TL;DR:
Spoken answers disappear as soon as they are said. Captions give you something to look back at.
Reading along helps in noisy places, with fast speech, and with names, numbers and spellings.
Captions show what was said, not whether it was right. Check anything that matters against the screen.
You ask a question, the assistant answers out loud, and halfway through a sentence a bus goes past. Or the answer includes an email address, a number with a lot of digits, or a word you have never seen spelled. Speech is quick and it is gone the moment it is said, so you cannot go back and look.
Captions fix that. They turn the spoken answer into text on your screen as it is said, so your eyes and your ears get the same thing at the same time. This article explains how that works, walks through one example, and covers what captions cannot do.
Why captions are worth turning on
Hearing something once is a weak way to hold on to it. Reading it as well gives you a second way in, and you can glance back if your attention slipped. That is useful for everyone, and it is essential for people who are deaf or hard of hearing.
There are plain practical reasons too. You may be in a library, a train or a shared office where you would rather keep the volume low. You may be listening in a second language, where reading the words helps you catch the ones you almost missed. Or you may simply be a reader who takes in text faster than speech.
What happens between the voice and the words on screen
An assistant that talks has already written its answer as text before it speaks, because a voice is produced from text. Captions reuse that text. There are two common ways to show it.
Shown as it is spoken. The words appear a phrase at a time, in step with the voice. This is the one that lets you read along.
Shown all at once. The whole answer appears as a block of text. It is easier to scroll back through, but you cannot easily tell where the voice has got to.
Keeping the text in step with the voice is the hard part. The voice speaks at its own pace, so the captions have to follow the audio rather than race ahead of it. Good captions also break lines at natural places, so each chunk is short enough to read in a glance.
Walk-through: captions while it talks
Say you are in a noisy cafe, working on a spreadsheet with the volume turned down. You ask your assistant how to freeze the top row so it stays visible when you scroll. This is a made-up but ordinary example. The answer is spoken, and the captions follow.
You ask the question out loud or type it.
The assistant writes its answer as text and starts speaking it.
A caption area shows the first phrase: the menu to open and the name of the option.
As the voice moves on, the next phrase replaces the first, and you read it a moment before or just as you hear it.
You miss the exact menu name because of the noise. You glance at the caption and read it.
The answer ends. The caption stays long enough to read the last line before it clears.
Two details make this work. First, you look at the spreadsheet, not at the captions, most of the time. Captions have to be small and out of the way, because the point is to keep your eyes on your task. Second, you can catch up. If you looked away for a few seconds, ask for the last step again instead of trying to reconstruct it.
Reading along when the speech is fast or the words are hard
Reading along is the strongest use of captions. A few kinds of content are much easier to read than to hear.
Names and spellings. Two names can sound identical and be spelled differently.
Long numbers and codes. Reading a string of digits is far easier than holding it in memory.
Commands and file paths. Slashes, dots and capital letters are lost in speech.
Lists of steps. You can see where step three starts.
A good habit: when the caption shows something you will need to type, copy it from the screen rather than from memory of what you heard.
What captions cannot do
They do not check the answer. Captions show what was said. If the assistant was wrong, the captions are wrong in exactly the same way.
They can lag or break. If your connection is slow or the audio stutters, the text may drift away from the voice for a moment.
They may mishear you. Captions of the assistant's answer come from its own text, but the question it received was turned into text from your speech, and that step can get words wrong. If the answer seems to address a different question, say it again more slowly.
They are not a transcript you can keep. Unless the app saves them, the words disappear. Copy anything you want to keep.
Tip: If you need something exactly, such as an address or a command, ask the assistant to say it slowly and spell it. Then read the caption to confirm it.
Privacy and control
Captions are text on your own screen, so they do not send anything extra anywhere. The things to think about are around them. If you speak a question in a shared space, other people can hear it, and captions on a shared screen can be read by people nearby. Turn captions off, or lower them, before you share your screen in a meeting.
You should also know when an assistant is listening. A well designed one listens only when you tell it to, for example while the pill is open, and not all day. Check the settings of whatever you use, and look for a clear sign on screen when the microphone is on.
Try it yourself
Pick an assistant you already use that speaks its answers, and find its caption or subtitle setting. If it has none, ask whether it can show the answer as text beside the voice.
Ask a question with something easy to mishear in it, such as a name, and read along.
Turn the volume down to a low level and see how much of the answer you can follow from the text alone.
Ask the same question in a noisy place and notice which parts you needed the captions for.
If you notice the same kinds of things needing the text every time, such as numbers and file names, that tells you when to look at the captions and when you can just listen.
Frequently asked questions
Do captions slow the answer down?
No. They show the same text the voice is speaking, so the answer takes as long as it takes. Reading along can help you catch things you would otherwise have to ask to repeat.
Are captions the same as a transcript?
Not quite. Captions are shown while the answer plays and may clear. A transcript is a saved record of the whole conversation, and not every app keeps one.
Get started
