Checking Automatically Generated Captions

In short

Treat automatic captions as a first draft, never as a finished product.
The usual errors are names and titles, mixed languages, wrong-sounding words and missing sounds.
A reviewer who speaks the language should read every line against the video.
Keep a short list of names and terms to give the reviewer before they start.

Treat automatic captions as a first draft. Read every line against the video, with the sound on, and fix what is wrong. W3C guidance on captions says automatic captions need accuracy checks and usually need correction. For an event video full of names, titles and mixed languages, that is especially true.

What goes wrong most often?

Problem	What it looks like	What to check
Names and titles	A guest of honour’s name misspelt, or Datuk Seri shortened or wrong	The programme, invitation or name list
Mixed languages	Malay words written as odd English words, or the other way round	Whether the reviewer reads both languages. See Malay-English code-switching in captions
Meaning	A word that sounds right but changes the sentence	Read the line in context, not word by word
Numbers and amounts	Dates, years, figures or percentages wrongly written	The slides or the speaker’s notes
Company and product names	Brand names turned into ordinary words	A short glossary from the organiser
Non-speech sound	Applause, laughter, music or an announcement not marked	Whether those sounds matter to someone who cannot hear them
Who is speaking	Two speakers merged, or an interpreter and a speaker confused	The run of speakers in the programme

For example, a speaker may thank “Yang Berhormat” and the software writes something unrelated. It looks small, but it changes the dignity of the sentence. This is an illustration of the kind of error to expect.

Who should review?

The best reviewer speaks the language and knows the event. That can be a colleague, the person who wrote the programme or a family member. If the speech mixes languages, the reviewer must understand all of them, not only one.

If the video is in more than one language, a separate reviewer per language is usually clearer than one person guessing.

A review process that works

Prepare: gather the programme, the speaker list with exact titles, and a short glossary of company, product and place names.
First pass: the reviewer plays the video with captions on and corrects every line, with the sound on. Do not read the text alone, because the meaning is in the voice.
Mark uncertain parts: when the reviewer cannot tell what was said, note the time and ask the speaker or organiser.
Check sounds: add or fix notes for applause, music or an important off-screen sound if the captions cover non-speech sound.
Second look: a different person reads the corrected text for typing errors and consistency of names.
Sign-off: the named decision-maker approves the text before it goes into the video or onto a platform.

The timecoded feedback tool lets reviewers note each correction against an exact moment. The revision planner helps you agree how many rounds you expect.

What standard should you agree?

Decide the standard before the review starts, not during it. Examples of rules to agree:

Names and titles must be exactly right.
Meaning must be correct in every sentence.
Fillers such as “um” may stay or go, but choose one.
Non-speech sounds are marked only when they matter to the viewer.
Each language is read by someone fluent in it.

Where checking fits in the whole plan

Checking is part of delivery, so it belongs in the brief and the quote, not at the end. See planning accessible event videos from the brief for who owns each step, and burned-in captions or separate caption files for why checking before burning text into the picture saves re-exporting. If the captions will be translated, review the original text first, then the translation. Our guides on Chinese subtitles and Tamil and other language subtitles cover that case.

What we confirm for your event

Whether the text is written by a person, generated by software, or generated and then corrected.
Who reviews each language, and what they need from you.
How much time to allow for review.
How corrections are returned and re-checked.

The caption and language planner helps you write the answers down, and our captioning and subtitles service page describes how we scope the work.

Tell us the languages of your automatic captions and who can review them, and we will confirm what checking we can do for your event.

WhatsApp us

Sources

W3C WAI, Captions/Subtitles, checked 2026-10-06
W3C WAI, Planning audio and video media, checked 2026-10-06

Checked on the date shown. Rules and venue requirements change, so confirm them for your own event.

Useful next steps

Planning accessible event video and languages

How to plan captions, subtitles, transcripts and readable graphics for Malaysian event video, and which accessibility guide to read first.

Captions and subtitles for your event video

Add captions or translated subtitles to your event video. What to specify about language, review and file formats so viewers can follow with or without sound.

Caption and language specification planner

Plan captions, translated subtitles and transcripts for an event video. Get a language and output matrix and an acceptance checklist before you brief an editor.

Captions, subtitles and transcripts: which do you need?

Captions, translated subtitles and transcripts do different jobs. Learn which one your event video needs, and what to settle with your videographer before quoting.

Plan an accessible event video from the brief

Write accessibility needs into your event video brief: who is responsible, what to specify and when to review, so captions and access are not left to the last edit.

A family video every relative can easily watch

Plan how relatives will actually watch your family video: on a phone, a TV or a shared link, with captions, clear sound and a version each person can use.

Burned-in captions or a separate caption file?

Burned-in captions are part of the picture; a caption file is separate and can be switched on or off. Learn which suits your event video and where it will be shown.

Questions people ask

Are automatic captions good enough to publish?

Usually not without correction. W3C guidance notes that automatic captions need accuracy checks and typically need to be corrected. Event speech often has names, titles, accents and mixed languages, which are the places software makes the most mistakes.

Who should check the captions?

Someone who speaks the language and knows the event, such as a colleague, the organiser or a family member. They should check every line against the video.

How long does checking take?

It depends on the length of the video and how clean the speech is. A useful rule is that checking takes longer than watching the video once, because the reviewer pauses to correct. We agree the review time and dates with you when scoping the event.

Do I need to correct every small error?

Fix anything that changes the meaning, misspells a person's name or title, or confuses the viewer. Agree beforehand how strict you want to be about small issues such as filler words, so the reviewer knows the standard.

Can you check captions I already made?

Review and correction of existing text can be agreed as part of the work, confirmed for each project. Send us the file and tell us which languages are involved.

Planning an event? Let’s talk it through.