AI Video SEO: What AI Engines Can Read on Your Video Pages

An AI engine answers from text. When ChatGPT search, Claude, Perplexity or Google's AI features use your page, they work from words they can extract from it. For a video page, those words come from the HTML around the player: the title, the description, the transcript, captions written into the page, and any facts you state in body copy.

The video file is a different matter. No crawler documentation in our evidence base describes an AI retrieval crawler watching, transcribing or weighting video content. That doesn't prove it never happens. It means you can't plan around it, and you shouldn't buy a service that claims to know how engines weight video.

What this means in practice: treat every video page as a text page that happens to contain a player. If a fact is only spoken in the video, assume an AI engine can't quote it. If the same fact is written in the page's HTML, the engine at least has a sentence it can use.

It helps to split AI visibility into four stages:

• Parametric: what a model knows from training, frozen at a cutoff. Nothing you publish this quarter changes it.
• Access: whether a crawler may fetch and store your page.
• Retrieval: whether your page enters the model's context for a given question.
• Generation: which retrieved pages the model uses and cites.

Video pages usually fail at the access stage, and they fail quietly. Many players, embeds and transcript widgets inject their text with JavaScript after the page loads. Vercel's crawler study found that ChatGPT and Claude crawlers fetch JavaScript files but don't execute them, so they can't read client-side rendered content. Later independent testing, as of June 2026, found the same for OAI-SearchBot, Claude-SearchBot, PerplexityBot and others. Googlebot is the main exception because it renders pages.

So a transcript that appears in the browser can still be missing from the raw HTML that ChatGPT's or Perplexity's crawler receives. The same problem hits sites built on client-side frameworks, which we cover in whether ChatGPT can read a React website. It is also the same logic as text baked into an image, explained in whether AI can read text inside images.

The table below shows the common places video text sits on a page and what the evidence says about AI crawlers reaching it. "Non-Google AI crawlers" means the retrieval and training bots from OpenAI, Anthropic, Perplexity and Meta covered by the rendering tests.

The pattern is simple. Text in your own server-rendered HTML is the only row where you have both control and evidence. Everything else is either blocked by rendering or undocumented.

If your videos live on YouTube, the YouTube page is YouTube's page. You control its title and description, not how it is crawled. Our guide to YouTube SEO with AI tools covers that side. This page is about the video pages on your own site.

In your control

• Rendering. Whether the transcript and description exist in the HTML before JavaScript runs.
• Crawler permission. robots.txt rules are per bot. OpenAI's crawler documentation says each setting is independent: blocking GPTBot stops training use, while OAI-SearchBot controls whether you appear in ChatGPT search answers.
• Network reachability. A firewall or CDN can block an AI crawler that robots.txt allows. OpenAI says eligibility needs both robots.txt permission and a host or CDN that allows its published searchbot IP ranges.
• Snippet controls. noindex, nosnippet, max-snippet:0 and data-nosnippet keep content out of AI answers on Google and Bing. A transcript wrapped in data-nosnippet by a theme is excluded from AI summaries.
• Google eligibility. Google says a page must be indexed and eligible to show with a snippet to be a supporting link in AI Overviews or AI Mode, with no additional technical requirements.
• Stated facts. A specific fact written on a fetchable page gives an engine a sentence it can quote.

Not in your control

• Whether an engine searches the web at all for a given question. Retrieval is discretionary, per query, on every platform.
• Which queries the engine actually runs. ChatGPT rewrites prompts into its own search queries.
• Whether your page is retrieved, and whether it is cited once retrieved.
• What a model already believes from training data.

The honest goal is to remove every block in the first list. Nobody can promise the outcome in the second.

Work from access outward. Each step only matters if the one before it passes.

1. Check crawler access. Read your robots.txt per bot. Confirm OAI-SearchBot, PerplexityBot and Claude-SearchBot are allowed if you want to appear in those engines' search answers. Then confirm your firewall or CDN doesn't return a 403 or a challenge page to them. robots.txt alone doesn't prove reachability.
2. Check the raw HTML. Fetch a video page with JavaScript disabled, or view source. Search for a sentence from the transcript. If it's missing, the non-Google AI crawlers probably can't see it either.
3. Server-render the transcript. Put the full transcript in the page HTML, not in a script-loaded widget or a collapsed panel that fetches text on click. A plain heading such as "Transcript" followed by paragraphs is enough.
4. Write a real description. Two or three sentences that say what the video covers, who it's for, and the main answer it gives. Avoid descriptions that only say "Watch our latest video."
5. State the key facts in body copy. If the video gives a price, a spec, a step count, an opening time or a definition, write that fact as one clear sentence on the page.
6. Check snippet directives. Look for noindex, nosnippet, max-snippet and data-nosnippet on video templates. Remove any that were applied by accident.
7. Confirm Google indexing. Use Search Console to check that the video pages are indexed and not excluded.

CMS choices shape steps 2 and 3. If you run WordPress, the theme and player plugin decide how transcripts render, and our WordPress AI SEO guide covers the settings to check. Decoupled setups have their own risks, covered in whether a headless CMS hides your site from AI crawlers.

• Relying on the video to carry the content. A page with a player, a title and nothing else gives an AI engine almost nothing to quote.
• Transcripts behind JavaScript. Interactive transcript widgets look good to visitors and can be empty to most AI crawlers. Keep a server-rendered copy.
• Blocking the wrong bot. Blocking all AI crawlers to stop training also blocks search bots such as OAI-SearchBot. Allowing GPTBot doesn't make ChatGPT search find you either, because GPTBot governs training only.
• Trusting robots.txt alone. A bot-management rule on your CDN can turn away a crawler that robots.txt allows, with no warning in any dashboard.
• Adding schema and expecting citations. VideoObject and other structured data describe a page, but Google says structured data isn't required for generative AI search. Ahrefs' matched-control study of 1,885 pages found adding schema produced no major uplift in AI citations on any platform, and AI Overviews citations fell 4.6%. Existing markup does no harm. It just isn't a lever for AI answers.
• Rewriting transcripts "for AI". Google's optimization guide says you don't need to write in a specific way for generative AI search. The best-designed lab studies found citation-oriented rewrites ineffective or harmful, as we explain in lab GEO studies vs live engines.
• Publishing an llms.txt for video. Google says Search ignores it, and no other engine documents reading it.

Start with one important video page and test it the way a crawler would. Fetch it as each AI crawler and record the status code. A 200 means the bot got through. A 403, a 429 or a challenge page means it didn't. The AI Crawler Checker requests a page as each AI crawler and shows which ones your firewall turns away.

Next, compare the raw HTML with what you see in the browser. Every sentence that only appears after JavaScript runs is a sentence most AI crawlers miss.

Then check the answer side. Ask the questions your video answers and see whether the engines name you or quote your facts. Read any single number carefully, because answers given with and without live web search get mixed together. A free GEO audit walks through access first and accuracy second, which is the right order for video pages too.

Can ChatGPT watch the videos on my website? OpenAI's crawler documentation describes what each bot is for, not how it handles video files. Nothing in the evidence we rely on shows ChatGPT's search crawler processing video content from your site. Plan as if it can't. Put the transcript, a clear description and the key facts in the page's server-rendered HTML, where the crawler can read them as text.

Does VideoObject schema help my videos appear in AI answers? There is no evidence that it does. Google says structured data isn't required for generative AI search. Ahrefs tracked 1,885 pages that added schema against matched controls and found no major uplift in AI citations, with AI Overviews citations falling 4.6%. If you already have VideoObject markup, leaving it in place does no harm. Just don't treat it as a route into AI answers.

Is an embedded YouTube video enough for AI visibility? Usually not. An embed shows YouTube's player inside an iframe, and the text inside belongs to YouTube's page, not yours. An AI crawler reading your page sees your HTML, which may contain little more than the embed code. Add a written description, the full transcript and the main facts to your own page so it stands on its own as text.

Should I put the transcript on the same page or a separate page? The evidence doesn't favor one layout. What matters is that the transcript sits in HTML that crawlers can fetch without running JavaScript, and that the page is indexable and not marked nosnippet. Many sites keep it on the same page under a plain heading, which keeps the facts next to the video. Google says there is no ideal page length, so a long transcript isn't a problem.

Do AI crawlers read caption files like .vtt? No AI vendor in our evidence documents whether its crawlers fetch caption files referenced by a video player. Some players load captions through JavaScript, which most non-Google AI crawlers don't execute. Captions still matter for viewers and accessibility. For AI readability, publish the same text as a transcript in the page's HTML instead of relying on the caption file alone.