Does a Video Transcript Help SEO? An Audit of 60 Videos

I attach a video to almost every article I publish, on the theory that a page carrying both text and video is the strongest thing you can hand a search engine. Sixty of them are live. This week I went looking for what those videos actually contribute to a machine reading the page, and the answer is close to nothing, because not one of the sixty carries a written record of what is said in it.

All 60 video articles on this site publish video markup with eight fields. None of those fields is the transcript.

Can a search engine read what a video says?

A search engine cannot hear a word of what a video says. A crawler is a program that fetches a page and reads the text in it, and reading is all it does. It does not press play, it does not listen, and it has no way to turn twelve minutes of speech into anything it can index. Whatever you said in the video is, to that program, simply absent.

That blind spot is easy to keep, because video feels like content in the way a paragraph feels like content. To a person watching, it is. To a machine reading the page, a video is a box with an address in it and some labels attached.

So the question worth asking about any video on your site is whether the information inside it exists anywhere on the page in a form the machine can take. Quality matters to the person watching and settles nothing here.

Does video schema already cover what a video says?

Structured data describes the video without containing any of what the video says. Think of schema as a block of machine-readable labels a page carries about itself, and for video those labels are things like the title, a short description, a thumbnail address and the address of the player.

Across every built article page on this site, I parsed what the schema actually carries. Sixty carry a video block, and every single one carries exactly the same eight fields: the type, an identifier, a name, a description, a thumbnail address, an upload date, an embed address and a content address. That set is consistent, correct, and entirely about the container.

Here is the part that took me by surprise. Schema.org defines a property for exactly the thing I am missing. On a video object, transcript is documented as “the transcript of that object,” and it has been available the whole time.¹ None of my 60 videos uses it. None uses duration either, which the same vocabulary makes available to every kind of media file.

My schema is doing exactly what I told it to do, which is describe the box.

What about the captions on YouTube?

Real as captions are, they sit in the wrong place. A caption track lives with the video on the player’s own servers and arrives through a separate request when a viewer turns it on, so it is never written into the page a crawler fetched. On this site the gap is wider still: the video only loads when a reader clicks it, so a program fetching the page receives no player at all, let alone its captions.

In the explanations I have read, that distinction usually gets bundled away, with captions and transcripts treated as two words for the same thing. For accessibility they overlap a great deal. For a machine reading your page they are not comparable at all, because one is in the document and the other is somewhere else entirely.

An on-page transcript is the version that counts here. It sits in the article, it gets fetched with everything else, and it is the only form of a video’s spoken content that either a search crawler or an AI answer engine can quote back.

Do I need a video sitemap, then?

Probably not, and being straight about that matters here, because a video sitemap was my first instinct and my first instinct was wrong. A sitemap listing video would tell Google nothing about these pages that my schema and my ordinary sitemap do not already say.

A sitemap is a file listing the addresses on your site so a search engine can find them. There is a video-specific version, and Google’s own description of it is careful: it “is a good way to help Google find and understand the video content on your site, especially content that was recently added or that we might not otherwise discover with our usual crawling mechanisms.”² That is a discovery tool, aimed at video a crawler would otherwise miss.

Its required fields are a title, a description, a thumbnail address and a player or content address.² Compare that to the eight fields my pages already publish in schema and the overlap is total. My video articles also sit in the ordinary sitemap and get crawled like any other page, so there is nothing here that Google would struggle to discover.

One oddity is worth a sentence. My built sitemap declares the video namespace, the bit of boilerplate that says video entries may follow, and then contains not one of them, because the tool that generates it emits that declaration by default. An empty slot, formally announced. It is a curiosity rather than a problem, and I mention it mostly because I spent an hour convinced it was the story.

What does the video content gap cost, and what fixes it?

Video keeps being accepted as an input on surfaces that cannot watch it. Google Merchant Center added a Product videos section this month, letting a merchant upload short product videos into the same feed system that already handles their photos and prices.³

The cost, on my own site, is that every claim I make on camera is unavailable to the systems most likely to be asked about it. When somebody asks an AI assistant a question my video answers well, the assistant has my title, my description, and whatever the surrounding article happens to say. The twelve minutes where I actually work through the thing are not in evidence.

The repair is simple enough that not having done it is mildly embarrassing. Put the spoken content on the page as text, in the article, near the video. Then point the schema at it using the property that already exists for the purpose.

Two things make this worth more than it sounds. A transcript is usually the most quotable material you have, because it is you explaining something in plain speech rather than in written register. And it is the same words twice, which costs nothing to produce once the video exists.

My own sixty articles are still unfixed. I found it this week, the sixty-article backlog is real work, and I would rather publish the audit now than wait until I can present it as a tidy after-picture. If you attach video to your writing, the check takes a minute: open your own article, view the page as text, and see whether anything your video says appears anywhere in it.