Turn on automatic captions, then spend ten minutes correcting them. The errors cluster on exactly the words a viewer needed to hear.

What automatic captions get right

Every video platform now generates captions without being asked, and they are considerably better than they were.

Ordinary conversational speech comes out largely correct, which is most of what most videos contain.

So the argument that automatic captions are useless is out of date and worth abandoning.

The problem is not the average accuracy; it is which words the errors land on.

Where they fail, predictably

Those are not random failures. They are precisely the words carrying the information, so a caption track can be ninety per cent accurate and useless for the purpose somebody is reading it.

The number problem is the worst

Since it can be actively misleading rather than merely unhelpful.

Automatic transcription regularly mishears figures, so fourteen hundred becomes forty hundred, and two point four metres becomes two point for.

Somebody relying on the captions then reads a price or a measurement that you never said.

That is worse than no captions, because it is wrong with confidence.

Check every number in the transcript specifically, which takes a minute and catches the errors that matter most.

The same applies to any address, telephone number, or date you mention.

A worked example

A business posted a short video explaining a common repair and left the automatic captions in place.

Reading the transcript, their trade term appeared as three different wrong words, their own business name was unrecognisable, and a price of eleven hundred and fifty was captioned as eleven fifty which read as a time.

Editing the whole thing took about twelve minutes for a four-minute video.

Afterwards the transcript was accurate enough to publish as text beside the video itself, which they did.

That page began appearing in search results for the repair question, which the video on its own never had.

The edit pass produced a page as well as usable captions.

The edit pass

Which is the whole method and is not difficult.

Open the caption editor on the platform, which every one provides.

Read the transcript rather than watching the video, since reading is faster and errors are obvious in text.

Fix the trade terms, the names, and every number.

Add punctuation where sentences run together, since automatic captions frequently produce one continuous stream.

Roughly three minutes of editing per minute of video is a realistic figure once you have done one.

Speaker changes and sounds

For anything with more than one person.

Where two people speak, the captions need to indicate who, which automatic ones generally do not.

Where a sound carries information, such as a click or an alarm, that belongs in the captions too.

Most small business videos are one person talking to a camera, where neither of those applies.

Check whether yours is the exception before spending any time on it, since for most trades the answer is no.

A walkthrough with a customer talking is the case worth checking.

So this matters for interviews and site walkthroughs rather than for the ordinary talking-to-camera case.

Where you do have two speakers, the platform's editor will let you mark the change with a name, which takes seconds per switch.

Publish the transcript as well

Which is the step that pays for the editing.

Once corrected, the transcript is a page of text saying everything the video says.

That is readable by anybody who would rather not watch, usable by search, and skimmable in a way a video is not.

Put it below the video with a heading, lightly tidied into paragraphs.

A business doing that turns one video into a video and a page, for ten extra minutes.

It also means the content survives the video platform, which is the recurring argument in all of this.

A video removed or a channel closed takes everything with it, and a transcript on your own site does not.

Burned-in captions are a separate thing

Worth distinguishing, since they are common on social.

Captions burned into the video image cannot be turned off, resized, translated, or read by anything.

They work well for silent autoplay in a feed and they do not serve the same purpose as a caption track.

Where a platform supports a proper caption file, use that as well rather than instead.

On platforms that do not, burned-in is better than nothing and should be high contrast and large.

Say the important things clearly on camera

Which reduces the editing work at source rather than after the fact.

Automatic transcription struggles most with speech over noise, fast delivery, and turning away from the microphone, all of which are normal on a job site.

Saying prices, product names, and your business name slowly, facing the camera, in a quiet moment, produces captions that are close to correct without any editing.

That is a habit rather than equipment, and it also makes the video better for everybody listening to it.

Where the site is genuinely loud, record the important sentence separately and put it at the start.

The counter-case

Most small business videos have very few viewers.

Editing captions on a video with two hundred views is a proportionate ten minutes and on one with eleven views it is not.

Automatic captions unedited are also considerably better than none, so the choice is between good and better rather than between something and nothing.

And a business that produces no video at all has nothing to do here.

Where you have three old videos, edit the one that gets watched and leave the other two, on the same reasoning as everywhere else in this run.

Turn on automatic captions, spend ten minutes fixing the terms and the numbers, and publish the corrected transcript below the video.

The ten minutes

  1. Turn on automatic captions.
  2. Open the caption editor.
  3. Read rather than watch.
  4. Fix the trade terms.
  5. Check every number.
  6. Add punctuation.
  7. Publish the transcript.

Step five is the one to do even if you skip everything else, since a mis-transcribed price is worse than no caption at all and takes a minute to find.

Adding them retroactively is covered in captions on a video you already published.


Frequently asked questions

Are automatic captions any good?

Better than they were. Ordinary conversational speech comes out largely correct, so the argument that they are useless is out of date.

Then what is the problem?

Which words the errors land on. Trade terms, product names, place names, numbers, and your own business name, which are precisely the words carrying the information.

Why are numbers the worst?

Because they mislead rather than merely fail. Fourteen hundred becomes forty hundred, and somebody reads a price you never said, which is worse than no caption.

How long does editing take?

Roughly three minutes per minute of video once you have done one. Read the transcript rather than watching, since errors are obvious in text.

What should I do afterwards?

Publish the corrected transcript below the video. It is readable by anybody who would rather not watch, usable by search, and skimmable in a way video is not.

What about burned-in captions?

They cannot be turned off, resized, or read by anything. They work for silent autoplay, but use a proper caption file as well rather than instead.

West Coast Media Solutions Inc. provides web design, web development, hosting, digital marketing, and business consulting to organisations across Canada, drawing on more than twenty-five years in the field.

Left the automatic captions as they came?

Read the transcript and check every number. That is where the damaging errors are.

Start a Conversation