You uploaded the video, the automatic closed captions switched themselves on, and the box in your publishing checklist says “captions: done.” The question worth asking is what that ticked box is supposed to certify. Does the platform consider the job finished once a caption track exists, or does it ask you to do something with that track?
The checkbox and what it stands in for
This piece is a reading of documents. It covers two YouTube Help pages and two W3C pages, each fetched with an HTTP request and dated below. No automatic captions were tested, sampled or compared here, and nothing in this piece describes how any captioning system performs on real audio. Where a claim is the writer’s reading of a page and not the page’s own words, the text says so.
One distinction runs through everything that follows. A platform enables something when it offers a feature that generates captions. A platform asks something of a publisher when its documentation contains an instruction, usually with “should” or “always” in it. A ticked box tells you the first thing happened. Only the documents can tell you whether the second thing is expected of you. In this piece, “closed captions” and “subtitles” mean the timed text that appears over video, and nothing else.
What YouTube’s automatic captioning page tells the publisher
YouTube’s “Use automatic captioning” help page opens by calling captions “a great way to make content accessible for viewers” and says YouTube can use speech recognition technology to create them automatically. Then it qualifies the feature. Its note reads: “These automatic captions are generated by machine learning algorithms, so the quality of the captions may vary. We encourage creators to add professional captions first.”
Later on the same page comes the sentence that matters most to a checklist: “automatic captions might misrepresent the spoken content due to mispronunciations, accents, dialects, or background noise. You should always review automatic captions and edit any parts that haven’t been properly transcribed.”
Two things stand out. First, the review is placed on the creator, in an instruction: “You should always review.” That is not a description of what the system does. It is a description of what you are expected to do. Second, the page names its own causes of misrepresentation, which means no one needs to assert an error rate to make the point. The platform documents that its output can be wrong and why, and it does not give a figure. Neither will this piece.
The page also lists reasons a video may get no automatic captions at all. The video’s audio may still be processing, the language may be unsupported, the video may be too long, the sound quality may be poor or the speech unrecognised, there may be a long silence at the start, or there may be overlapping speakers or several languages at once. Read plainly, that is the platform stating its own limits.
Accessed 29 September 2026 at support.google.com/youtube/answer/6373554. No date was visible on the page.
Where the review actually happens: what the edit page says
The instruction to review would be empty without a place to do it. YouTube’s “Edit or remove captions” page opens its editing section with a plain statement: “You can change the text and timestamps of your captions.” Its note on automatic captions adds: “If you’re editing captions created automatically, a new caption track that includes your revisions will be generated.”
The page then walks through the route in YouTube Studio: Subtitles, the video, Edit, and for automatic captions Duplicate and edit, then Publish. A separate section covers adjusting timing in the captions editor. An editing route through a downloaded file also exists, and this piece stops there.
Here is the writer’s reading, labelled as a reading: because YouTube documents both text and timestamps as things you can change, timing belongs in the review as well as wording. The page does not say “check the timing” as an instruction. It documents that timing is editable, and a reasonable publisher can take that as a sign that timing is part of what the platform considers worth correcting.
Accessed 29 September 2026 at support.google.com/youtube/answer/2734705. No date was visible on the page.
What W3C says about automatic captions on their own
The W3C Web Accessibility Initiative is not a platform and does not set platform rules. Its “Captions/Subtitles” page, listed as updated 17 September 2024, has a section headed “Automatic Captions are Not Sufficient.” Its first sentence: “Automatically-generated captions do not meet user needs or accessibility requirements, unless they are confirmed to be fully accurate.”
The same section says that “often the automatic caption text is wrong and does not match the spoken audio,” and describes the errors as happening “sometimes in ways that change the meaning (or are embarrassing).” The page gives its own worked example of a caption that reverses a meaning. It is not reproduced here. Read it on the page. The page also says automatic captions “can be used as a starting point for developing accurate captions and transcripts,” which is a description of a draft.
The second W3C document is the WCAG 2.2 Understanding page for Success Criterion 1.2.2, Captions (Prerecorded), a Level A criterion. The criterion says captions are provided for all prerecorded audio content in synchronized media, with a narrow exception for media that is an alternative to text on the page. Its intent sentence is the useful part for anyone thinking about what a good track contains: “Captions not only include dialogue, but identify who is speaking and include non-speech information conveyed through sound, including meaningful sound effects.”
Be clear about what this is. W3C guidance is not a platform rule, and none of the YouTube pages fetched for this piece cite it as a requirement. Nothing here is a legal or compliance claim. The reader is being shown two different kinds of document that point the same way: one from a platform, written as help for creators, and one from a standards body, written about what accessible captions are.
Enables versus asks: a side-by-side reading
| Document | What it enables or describes | What it asks of the publisher | Exact wording that shows it | Anything beyond turning captions on? |
|---|---|---|---|---|
| YouTube, Use automatic captioning | YouTube can use speech recognition technology to create captions automatically. | Review the automatic captions and edit what was not properly transcribed. | “You should always review automatic captions and edit any parts that haven’t been properly transcribed.” | Yes. It asks for review and editing. |
| YouTube, Edit or remove captions | Caption text and timestamps can be changed; editing automatic captions generates a new track with the revisions. | Nothing stated as an instruction. It documents the means. | “You can change the text and timestamps of your captions.” | No instruction, but it supplies the place where the review happens. |
| W3C WAI, Captions/Subtitles | Tools can turn a soundtrack into a timed caption file, and some video sites provide automatic captions. | Confirm that automatic captions are fully accurate before treating them as adequate. | “Automatically-generated captions do not meet user needs or accessibility requirements, unless they are confirmed to be fully accurate.” | Yes. It asks for confirmed accuracy, as guidance and not as a platform rule. |
| WCAG 2.2, Understanding SC 1.2.2 | Captions are provided for prerecorded audio content in synchronized media. | Captions should carry more than dialogue: speaker identity and meaningful non-speech sound. | “Captions not only include dialogue, but identify who is speaking and include non-speech information conveyed through sound, including meaningful sound effects.” | Yes. It describes what the content of the captions should be. |
Where Meta’s documentation could not be read here
Facebook and Instagram have help pages about captions. For this piece they were requested by HTTP and returned a success status, but they are rendered in the browser and yielded no readable article text to the method used. A page that cannot be read cannot be quoted, paraphrased or characterised, so this piece makes no claim about what Meta’s documentation asks of publishers on captions, accuracy or review. That is a gap in this reading, and it is left as a gap on purpose.
What a review step looks like when you take the documents at their word
Here is a short pre-publish practice of five checks. Each one is tied to a sentence quoted above, and where a check rests only on W3C context and not on a platform page, it says so.
- Review the whole automatic track, not a sample of it. Source: YouTube’s automatic captioning page, “You should always review automatic captions.”
- Correct the passages that were not transcribed properly. Source: the same sentence, “edit any parts that haven’t been properly transcribed.” Names, accents and noisy stretches are the causes the page itself lists.
- Check the timestamps. Source: YouTube’s edit page, “You can change the text and timestamps of your captions.” That timing belongs in the review is the writer’s reading.
- Check that speaker changes and meaningful non-speech sound are represented where the audio carries meaning. Source: WCAG 2.2 SC 1.2.2 intent. This comes from W3C context only; the YouTube pages fetched do not state it as an instruction for automatic captions.
- Treat “captions on” as the start of the checklist item. Source: W3C’s “unless they are confirmed to be fully accurate,” and YouTube’s own description of the output as something that “may vary.” This is W3C guidance, not a platform rule.
What the documentation does not say matters just as much. None of the pages fetched gives an accuracy rate for automatic captions, so this post offers none, and you should be wary of any that appears without a named, dated source. If you also publish video on your own site, Google’s own video markup documentation is another primary-source reading about the steps around a video post.
The image counterpart
The same gap between a switch and a practice shows up with images, and the image counterpart, alt text, covers it in its own terms.
FAQ
Does YouTube require creators to review automatic captions?
YouTube’s help page says “You should always review automatic captions and edit any parts that haven’t been properly transcribed.” That is guidance written as an instruction. The page as fetched on 29 September 2026 describes no penalty or enforcement for skipping it.
Do automatic captions meet accessibility standards on their own?
W3C’s Web Accessibility Initiative says “Automatically-generated captions do not meet user needs or accessibility requirements, unless they are confirmed to be fully accurate.” That is W3C guidance, not a legal ruling and not a platform rule.
Sources
- YouTube Help, Use automatic captioning, HTTP 200, fetched 2026-09-29
- YouTube Help, Edit or remove captions, HTTP 200, fetched 2026-09-29
- W3C WAI, Captions/Subtitles, HTTP 200, fetched 2026-09-29
- W3C, Understanding SC 1.2.2 Captions (Prerecorded), HTTP 200, fetched 2026-09-29
