drafts — that's Suggest mode with company info added. If not, start with the AI Replies tutorial.
Step 1 — Send your first voice reply
Wherever a draft waits for your review — the approval pop-up in the chat window, or the pending-approvals card on the AI Nexus Dashboard — there's a 🎤 Voice button next to the usual buttons.
- Tap 🎤 Voice. The button shows Generating… for a moment while the draft is turned into audio.
- The preview starts playing automatically. Listen through — tap to pause and resume whenever you like.
- When you've heard the whole thing once, the button becomes ✅ Send as voice.
- Tap it. The customer receives an ordinary WhatsApp voice note.
The text path is untouched: Approve & Send still sends the draft as text, exactly as before. Voice is an extra option, never a replacement.
✓ Done when the voice note shows up in the chat as sent.
The listen-first rule
✅ Send as voice only unlocks after one complete listen. That's deliberate: with a text draft your eyes are the review, with a voice note your ears are. Two things people expect to be stricter than they are:
- Pausing is fine. The gate needs one full listen, not an uninterrupted one. Pause, answer another chat, come back, resume.
- Replaying is fine. Listening again never re-locks the button.
And two edges worth knowing:
- If the preview doesn't start on its own, tap ▶ Play preview to start it manually. Same rule applies.
- If you edit or refine the draft after generating, the preview is discarded — generate a fresh one. The voice note that goes out is always exactly what you heard, never an older take.
When the app suggests voice
You decide text-or-voice per reply, but the app gives you a nudge when voice is likely the right call. If the customer has been sending voice notes recently and the draft carries no price or link, the button highlights and reads 🎤 Voice · suggested — hover it and you'll see why: "Customer uses voice notes." People tend to reply in the medium they're spoken to in; mirroring it feels natural.
The reverse also holds — some drafts belong in text:
- Prices, links, and long numbers (order numbers, phone numbers) — customers scroll back to re-read these. You can't skim a voice note for a figure. Drafts like this simply don't get the highlight; the button still works if you choose voice anyway.
- Long drafts — over 500 characters, the voice button disables with "Over 500 characters — send as text." Nobody wants a two-minute synthetic monologue.
The suggestion is only ever a highlight. It never auto-sends, never pre-selects voice, and an un-highlighted button behaves exactly like a highlighted one.
Pick your AI voice
Open AI Nexus → Takeover Settings. Right below the AI mode switch is the AI voice section: four voices — Mia and Ava (female), Daniel and Leo (male), named in your interface language — each with a ▶ Preview button so you can hear it before committing.
You choose one voice for the whole workspace. Every voice note — every chat, every agent — speaks in it, so your business sounds like one consistent person, not a different stranger each message.
Plans: sending voice replies works on every plan. The plan only affects the voice picker:
- Free — the default female voice, fixed. You can see and preview all four, but the selection is locked.
- Pro and Enterprise — pick any of the four.
What this deliberately doesn't do
- Auto mode never sends voice on its own. Even with AI mode on Auto, voice notes remain manual — a human listens first, every time. Automatic replies stay text.
- Telegram approval stays text. Approving from your phone via Telegram remote sends the draft as text — there's no voice button in Telegram.
- No voice cloning. The four voices are a curated set. The app never imitates your voice or anyone else's.
Teams on the Enterprise Inbox get the same button: draft cards in the web workspace carry 🎤 with the identical listen-through rule.
Things that trip people up
"Send as voice won't unlock." You haven't heard the whole preview yet. Pausing is fine; skipping to the end isn't a thing — let it play through once.
"The voice button is greyed out." The draft is over 500 characters. Shorten it, or send as text.
"I generated a preview, got pulled away, and the button is back to 🎤." Previews don't stay valid forever. Just generate again — it takes a couple of seconds.
"The preview won't play." Use Regenerate voice on the error state and it re-synthesizes from scratch.
Got a customer who only ever voice-notes you? Download WhatsApp AI Pro, wait for their next message, and answer it in kind — draft, 🎤, listen, send.
Related guides:
- One WhatsApp AI for Text and Voice Messages — the inbound half: transcription, translation, and answers from your own facts
- AI Replies Tutorial: Add Company Info, Get Your First Draft
- Telegram Remote Control: Approve Drafts From Your Phone