How to get your videos cited by ChatGPT, Claude, Gemini, and Perplexity
There is no rank inside an AI assistant. There is a chain: a page you published, that an engine read, that it chose as a source. The seven-step routine for making the first two true for every video, measuring the third honestly, and reading "not yet" correctly.
Somebody asks ChatGPT a question you have answered on camera a dozen times. It answers confidently, cites three pages, and none of them is yours. That is the moment most people first notice the new discoverability problem, and the first instinct is to ask how to “rank” in an AI assistant. There is no rank. There is a short chain of things that have to be true before an engine can cite you, and a routine for making them true, one video at a time. This is that routine, in order, with the honest limits marked.
The chain: publish, read, cited
An assistant that searches the web before answering does three things with your content. It has to be able to read a page you published, and then it has to choose that page as a source when it composes its answer. You fully control the first link. You can see evidence of the second. The third is the engine's decision, shaped by whether your page answers the question better than the others it read. Every step below strengthens one link in that chain. None of them can force the last one, and anyone who says otherwise is describing the part they do not control. If you want the background on why this replaced ranking, the AEO primer covers it.
You control whether you can be cited. The engine decides whether you are.
Record videos that answer a question someone would type
The engine matches questions to answers. A video titled “Q3 update” that wanders across six topics gives it nothing to match. A video that exists to answer “how long does implementation take for a team of ten?” and answers it in the first ninety seconds gives it everything. Before you press record, write the question down. Say it out loud near the start. Answer it plainly, then elaborate. The ninety-second script structure works for this because it forces the answer forward. Very short clips and music-only videos have nothing to extract from; substance is the input.
Make sure the video is findable
Nothing downstream happens for a link-only video. In StreamAgent the switch is on each video's Discoverability tab, “Findable in search & AI,” and it is on by default unless your workspace default says otherwise. Two things silently stop the whole chain: a video that was link-only when its captions finished, and the workspace-level “Allow AI assistants to read published videos” switch being off. Check both before you check anything else. A check on a link-only video is refused on purpose, because an engine cannot cite a page it is told not to read.
Let the questions publish, then read them
When captions finish on a findable video, the questions the video answers are extracted from the transcript and published without a click: up to eight, written the way a person would type them into a search box, each with a short answer and the timestamp where it is said, ordered by where they occur in the video. They appear as an FAQ on the watch page, on the readable transcript page, and in the AI guide file. Open the Discoverability tab and read them as an editor would. Are these the questions a buyer actually asks? Is the answer the one you would give? If the transcript was thin, the card will tell you it did not yield publishable questions, and the fix is a better video, not a better button.
Fix the spelling of your own name
Speech-to-text splits and mangles brand names. The extracted questions do not trust the transcript for names; they spell brand and product names the way the video's title and description spell them. So put the correct spelling in the title or description of every video, then press Re-run on the Q&A card. Re-running when nothing changed costs nothing and does nothing; it only does new work when the transcript or the text it reads from has actually changed. A citation attributed to a misspelling of you is a citation you did not get.
Wait for the reads, and know what they mean
Engines take days to weeks to discover, fetch, and digest a new page. The first evidence you will see is not a citation but a read: the Discoverability tab's status cards flip from a capability statement to “Crawled · 2 days ago” when a named crawler has actually fetched the video's pages, and the Read by AI section on Analytics rolls that up across the library. Two honesty rules make the number worth trusting. A crawler's claimed identity is verified against where the fetch really came from, so a faked header records nothing. And only the crawlers that fetch pages to answer a user count as an assistant read; training-data collection is listed separately under the company's name and never inflates the count. A read is not a view and never touches your analytics. It is infrastructure evidence that the door is open.
Ask the engines yourself
The only way to know whether you are cited is to ask the question and look at the sources. StreamAgent does this for you: a citation check asks ChatGPT, Claude, Gemini, and Perplexity the video's top published question through each engine's own interface with web search on, and records whether any source the engine cites in its answer is your page. Three verdicts, kept honest. “Cited this video as a source” means that exact page. “Cited your site as a source” means one of your other pages, another video or your channel, which still counts as you. “Not citing you yet” means the engine answered without naming you. A page the engine merely looked at while searching and then ignored is never counted as a citation.
On every paid plan a check runs automatically each week, rotating through the library, never-checked videos newest first, so every video gets a verdict without anyone pressing anything. Pro adds the “Run a check” button for any video, any question, right now. Results collect in the Cited by AI section on Analytics with cited videos sorted to the top.
Read “not yet” correctly, then publish the next one
“Not citing you yet” is the normal starting state and the correct baseline. It means one of three things: the engine has not digested the page yet, the engine found a page that answered the question better, or the question is one nobody is asking the engine in that form. The first resolves with time. The second and third resolve with the next video. Each findable video is another set of questions an engine can match, so the strategy that actually moves the needle is boring: keep recording videos that answer real questions in your niche, keep them findable, and let the weekly check tell you when a verdict flips. When it does, the card and the analytics section update on their own.
What being cited is for
A citation sends a stranger with a question to a page you control, at the exact moment they were asking it. That page should not be a dead end. In StreamAgent it is a watch page with a player that can ask who they are, score the answer, and offer a call, and the AI guide file even tells an assistant how to book one on the visitor's behalf. Getting cited is how the video gets itself found. Everything after is what it does with the person it found.
Publish so you can be read. Check whether you were. Then record the next answer.
One platform, seven layers
This article covered one slice. The machine ships whole: record it, get it found, arm it, test it, rank the leads, train the ads, and ask your AI how it is going.
Common questions
- How do AI assistants decide which sources to cite?
- An assistant with web search on reads a handful of pages it can access, composes one answer, and names the pages it drew on. It can only cite a page it was able to read, and it picks the pages that answered the question most directly. You control whether your page is readable and answers the question; the engine decides whether to choose it.
- How long does it take for ChatGPT or Perplexity to cite a new video?
- Days to weeks. Engines have to discover, fetch, and digest a new page before it can appear as a source. The first evidence is usually a verified crawler read of the page; citations follow later, and some never arrive because another page answered the question better. A video published this week showing "not cited" is expected.
- Can a video platform guarantee that AI assistants will cite my videos?
- No. A platform can make every video citable without setup, show you which engines actually read the pages, and tell you when a citation arrives. It cannot make an engine choose your page over another. The part you control is recording videos that answer real questions, keeping them findable, and publishing the questions they answer.
- How do I check whether an AI assistant cites my video?
- Ask the engine the question your video answers, with web search on, and look at the sources it names. StreamAgent automates this: a weekly check asks ChatGPT, Claude, Gemini, and Perplexity each video's top published question and records whether any cited source is your page, with on-demand checks on Pro. Only pages the engine actually cites in its answer count.