A product video that AI cites is a video a model can find, watch in words, and quote with confidence. The video itself shows the product doing its job, the page around it says in plain words what the video shows, and the markup underneath names the video, describes it, and points to the file and the thumbnail. No vendor publishes a formula for how much a video moves an AI citation, and this guide does not invent one.
A video helps in two ways, and they are easy to confuse. The first way is human: a shopper watches the product work and trusts it more. The second way is machine reading: a model reads the title, the description, the transcript, and the markup, and uses those words to answer a question about the product. A beautiful video with no words around it helps the first way and does almost nothing for the second.
Where a product video fails
Most product videos fail for AI reading in the same few ways, and none of them look like failures to the person who made the video.
- The video lives only on a social feed. The video is posted to a social platform and never placed on the product page or on a page of its own, so a crawler that reads your site never meets it, and the words that describe it live on someone else's platform.
- The video is embedded but the page says nothing about it. The player sits on the product page with no heading, no summary, and no transcript, so a model that reads the page text finds a player and no words to quote.
- The title is the product name and nothing else. Every video on the site carries the same short product name as its title, so a model cannot tell the unboxing video from the setup video from the care video, and neither can a shopper.
- The thumbnail is blocked or missing. The thumbnail sits behind a login, is blocked in robots.txt, or is served in a way a crawler cannot fetch, so the video has no image to show in results.
- The markup does not match the page. The structured data names a video that is not on the page, or describes a different product, or gives a duration that does not match the file, so a system that checks the markup against the page loses trust in both.
- The video is too short to say anything. The clip runs for a few seconds and shows the product from one angle, so there is no spoken explanation, no step shown, and no sentence a model could quote.
- The key moments are left to chance. The video shows three distinct steps, fitting, charging, and cleaning, and nothing on the page or in the markup names those steps, so a model that wants to answer one step cannot point to it.
The steps
Step 1. Give each video one job and one watch page. The marketer picks one question the video answers, such as how to fit the product, how to charge it, or how to clean it, and writes that question down in plain words. The developer puts the video on a page where a visitor can watch it, which Google calls a watch page, and links that page from the product page. Do not point a video at a page where a visitor cannot watch it. Google's video structured data guidance says VideoObject markup must be added to a page where users can watch the video.
Step 2. Write the title as a specific promise. The marketer writes a title that names the product and the job, such as how to fit the rechargeable bike light on a handlebar, instead of repeating the product name alone. Google's guidance says to use unique text in the name property for each video on your site, so two videos about the same product need two different titles. Keep the title short enough to read at a glance, and use the same title in the visible heading, in the video description, and in the markup. A title that promises a job tells a model what question the video answers.
Step 3. Say the same facts in words on the page. The marketer writes a short summary under the video, in plain sentences, that says what the video shows, in what order, and what the viewer will know at the end. List the steps in the order they appear in the video, name the parts shown, and state the result. Keep the summary close to the player, not buried at the bottom of the page. A model that reads the page needs words it can quote, and the summary is the quotable version of the video.
Step 4. Add a transcript and captions. The marketer writes out what is said in the video, word for word, and puts the transcript on the watch page or links to it from the page. The developer adds captions to the video file or the player so the spoken words exist as text. A transcript turns a spoken demonstration into readable text, which is what a model can actually cite. Do not write a transcript that says something different from the video.
Step 5. Make the thumbnail and the file crawlable. The developer checks that the thumbnail image is reachable by a crawler, is not blocked in robots.txt, does not need a login, and is a stable address that will not change. The developer also checks that the video file or the embedded player is reachable in the same way. Google's video guidance lists the thumbnail as a required property, and a thumbnail a crawler cannot fetch leaves the video without an image in results. Use a thumbnail that shows the product in use, not a logo and not a blank frame. One thumbnail per video, and a different thumbnail for each video.
Step 6. Mark the key moments by name. The marketer watches the video and writes down the moments that matter, such as fitting the mount, charging the light, and switching modes, with the time each moment starts. The developer adds those moments as Clip entries in the VideoObject markup, with a name, a start offset, and a URL that points to that moment, or adds timestamps to the video description on the hosting platform where that is supported. Google's video guidance says key moments can be set by the publisher through Clip structured data, and that Google will prioritize key moments set by the publisher. Name each moment in the same words the page uses for that step.
Step 7. Check the page the way a crawler sees it. The developer opens the watch page, checks that the video, the title, the summary, and the transcript are present, and runs the page through a structured data testing tool to check the markup. The marketer reads the summary and the transcript again and fixes any place where the words disagree with the product page, such as a different weight or a different item in the box.
The machine-readable piece
This is VideoObject markup for a hypothetical product video. Swap the placeholders in square brackets for your own values. The name is the specific title from Step 2, the description is the short summary from Step 3, and the thumbnailUrl points to the thumbnail from Step 5. The uploadDate placeholder must be replaced with the real publication time in ISO 8601 format, with timezone information. The duration uses ISO 8601 duration format, so a video of two minutes and thirty seconds is written as PT2M30S. The hasPart entries are the key moments from Step 6, with startOffset and endOffset in seconds. Only include a view count if it is a real count from your own records. Only include a contentUrl or an embedUrl that points to the video a visitor can actually watch on this page.
{
"@context": "https://schema.org",
"@type": "VideoObject",
"name": "How to fit the rechargeable bike light on a handlebar",
"description": "This video shows how to fit the rechargeable bike light on a standard handlebar, charge it with the included cable, and switch between steady and flashing modes.",
"thumbnailUrl": "https://[yourdomain.com]/images/bike-light-fitting-thumbnail.jpg",
"uploadDate": "[publication date and time in ISO 8601 format, with timezone]",
"duration": "PT2M30S",
"contentUrl": "https://[yourdomain.com]/videos/bike-light-fitting.mp4",
"embedUrl": "https://[yourdomain.com]/videos/bike-light-fitting",
"hasPart": [
{
"@type": "Clip",
"name": "Fit the mount on the handlebar",
"startOffset": 10,
"endOffset": 55,
"url": "https://[yourdomain.com]/videos/bike-light-fitting?t=10"
},
{
"@type": "Clip",
"name": "Charge the light",
"startOffset": 56,
"endOffset": 95,
"url": "https://[yourdomain.com]/videos/bike-light-fitting?t=56"
},
{
"@type": "Clip",
"name": "Switch between steady and flashing modes",
"startOffset": 96,
"endOffset": 150,
"url": "https://[yourdomain.com]/videos/bike-light-fitting?t=96"
}
]
}
Google lists name, thumbnailUrl, and uploadDate as the required VideoObject properties, and lists description, duration, contentUrl, embedUrl, and hasPart with Clip entries among the recommended properties. Adding the markup makes the video eligible for video features in search results. It does not guarantee the video will be shown.
What it looks like on the page
On the site, a shopper sees a product page with a video player near the top, a title that names the job the video does, a short summary under the player that lists the steps in order, and a transcript link beside the player. Under the surface, the same title, description, thumbnail, and key moments sit in the VideoObject markup, and the thumbnail and the video file are reachable without a login. A shopper who cannot watch the video can still read the summary and know how to fit the product.
Say you run a hypothetical shop that sells a rechargeable bike light. You make a short video that shows fitting the mount on a handlebar, charging the light, and switching between steady and flashing modes. You put the video on the product page, write a title that names the fitting job, add a summary that lists the three steps in order, and publish a transcript that says what the presenter says at each step. You add VideoObject markup with the same title and summary, a thumbnail that shows the light on a handlebar, and three Clip entries, one for each step. A shopper asks an assistant how to fit the light on a handlebar. The assistant can read the summary, the transcript, and the markup, find the fitting step by name, and cite your page instead of a generic video about bike lights.
The fundamentals
These mechanics are shared across every site and written in full in their own guides, so this guide points to them instead of repeating them:
- •How to write the llms.txt file for your site, in our llms.txt guide
- •How to add the Product markup AI systems read, in our product schema guide
- •How to write the crawler rules for AI systems, in our robots.txt guide for AI search
What this guide does not cover
This guide covers making one product video readable and quotable: the watch page, the title, the summary, the transcript, the thumbnail, the key moments, and the VideoObject markup. It does not cover video production itself, such as lighting, editing, or sound, and it does not cover paid video distribution or social video strategy. It does not cover the transaction layer that lets an agent complete a purchase.
Check your own site
Run the free scan at aizii.ai/aeo-scan. One click, no signup, and it checks whether AI crawlers can read your site today.
Run the free scanSources
- Google Search Central, Video structured data (specific): https://developers.google.com/search/docs/appearance/structured-data/video
- Google Search Central, Optimizing your website for generative AI features on Google Search (specific): https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
- Google Search Central, AI features and your website (specific): https://developers.google.com/search/docs/appearance/ai-features
- Schema.org, VideoObject type (specific): https://schema.org/VideoObject
- Schema.org, Clip type (general docs): https://schema.org/Clip
Note: links marked specific point to the page that supports the claim in this guide. Links marked general docs point to the vendor's general docs section for that topic, not to a page that proves every step. Check the vendor docs for the current path before relying on a menu name or click path in this guide.
