Key Takeaways
- Assume the video itself is never read, only the words attached to it. An assistant reaches a video the way it reaches any page: through text. Your captions, title, description and chapters are the video as far as it is concerned.
- Publish a real transcript rather than trusting automatic captions. YouTube generates captions with speech recognition, and Google says accuracy varies with audio quality, accents and background noise. A transcript you control is the version you want quoted.
- Say the specifics out loud, because only spoken words become text. The price on your slide and the result you point at on screen leave no trace. If it matters, say it in a sentence a stranger could repeat.
- Put the video on a page of your own, not only on YouTube. A watch page you control can carry the transcript, the specifics and the schema markup around the video, which gives the same content a second and better address.
- Expect video to support a citation rather than win one alone. Text pages remain the easier thing to lift. A video earns its keep by proving a person and a practice exist behind the claims, which a page alone cannot do.
You have forty videos. They took real time to make, they explain the work better than your website does, and an assistant asked about your field has apparently never heard of any of them.
The reason is less about quality than about format. Something has to turn your video into words before anything else can happen, and that step either happened well, happened badly, or did not happen at all.
This post covers what actually travels out of a video, why automatic captions matter more than they should, and what to do with the library you already have.
Do YouTube videos get cited by AI?
Sometimes, and nearly always by way of their text. The route runs through captions, title, description and the page the video sits on.
This is the same constraint that governs every other format. An assistant works with what exists as words, which is why a slide deck and a scanned PDF behave the way they do in whether AI assistants read PDFs and slides. Video is one more container. What matters is whether the contents were ever written down.
So the practical question is not whether your video is good. It is whether a stranger reading only the text attached to it would learn what you charge, who you work with, and what happens next.
What an assistant actually reads
Four things: the transcript, the title, the description and the chapters. Two things it does not read: what is printed on your slides and what you point at on camera.
That split is the whole post, and it catches people out because the two invisible items are usually where the good stuff is. The number is on the slide. The before-and-after is on screen. Neither leaves the video.
A filled track is a part of the video that exists as words somewhere. An empty dashed track is a part that never becomes text, however good it is on screen.
Google's own video SEO best practices describe the same requirement from the other side. To use a video at all, it has to be able to find it, fetch it and understand what it is about, which in practice means a page it can reach and text it can read.
Why automatic captions decide how you are understood
Because for most videos, the automatic caption track is the only text version of what you said, and it is generated by speech recognition rather than by a person.
Google's help documentation on automatic captioning is candid about the limits: captions are produced by machine learning, they are not available for every video or every language, and accuracy varies with audio quality, accents, background noise and how clearly people speak.
Read that list against what you say in a video. Business names, place names, prices and specialist terms are precisely the words that get mangled, because they are the words a general speech model has seen least. The sentence you most wanted quoted is the one most likely to come out wrong.
The fix is unglamorous and permanent. Upload a corrected transcript, and publish the same text on a page of your own beside the embedded video. The schema.org VideoObject type includes a transcript property, so there is a defined place to hand the words over rather than hoping they are inferred.
Where video helps and where a page still wins
Video wins at proving a person exists and works the way they claim. Pages still win at being quoted, because a page is already the thing an assistant is built to lift from.
It is worth being honest about this rather than selling video as a shortcut. If your goal this quarter is to be named in an answer, a page that states a price plainly will get you there faster than ten more videos. Video earns its place in the second half of the buyer's decision, when they have your name and want to know whether they like you.
| Part of the video | Exists as text | How it gets there | What it is worth |
|---|---|---|---|
| The words you say | Yes, if captions or a transcript exist | Automatic captions, or a transcript you upload | The only part long enough to quote |
| The title | Yes | You wrote it | Carries the subject in the buyer's own words |
| The description | Yes | You wrote it | Usually the only place a price or a city appears |
| Chapter markers | Yes | Timestamps you added | Turns one long answer into several findable ones |
| Text on your slides | No | Nothing carries it out | Retype it into the page or lose it |
| What you show on camera | No | Nothing carries it out | Describe it out loud or it did not happen |
One more thing worth knowing. Forums and discussion pages are quoted heavily because they are dense with plain, specific sentences, which is the argument in why AI cites Reddit so often. A transcript has the same quality when you speak plainly, and none of it when you speak in generalities.
What to do with the videos you already have
Pick your three best videos, the ones that answer a question a buyer actually asks. Read their automatic captions. Fix the places where your business name, your city or your prices came out wrong, and upload the corrected track.
Then give each of the three a page on your own site: the embedded video, the corrected transcript underneath it, and one short paragraph at the top that answers the question in plain words. That page is now something that can be found and quoted on its own terms, which is the point covered in what kinds of pages AI assistants quote most.
Then leave the rest of the library alone. Forty transcripts is a project. Three is an afternoon, and three is enough to find out whether it changes anything.
If you would rather have the first reading done for you, the free check asks ChatGPT 25 real buyer questions from your niche at $0 and emails a plain-English readout: your score, who got named in your place, and the one fix to start with. It runs as soon as you ask for it, and the report lands within the hour between 7am and 5pm Central, or by 10am the next morning outside those hours.
Questions we hear the most
Do YouTube videos get cited by AI?
Sometimes, and almost always through their text. Assistants reach a video by way of its captions, title, description and surrounding page, so a video with no transcript and a vague description has very little for anything to quote.
Does the AI actually watch my video?
Treat the answer as no. The practical route from your video to an answer runs through words: the transcript, the metadata you wrote and the page the video sits on. Plan for the text version to be the version that travels.
Are automatic captions good enough?
They are better than nothing and worse than a transcript. Google says automatic caption accuracy varies with audio quality, accents, background noise and how clearly people speak, so anything specific is exactly what tends to be garbled.
Should I put the transcript on my own website?
Yes, on a page that holds the embedded video. That gives the same words an address you control, lets you add a price or a location beside them, and turns one video into a page that can be quoted on its own.
Does video schema markup help?
It helps machines understand what the video is and where it sits. Google documents VideoObject markup for its video features, and the schema.org type includes a transcript property, which is a direct place to hand over the words.
Is a long video better than a short one?
Length is not the variable. Specifics are. A four minute video that states a price, a place and a timeframe gives more to quote than an hour of general advice with none of those in it.
How do I find out whether AI is naming me right now?
The AnswerHalo free check asks ChatGPT 25 real buyer questions from your niche and emails a plain-English report: your score, who got named in your place, and the one fix to start with. It costs $0 and it runs as soon as you ask for it.
Does the audit look at my video content too?
It looks at what the assistants surfaced about you, whatever the format. The $500 audit runs 125 checks across five AI tools and records what got named and what got cited on each, then ends with the page that wins your worst question.
Your buyers are already asking. Find out what AI tells them.
The free check asks ChatGPT 25 real buyer questions about your niche and sends you the report within the hour, at $0. If it shows you are already getting named everywhere, we will say so in plain words.
Get my free check →