Video schema generator

How to generate video schema

Three required properties, one date format that trips almost everyone, and two URL fields that are not the page you are on. The generator writes the markup and a caption block from the same values.

1. Title and thumbnail first. Both are required, and both must be unique to this video. A thumbnail is only useful to Google if the URL stays the same next month — a CDN that rotates URLs quietly costs you video indexing.
2. Give the upload time an offset. The date alone is valid; the date with a time and a UTC offset is unambiguous. Without an offset Google reads the timestamp in Googlebot's timezone, which is not yours.
3. Type the runtime however you think of it. Eighteen forty-two, one-oh-two-thirty, a raw second count, or the ISO form. The output is always the ISO 8601 duration the property expects.
4. Point contentUrl at the file, embedUrl at the player. Neither is the page a reader lands on. The generator refuses a contentUrl that looks like a web page, because that is the single most common mistake in video markup.
5. Add chapters and paste both outputs. Chapter links are built from the watch page URL with a time parameter, which is the form Google requires. The JSON-LD goes in the watch page; the caption block goes under the player.

uploadDate: the property that is wrong on most pages that have it

A video schema generator turns a video's details into VideoObject JSON-LD — the structured data Google reads for a video's title, thumbnail, runtime and key moments. Three properties are required: name, thumbnailUrl and uploadDate. The third is where markup goes subtly wrong.

Google's documentation says uploadDate is the date and time the video was first published, in ISO 8601 format, and adds: we recommend that you provide timezone information; otherwise, we will default to the timezone used by Googlebot. A bare 2026-08-14 is accepted. A premiere at 21:00 local time, written without an offset, can be read as a different calendar day — which matters when freshness is the whole point of publishing it.

Three shapes, in increasing order of how much you have told Google:

"uploadDate": "2026-08-14"                    // valid, no time at all
      "uploadDate": "2026-08-14T09:30:00"           // time, but whose clock?
      "uploadDate": "2026-08-14T09:30:00+02:00"     // unambiguous

The same recommendation applies to expires, and Google adds one instruction people ignore: do not supply expires at all if the video does not expire. An empty or far-future expiry is not neutral — it is a claim.

Duration is a duration, contentUrl is a file

duration takes an ISO 8601 duration, not a clock time and not a number of seconds: PT00H30M05S is thirty minutes and five seconds. This form accepts whatever you actually have — 18:42, 1:02:30, 742, or the ISO string — and emits the ISO one. There is no reason for a human to hand-convert this, and every reason for a tool to refuse a value it cannot parse rather than passing it through.

The two URL properties are where video markup most often fails silently:

• contentUrl is the video file's actual content bytes. Google's page says it in bold type: don't link to the page where the video lives. It should end in one of the file types Google can process — mp4, webm, mov, mkv and eighteen others — and it must not be a data URL.
• embedUrl is the player, usually whatever sits in the src of your embed element. Also not the page.
• Google recommends providing contentUrl when you can, because it is the most effective way for it to fetch the video; embedUrl is the fallback.

This generator treats a contentUrl with no file extension, or with an HTML-ish one, as an error rather than a warning. On a page that already validates everywhere else, that one field is usually the reason the video never appears.

Thumbnails and chapters, with the numbers Google actually publishes

Thumbnail rules are concrete enough to check, and rarely checked:

Chapters are the Clip type nested in hasPart, and they carry a rule that most tools skip: the clip URL must point to the same URL path as the video with additional query parameters that specify the time. Give this form your watch page URL and each chapter comes out as that URL plus ?t= and the offset in seconds — not a fragment, not a different host, not a bare page link. The tool also rejects a chapter that starts after the video ends, which is what a mistyped timestamp looks like from Google's side.

One thing markup cannot fix: the video has to be embedded on the page, visible, and not hidden behind other elements. Structured data describes a video that is there; it does not conjure one.

Why there is a second output

Most video generators stop at the JSON-LD. This one also writes a caption block — title, description, publication timestamp, runtime, chapter table — carrying exactly the strings the markup declares.

That is partly the house rule behind every tool here: one source, two outputs, no drift. It is also practical. A watch page whose only description of the video is inside a script tag gives a reader nothing, gives an AI answer engine nothing to quote, and gives you no way to notice when the markup has gone stale except by opening a validator. A visible chapter list with the same timestamps is checkable at a glance.

The same principle, applied to questions and answers, is the FAQ schema generator; applied to a recipe's ingredients and steps, the recipe schema generator, which has the tighter version of this rule because a recipe's marked-up steps are the ones Google shows.