YouTube and GEO: what role does video content play in language model responses

For years, when we talked about video in SEO, we usually referred to a complementary format. A video could enhance a product listing, help explain a service better, increase time on page, or open an additional visibility channel on YouTube.

But that approach is starting to fall short.

With the arrival of AI Overviews, AI Mode, ChatGPT, Perplexity, and other generative response systems, content no longer competes only to appear in a list of results. Increasingly, it competes to be understood, retrieved, cited, or used as support to construct a response.

And in that scenario, YouTube plays a much more important role than might appear at first glance.

It’s important to understand that some models are already fully multimodal and can process video without issue, in addition to images, audio, and text. Others, however, rely on intermediate layers such as transcriptions, subtitles, descriptions, frames, metadata, or textual content associated with the video.

This is precisely why YouTube is so relevant: it converts a complex and difficult-to-process format like video into a piece that is much more readable for machines. A YouTube video is not just an audiovisual file. It’s a title, a description, a transcription, chapters, comments, interaction signals, a thumbnail, a channel, an entity, etc.

And all of that, when properly developed, can carry significant weight in your company’s online findability.

Context: From classic SEO to findability in generative models

At Human Level, we typically use the concept of findability to refer to something that goes beyond traditional organic positioning. It’s not just about appearing on Google for a specific keyword, but about ensuring that a brand, content, entity, or response is easy to find, interpret, and use by the different systems that today mediate between the user and information.

And this is where GEO comes in, or in other words, Generative Engine Optimization.

The underlying idea is not that SEO is dead or that we need to throw away everything we knew. In fact, Google has indicated that to appear as a supporting link in AI Overviews or AI Mode, traditional fundamentals still apply: the page must be crawlable, indexable, and displayable in Google Search with snippets.

The technical requirements at the indexability and WPO level barely change.

And it’s important to note this because it avoids falling into an overly “magical” interpretation of the topic. There is no secret markup for ranking in AI. There is no miracle tag to get a model to cite us. And there is no closed checklist that guarantees appearing in a generated response.

However, what does change is the way the response is constructed.

Google explains that both AI Overviews and AI Mode can use a technique known as query fan-out, which basically consists of launching several related searches on different subtopics and sources to develop a more complete response.

In other words, the system no longer always limits itself to searching a single query and returning an ordered list of URLs, but can now break down the user’s intent into several informational needs and combine signals from different sources.

This significantly changes the way we think about an SEO strategy.

Until now, the most common question in a project was: “Am I in the top 3 for this keyword?” That question remains important, of course. But now another equally or more relevant question appears: “Is my content clear, reliable, and retrievable enough for a generative system to use it as a source or support when someone asks about this topic?”

And it’s at this point where video begins to gain great importance.

Why YouTube is not just a video platform

YouTube is usually seen as a distribution, entertainment, or discovery platform. And it really is. But from a findability perspective, it also functions as a massive structured audiovisual knowledge base in a way that’s quite convenient for search engines and models.

This doesn’t mean YouTube is always the best source, far from it. But its videos have several layers that can help systems understand them.

A well-optimized video on YouTube can include:

• A descriptive title
• A description with context
• Chapters or timestamps
• Manual or automatic subtitles
• Transcription
• Related links
• Engagement signals
• Comments
• Channel information
• View and retention data

YouTube’s own documentation explains that to estimate the relevance of a video in Google Search, factors such as the match of the title, tags, description, and video content with the user’s query are taken into account, in addition to interaction signals such as watch time for a specific search.

However, we should not return to the old obsession of filling in 1,000 tags as if we were in 2012. The important point is different: YouTube needs to understand what the video is about, and for this it uses not only what is seen on screen, but all the context we can provide about that content.

We already know this. An article can be very good, but if it’s poorly structured, doesn’t answer clearly, doesn’t properly mention important entities, or gets lost in digressions, it will be less useful for a model that needs to extract information quickly.

Something similar happens with video, to some extent, but with an added difficulty. If we don’t provide a clear textual and semantic layer, much of its content will remain far more hidden.

That’s why YouTube has an advantage. Not because the video format is “magical”—it’s not—but because YouTube wraps the video with many signals that will help interpret it.

Video as a source for generative responses

As I’ve already mentioned, not all language models or LLMs work the same way. Some can work directly with multimodal content. Others can use transcriptions, fragments, metadata, or results retrieved from search engines. Additionally, we don’t always know exactly which part of the content influenced a specific response.

But we do know something: the more structured, clear, and accessible a piece of content is, the easier it will be for a system to understand and reuse it.

In the case of video, this makes seemingly secondary elements quite important.

Subtitles, for example, are not just an accessibility improvement. They also convert spoken content into processable text. YouTube indicates that transcriptions contain the text of what is said in the video and can include chapters, which helps convert a video into a much more interpretable structure.

The same applies to chapters.

A 20-minute video without structure can be too broad a piece. In contrast, a video divided into clear blocks can function almost like several potential responses. If each chapter addresses a specific question, the system can better understand which part of the content answers which intent.

Google, in its video recommendations, also emphasizes technical issues that remain very common in SEO: stable URLs, dedicated pages for each video when appropriate, or the ability to properly access the file or thumbnail so it can be processed and indexed.

This brings us back to an idea that is sometimes forgotten: findability is not just content. It’s also architecture, crawling, indexing, WPO, and well-aligned signals.

The current situation: YouTube appears increasingly in AI responses

The official documentation tells us that Google is integrating AI experiences into Search, that AI Mode can break down questions into subqueries, and that SEO fundamentals remain necessary. We also know that Google recommends accompanying important content with visual formats when appropriate, but without replacing essential textual information.

Third-party studies, meanwhile, are finding quite interesting patterns.

Ahrefs, for example, states that YouTube is currently the most cited domain in AI Overviews within their data, and that it has grown 34% in the last 6 months. It also indicates, in its research on 75,000 brands, that mentions on YouTube were the factor with the highest correlation with visibility in AI Overviews.

In another analysis, Ahrefs also found that mentions on YouTube showed the strongest correlation with visibility in ChatGPT, AI Mode, and AI Overviews, above other classic factors such as some authority metrics or page volume. The company itself clarifies something important: correlation does not mean causation. And that caution is necessary.

But even with that caution, the data is hard to ignore.

We cannot claim that if you upload videos to YouTube you will appear in AI responses. That would be an absurd simplification. But we can say that YouTube is behaving as a very present source in generative search. And that, for any serious content strategy, should make us think.

The most delicate case appears in sensitive sectors, such as health. A study by SE Ranking analyzed more than 50,000 health queries in Germany and found that YouTube was the most cited domain in AI Overviews for that set of searches, with 4.43% of citations.

This data generated great concern because YouTube is not a medical source in itself and, although there are great communicators and professionals, it remains an open platform where creators without the slightest health qualifications also coexist.

This example is very useful because it shows both sides of the problem. On one hand, it confirms that video can gain significant presence in generative responses. On the other, it demonstrates that visibility and reliability are not the same thing. That content is easy to retrieve does not automatically mean it is the best source.

And in YMYL (Your Money, Your Life) sectors—those that can affect people’s lives—this is especially important.

What this means for your company

If a significant part of your knowledge, product, or service can be better explained through video, not working on that format may make your brand less visible in a growing part of the search ecosystem.

Let’s consider some clear cases:

• A software company that can explain complex functionalities with demonstrations.
• An ecommerce business that can resolve product questions with comparisons or visual guides.
• A clinic that can explain procedures with medical rigor.
• A B2B brand that needs to build trust before the user makes contact.

In all these cases, video is not an ornament. It can be a way to demonstrate knowledge, reduce ambiguity, and reinforce entity.

But be careful, there is an unavoidable condition: the video must be designed as retrievable content.

An improvised video, without structure, with a generic title and a poor description may work for an audience that already knows the brand, but it will be much more difficult for it to appear in searches.

In contrast, a video that answers a specific question, properly names entities, explains the context, divides content into chapters, and is supported by a well-optimized web page will have much better chances of being understood by search engines and generative systems.

Therefore, when recording your videos, you must think about how to turn them into a piece of strategic content.

YouTube or website: the false dilemma

A very common question is whether it’s better to publish the video on YouTube, on my website, or in both places.

And here the most reasonable answer is that we should not frame it as a war between channels.

YouTube has an extremely powerful discovery infrastructure. It’s a platform where people search, learn, compare, and consume content with often very clear intent. Additionally, as we’ve already discussed, it offers a layer of metadata, transcriptions, subtitles, and chapters that facilitates content interpretation.

The website, on the other hand, gives us control. Control over context, conversion, structured data, internal linking, the relationship with other content, lead capture, business, etc.

That’s why, in many cases, the smartest approach will not be to choose between YouTube or the website, but to coordinate both.

For example, we can publish the video on YouTube and, at the same time, create a page or article on our website that:

• Develops the topic in greater depth.
• Includes the embedded video.
• Adds a reviewed transcription.
• Incorporates structured data when appropriate.
• Links to related resources.
• Resolves complementary questions.
• Guides the user toward a conversion.

This way, YouTube can act as an entry point, as an initial touchpoint and discovery channel, while the website can act as a space to organize, expand, and direct that interest toward the business and conversion.

Additionally, it’s important to note that if the video is also published on our website, it should not appear just any way. Google recommends that the video be on a stable URL and, when appropriate, on a page where that video is an important part of the content.

It’s not about embedding it without more, but about accompanying it with context, useful text, related links, and a structure that helps understand what that piece contributes.

This is especially important in large projects. If a company starts working on video seriously, it cannot treat each piece as something isolated. Each video should be part of a content architecture connected with articles, categories, products, services, or related resources.

What elements should be addressed in a video strategy for GEO

If we think about video from a findability perspective, there are several points worth reviewing.

Topic

Not everything deserves a video. There is content that is better resolved with text, a table, an image, a tool, or a technical sheet. But there are topics where video has a natural advantage: demonstrations, tutorials, product analysis, visual explanations, processes, comparisons, expert interviews, training…

Structure

A video for GEO should not simply be a disorganized conversation. It can have a natural tone, of course, but it needs a clear structure: introduction, context, development, examples, conclusions, and, if appropriate, well-marked chapters.

Language

There’s an important detail to consider here. Models work very well with explicit text. That is, if in the video we talk about “this,” “that,” “the tool,” or “the problem” without properly naming entities, we’re losing context. It’s advisable to clearly mention the topic, brand, product, category, and questions being resolved.

Textual layer

Automatic transcription can be useful, but it’s not always sufficient. YouTube allows working with transcriptions and subtitles and also indicates that translated titles and descriptions can appear in searches in other languages when the video has those translations.

For international projects, taking care of translations allows us to adapt the video’s findability layers to the markets where we truly want to compete.

Connection with the website

A video that doesn’t link to anything, that isn’t connected to a relevant page, and that isn’t part of a broader content strategy falls short. It may get views, yes, but it may not help as much to build authority, capture, or business.

Possible consequences for SEO strategies

If this trend consolidates, there will be several clear consequences.

The first is that video will stop being seen as a simple extra in many content strategies to become a fundamental pillar.

Of course, not in all. But in those where the user needs, for example, to see, compare, understand a process, or gain confidence before making a decision. In these cases, not having video content may start to be a disadvantage, no longer just on YouTube, but also in generative environments.

The second is that SEO teams will need to work more closely with content, video, brand, and product teams.

Optimizing a video for GEO does not consist of asking someone to put a keyword in the title at the end of the process. Optimization starts much earlier: in choosing the topic, in the script, in how entities are named, in what questions are answered, in how the piece is structured, and in how it connects with the website.

Brand authority will be increasingly distributed.

In addition to links or mentions in articles, we now have other ways to achieve authority for our brand: appearances in videos, interviews, tutorials, transcriptions, podcasts, forums, comparisons, and other contexts that systems can process.

Finally, there will be more competition to become a reference.

In classic SEO, we often competed for position. In GEO, we also compete to be a sufficiently clear and expert reference to be part of a response. And that requires something more than publishing by volume.

Ahrefs, in fact, found a very weak correlation between page volume and AI visibility, which reinforces the idea that it’s not simply about creating more content, but also about obtaining relevant mentions and signals.

So, should we make more videos?

As a good SEO professional, there’s only one possible answer: it depends. 😉

You should make more videos if video helps explain better, demonstrate better, generate more trust, or better resolve a search intent.

You should make them if the brand has something relevant to show or explain.

You should make them if they are part of a strategy connected with the website, entity, authority, and business.

But you should not make video because of trends.

A bad video does not become strategic just by being on YouTube. Just as a bad article does not become useful just by being indexed.

The key is to think of video as a piece of knowledge.

Basically, if the content responds to a real need, if it’s well structured, if it has a reviewed textual layer, if it properly names important entities, and if it integrates with the rest of the brand’s digital assets, it can indeed become a very valuable source for SEO and GEO.

And YouTube is a platform that, in addition to allowing us to distribute our content easily, will help us have all the necessary information as complete as possible.

What all this teaches us

We’re now reaching the end of this article and, if you’ve made it this far, you probably already have a clear idea of the major role YouTube plays in the new landscape that AI-related searches have in store for us. But let’s analyze some conclusions.

The main lesson is that video content will not replace written content, but it is gaining increasing relevance.

Text remains fundamental. It remains the most direct way to explain and make information accessible. Additionally, we’ve already seen that search systems themselves still need textual content, clearly presented, to understand many pieces.

But video is no longer just a format for capturing attention. It’s also starting to behave as a source. A source that can help models better understand a product, brand, explanation, demonstration, or entity.

Until now, many companies wondered whether it was worth making videos because “YouTube doesn’t convert like the website” or because “our audience isn’t there to buy.”

However, if LLMs, AI Overviews, or AI Mode are increasingly using video sources from YouTube, we’re leaving out an important part of our authority if we don’t work on them.

Therefore, if I had to summarize it in one idea, it would be that YouTube should no longer be seen solely as a distribution channel, but as another layer within the findability strategy.

It’s not enough to upload videos. They must be made understandable, crawlable, contextualized, and connected with the website.

Because in GEO, the winner is not necessarily the one who publishes the most, but the one who manages to make their knowledge easier to understand, retrieve, and cite.