Start Free Trial
← All posts
Voice AI's caption gap

How to caption an AI chatbot's spoken reply

ChatGPT talks. Gemini talks. Almost none of them caption themselves. Here's how to read any AI chatbot's voice reply in real time — translated, if you need it.

7 min read Published September 2026

AI chatbots talk now. ChatGPT has a voice mode, Gemini has Live, and a growing list of assistants answer out loud instead of typing. None of them reliably caption their own reply for you — the transcript you see, if there is one, is often just your side of the conversation. If you need to read what an AI chatbot said instead of just hearing it, something listening to your speakers has to produce that text. That's true whether you're deaf or hard of hearing, working in a language that isn't your first, or simply in a room where audio isn't an option.

Why AI chatbot voice replies go uncaptioned

Voice interfaces spread fast this year. ChatGPT's Advanced Voice Mode, Google's Gemini Live, and a wave of assistant apps now answer out loud by default, the same way a phone call or a video meeting has audio. But a phone call and a video meeting both eventually grew captions, because a platform decided to own the transcript. An AI chatbot's voice reply usually has no equivalent: the app is generating speech, not committing to show you the words behind it.

Some chatbot apps show a running transcript of your own side, because they have to turn your speech into text to answer you at all. Fewer show a clean transcript of their own reply, and whether they do varies by app, by platform, and by whether you're on a phone, a browser, or a desktop client — and that can change with any update. If you need the reply as text every time, the safer approach is the one that already works for an uncaptioned call or meeting: something listening to your speakers turns the audio into a transcript, independent of whatever the chatbot app itself shows.

Where this actually matters

Accessibility

A deaf or hard-of-hearing user has no way to follow a voice-only reply unless something puts it in text — the same gap that makes an uncaptioned meeting a barrier, not a convenience.

Language practice and non-native use

Chatbot voice modes are popular for practicing a second language, but hearing a fast reply in an unfamiliar accent is harder than reading it — and reading it translated is easier still.

Streaming an AI conversation

Creators who talk to an AI chatbot on stream need captions for viewers watching muted, the same reason they'd caption any other stream audio.

Quiet offices and shared spaces

Playing a chatbot's voice reply out loud isn't always an option. Captions let you keep talking to it with the sound off.

Setup: captioning ChatGPT, Gemini, or any voice chatbot

Open the chatbot's voice mode

Start a voice conversation in ChatGPT, Gemini, or whichever assistant you use — in its app, or in a browser tab.

Open Unicaption and pick your audio source

In Unicaption, choose system or tab audio so it reads the chatbot's voice directly — not your microphone, which would also try to caption your own side of the conversation.

Set your languages

Choose the language you expect the reply in, or leave auto-detect on. Set a target language if you want a live translation next to the original.

Talk normally

Ask your question, let the chatbot answer out loud, and read the caption as it appears — moments behind the reply. Add any unusual terms to the custom dictionary first if the topic is technical.

One mic, one purpose. Caption a voice chatbot through your microphone instead of system audio, and the transcript often picks up your own spoken question along with the reply. System or tab audio keeps the transcript to just the chatbot's side.

Reading a reply in a language you're more comfortable in

AI chatbot voice modes typically reply in whatever language you spoke in — useful if you're practicing that language, less useful if you asked a quick question in your second language and want the nuance of the answer in your first. Live translation solves that without changing how you talk to the chatbot at all: you speak to it exactly as you normally would, and the caption on your side shows the reply translated.

It also works in reverse for practice: read the original alongside the translation, catch words you didn't recognize by ear, and build vocabulary from an answer you already understand the meaning of.

Getting a clean transcript of an AI voice reply

Captioning through a microphone instead of system audio turns your own spoken question into part of the transcript and adds room echo to the chatbot's reply.
Playing the chatbot through small laptop speakers next to an open mic invites feedback — switch to system or tab audio, or use headphones if you must use a mic.
Bluetooth audio can lag slightly before it reaches your speakers, which shows up as captions a beat further behind than usual — wired audio is more consistent to transcribe.
Fast, run-on chatbot answers are easier to caption accurately if you ask the assistant to answer in shorter turns, the same way a human speaker's pacing affects transcript quality.

Which setup fits your situation

Your situationBest setupWhy
Chatbot voice mode running in a browser tabTab audioCleanest source — captions only the chatbot's side
Chatbot app on desktop, not a browser tabSystem audioReads everything your computer plays, including a native app
Chatbot on your phone, played through a laptopMicrophone, close rangeKeep the phone speaker near the mic and the room quiet
Practicing a language or reading in translationSystem or tab audio + live translationOriginal and translated reply side by side
Whichever setup fits, the first 30 minutes of Unicaption are free every week, no credit card — enough to caption a real AI conversation and judge the transcript yourself. Start a session.

Frequently asked

Can I get live captions on ChatGPT's Advanced Voice Mode?

Yes. ChatGPT doesn't need to cooperate for this to work — run Unicaption alongside it with system or tab audio as the source, and it captions whatever ChatGPT says out loud, independent of whatever transcript ChatGPT's own interface does or doesn't show.

Does this work with Gemini Live or other AI voice assistants?

Yes, and with any voice chatbot, current or future. Unicaption reads audio your computer is playing rather than integrating with a specific app, so it works the same way for Gemini Live, a phone assistant played through a laptop mic, or any voice interface that comes next.

Will the AI chatbot or its app know it's being captioned?

No — Unicaption doesn't join, connect to, or interact with the chatbot app in any way. It listens to audio your own computer is already playing, the same as it would for a video call or a stream, and that audio is never stored.

Can I get a translated caption of what an AI chatbot said?

Yes. Unicaption shows the original transcription and a live translation side by side, in 60+ languages, so you can talk to a chatbot in one language and read its reply in another, moments behind the spoken answer.

Caption every AI conversation you have

Live captions and translation for any chatbot's spoken reply your computer can hear. 30 free minutes every week — no credit card.

Start Free Trial →