Nov 9, 2026

Who Gets Cited in AI Answers, and What Video Creators Can Do About It

7 minute read
Michael Aubry

Ask ChatGPT, Perplexity or Google's AI Mode a question your video answers, and the reply names a few sources. Here is what the companies behind those engines say about how they choose, what independent research has measured, and what you can change on the pages that carry your videos.

Most advice about getting cited by AI reads as if the engines handed out a rulebook. They haven't, and every hour a video team spends chasing a guessed rule comes out of the edit. This guide sticks to what the companies have written down and what has actually been measured up to November 2026.

In short

  1. Keep each engine's search crawler allowed, and block only the training crawlers if you want out of training.
  2. Put the words of your video on its page: a transcript, a plain summary, and the main facts near the top.
  3. Give every figure a date, and give every product, channel, and person the same name each time.
  4. Measure with the reports that exist, and treat a single prompt screenshot as an anecdote.

Dark video editor with a pottery clip in the preview, an answer panel of three source cards with one highlighted in violet and linked to a clip on the timeline, and a transcript track below

Quick answer:

  • Most engines don't publish how they pick sources. Microsoft is the exception, and Google says its AI features run on normal Google Search with no extra requirements.
  • Google lists making important content available in textual form among its best practices for AI features. A video with no words on its page gives an engine very little to quote.
  • No public study has shown that building links causes more AI citations. In the largest dataset reviewed so far, brand mentions across the web tracked AI mentions more closely than links did.

What the engines say about choosing sources

The clearest side by side I've found comes from Jacob Whitmore, who has worked in SEO for around 12 years and now publishes dated studies on how AI is changing search. In How do ChatGPT, Claude, Gemini, Perplexity and Copilot choose their sources? What the companies actually say, he read the documentation from OpenAI, Anthropic, Google, Perplexity, and Microsoft on 8 November 2026. His reading is that only Microsoft had published what makes a page more likely to be chosen as a source.

Bing's guidelines, as he summarises them, favour pages on a single topic, facts stated outright, the important information near the top, and people and organisations named the same way throughout. Google's page on AI features and your website says there are no additional requirements to appear in AI Overviews or AI Mode, beyond a page being indexed and eligible to show in Search with a snippet.

OpenAI, Anthropic, and Perplexity, in Whitmore's reading, document how to keep your site out and say nothing about how they pick what goes in. So when someone tells you ChatGPT prefers a certain kind of page, they're inferring it from outputs, because the company hasn't published a rule.

What the research has measured so far

In Do backlinks get you cited in AI answers? What 16 studies actually measured, Whitmore scored 16 published studies. All 16 were correlations, none was a controlled test, and none published its raw citation data. In the largest dataset, Ahrefs' 75,000 brands, branded web mentions correlated with AI mentions at 0.66 to 0.71, against 0.27 to 0.33 for Domain Rating.

The share of AI citations that come from Google's top 10 runs from about 4% to 76% across the studies he lists, depending on the engine and the date. Google's own AI features overlap with Google rankings far more than any other engine does. One Seer Interactive study he cites found ChatGPT search had an 87% match with Bing's top results and 56% with Google's.

Google's guide to generative AI features says those features can show what's said about products in blogs, videos, and forum discussions, and it warns that chasing inauthentic mentions isn't as helpful as it might seem.

Why a video on its own is hard to cite

Every engine here answers in text and links out to pages. Google's best practices for AI features include making sure important content is available in textual form, then supporting that text with high-quality images and videos where they apply. In that order, the text carries the page and the video backs it up.

Google's generative AI guide adds that these features can bring in relevant images and video, which gives a page with good video more ways to appear than a plain link. Google's video documentation recommends a dedicated watch page for each video where it makes sense, with a title and description unique to that video.

So the practical job is getting the words out of the video and onto the page. You can transcribe a YouTube video, then edit the raw text into a readable summary with the steps in order.

Burned-in captions help viewers, but words inside the pixels aren't page text. Keep the transcript on the page for the engines, and add subtitles to the video on their own track for the people watching.

Write the page the way the engines read it

Whitmore calls Bing's list the only first-party description of how sources get selected. It works as a brief for a video landing page:

  • One topic per page, so an engine has one clear thing to quote.
  • Facts written out. If the video walks through export settings, write the settings on the page too.
  • The answer near the top, before the reader has to press play.
  • The same name for your channel, your product, and the people in it, every time.

Dates matter as well. In Do AI answers say when a "current" figure is from?, Whitmore asked four engines ten UK figure questions on 10 October 2026. Seven of the 40 answers didn't say when the figure was from, and all seven came from Google AI Mode and Gemini. His advice is to put the period next to the number, so an engine can lift the sentence with its date intact. On a video page, write "price as of November 2026" in the description as well as saying it out loud.

Keep the search crawlers in, and measure the right thing

If you've blocked AI bots, check which ones. OpenAI's overview of its crawlers separates OAI-SearchBot, which surfaces sites in ChatGPT search, from GPTBot, which crawls content for model training. Perplexity says PerplexityBot surfaces and links sites in its search results and isn't used to crawl content for AI foundation models.

Whitmore's suggestion is to leave the search crawlers open (OAI-SearchBot, Claude-SearchBot, PerplexityBot, Googlebot, and Bingbot) and block the training ones if you want out of training. He also says to check your CDN's bot settings as well as robots.txt.

Then measure with what exists. Google counts AI Overviews and AI Mode in the Search Console Performance report. Bing Webmaster Tools has an AI Performance report with citation counts and sample grounding queries. ChatGPT tags referral clicks with utm_source=chatgpt.com, so you can filter for it in analytics.

Treat a screenshot of one prompt with care, because AI answers are sampled and the same question can return a different list. In How many times should you run an AI prompt before you trust the number?, Whitmore works out that pinning a brand's appearance rate near 50% to within 10 points takes about 97 runs of one prompt.

Frequently asked questions

Do I need llms.txt or special schema to show up in AI answers?

Not for Google. Its generative AI guide says Google Search doesn't use llms.txt or other AI text files, and that no special schema.org markup is needed.

If I block GPTBot, will ChatGPT stop citing me?

It shouldn't, because GPTBot is OpenAI's training crawler. Showing up in ChatGPT search depends on OAI-SearchBot and on the third-party search providers OpenAI uses.

Are AEO and GEO different from SEO?

Google says work on generative AI search is work on the search experience, so for Google it's still SEO. In Whitmore's reading, OpenAI, Anthropic, and Perplexity haven't said how they pick sources at all.

Where to start

Open the pages for your three most watched videos. Check that the search crawlers can reach them, that a transcript or summary sits on each page, and that the main facts are written out near the top with their dates. Most of that text is what viewers need anyway.

Then get back to the video itself. Our guide to making YouTube Shorts with AI covers the hook, the captions, and the export.

Make this in Wireflow
Motionbox now runs on Wireflow. Type what you want and get motion graphics back. Free to start.
Start free →
Michael Aubry

Founder of Motionbox and Gluely. Building tools for creators.