Keywords Aren’t Enough. Start Using Vectors to Rank on Google and Other AI Engines

In this video, we're diving into the future of SEO. We're moving beyond the world of simple keyword matching into something far more sophisticated. For years, SEO was about strings or literal text matches. Search engines would ask, does this page contain the exact words that somebody is searching for? Now that's a very brief discussion, but you get the idea. We saw this beginning with Hummingbird, where we moved away from strings to things, where Google understands entities, relationships, and context. But today, we're taking the next step. Search engines and large language models can process and understand meaning through vector embeddings. This isn't theoretical. It's the foundation of how modern search engines work right now. So here's the problem. Traditional keyword analysis is becoming insufficient. When somebody searches best smartphone twenty twenty five, they're not just looking for those exact words. They want a comprehensive comparison of things like battery life, camera quality, and price versus performance. Your content may be incredibly relevant to their intent, but if you're still thinking in terms of exact keyword matches, you're missing the bigger picture of what search engines actually understand

about your content. This is where vector embeddings come in. Vectors use mathematical language to help machines understand both context and relationship between concepts and allowing them to really extract meaning. So you might be asking yourself, what are vector embeddings? Vector embeddings are numerical representations of text. Instead of seeing a word as individual isolated entities, embeddings place them in a mathematical space where relationship and meaning is preserved. Imagine a vast multidimensional space where every word, every phrase, and every concept has a specific location. Words with similar meaning are clustered together, while unrelated concepts sit further apart. For example, term like king and queen would be positioned closer together because they're related, but apple and banana would be near to each other because they live in the fruit neighborhood. But king and banana would be much further apart. So how are vector embeddings made? Well, advanced AI models like Google's Gemini, ChatGPT, and others analyze massive amounts of text to understand how each word relates to one another within context.

They do this by converting text into multidimensional vector points. Then they apply something like cosine similarity to measure the relationship between these concepts or really the distance between these concepts within that multidimensional space. They can do this with either words, sentences, paragraphs, documents, and more. Now, a key concept to understand is cosine similarity. Cosine similarity measures how far two embeddings are from each other. A score of one means that they have an identical meaning. Zero means that they have no relationship, and negative one means that they have opposite meanings or that they're opposite concepts. In SEO terms, higher the similarity score indicates a stronger topical relevance. Now don't worry, you won't have to learn any math. You just need to grasp the overall concepts. Now let's talk about some core applications within advanced SEO. One area that vector embeddings make a huge impact is in content optimization. Traditional SEOs may ask, does my content contain the right keywords, phrases, topics? But vector based SEO is much more concerned with how well the content's meaning

aligns with the user's intent. Search engines now understand that somebody searching for the term reliable family vehicle and dependable car for parents are essentially looking for the same thing even though they share no exact keywords. You can now mathematically measure how well your content matches target queries by comparing them with vector embeddings. This goes far beyond keyword density or exact match phrases. For instance, if your target query is something like sustainable web design practices, you can score well with your content if it covers concepts like green hosting, efficient code, renewable energy servers, eco friendly UX design, even if those phrases aren't actually in your original keyword list. Another area that vector embeddings are extremely powerful is in content gap analysis and topic clustering. Vector embeddings reveal content gaps that keyword analysis might miss. You can compare your content's semantic coverage against top ranking competitors to identify missing subtopics or even concepts. Topic clustering becomes incredibly powerful here. Instead of grouping content by shared keywords, you can group by shared meaning. This might reveal that your email marketing and

your customer retention content should be more closely linked or that your local SEO article could benefit from connecting to your Google My Business guide. So now we're gonna go into a demo. I'm gonna show you how we can pull keywords from SE Ranking and do a SERP analysis and take that data, vectorize that data, and reveal some deep insights. Alright. Let's say I want to rank for a term like what is SEO. Good basic search engine marketing, search engine optimization practices are still at play. I want to do keyword research. I want to understand the different similar terms, the related terms going on. And I really want to understand as well who is competing for this term. So you still need to have your traditional SEO tools because they're very valuable. So I get the score for the term here. What I really want to look at is who is ranking for this term. In order to run this analysis, not everybody's going to give me insights into their content using this tool I'm going to show you here in a second. So I need to pull a site that I know I can actually pull the content from. So what we're going to do is we're going to this term and then we're also going to take the site where we have or the page we have on our site as well as one of these competing

sites to see how well we stack up using vector embeddings. Alright. So before I get to the tool, I'm going show you a little bit of the back end of how it works. Don't worry. You don't have to know Python in order to make this work because I gave you a quick front end just to check this out with. But what we're going to be doing here is we're going to be scraping content from a site. We're going to look at the text and then we're going to split that text into chunks. This is really important because passage indexing in search engines, as well as passage indexing or chunk retrieval in LLMs is kind of the same thing. They don't look at the whole page entirely all the time. They're actually looking to retrieve specific chunks and understanding how well that chunk may answer a part of the intent of the query. So we're taking the content, we're loading the content, and then we are going to chunk the content using a recursive text splitter. So what this is going to do is going to split the content within essentially paragraphs. Then we're going to vectorize each of the chunks on both of the pages. We're also going to vectorize the keyword itself and to see how closely each of the chunks on our page relate to the main keyword as well as our page as a whole. So I hope your head doesn't hurt yet. Now let's go over to the app itself.

We're gonna take our keyword research, what is SEO, and put it right here. I'm also gonna take the page that I have on my site pulled up over here. And then I wanna take one of the competitors. I know that I can retrieve chunks from this, so I'm just gonna copy the link address and put it here and just click analyze content. Now with this one, I'm using an LLM from Hugging Face. You can do these with a number of language models. So here's the analysis. My page score is point seven three. My competitor's is point seven six. They're slightly more relevant than my page. On this first one, it's going to show me the content similarities. So these are the two sections of content. If you scroll over here, it'll show you where my pages or my sections are similar. And then I can go to this page and pull the gaps. The competitor has frequently asked questions. They talk more about linking than I do. This section here, which talking about paid versus free, auditing my website and start learning. So I can then take these different gaps, understand kind of what these different sections are and go apply that into my content if that's going to help me optimize it more. Now,

these were all found using vector similarity. What it's doing is it's taking these these chunks of content and is saying this is how relevant it is to the keyword. So like this first section has a relevance of point six out of one, and then my max coverage is slightly under that. So in all these areas, falling a little bit under except for these two sections here. But I could still cover them better if I incorporate something similar to what my competing site is doing. What's cool about the vector embedding is now I can actually go in and look at these specific sections on that page if I wanted to and analyze those and see how they structured them. It's not a copy and paste type thing. It's it's helping us better understand the relevance of our content in connection with this main query right here. So it goes a little bit deeper than just looking at h one or h two. It's looking at specific sections, how those sections are organized and structured and how the LLM will retrieve each section based on a similarity score at the vector level. Another area that vector embeddings can be extremely helpful is in advanced internal link analysis. Traditionally, internal linking focuses on anchor text and keywords. But vector based linking identifies pages that are semantically related

even when they don't share obvious keywords. For example, a page about voice search optimization might be highly relevant to link to a page about conversational content strategy or featured snippet optimization. These connections become obvious when you analyze their embeddings, but you might miss them when you do a traditional keyword based linking strategy. Tools leveraging embeddings can automatically suggest internal linking opportunities based on semantic similarity. This helps build comprehensive topic authority by connecting related concepts across your website. So instead of manually searching for pages that mention similar keywords, you can identify pages that discuss related concepts creating a more natural and comprehensive linking structure. Using vector embeddings, you can also get varied contextually appropriate anchor text that maintains semantic relevance while avoiding over optimization. This creates a more natural linking pattern that search engines actually prefer. For example, when linking to your WordPress security page, you could use an anchor text like protecting your website or security best practice or safeguarding your CMS. All of these are semantically related but naturally varied. Alright. So let's see how we

can actually apply this in the real world. We're gonna be leveraging Screaming Frog's new analysis that allows us to pull vector embeddings within our content and then run some very quick internal link optimization with those vector embeddings. Finding internal link opportunities can often be hard. Most of the systems today are really built either around keywords or titles or things of that nature. But like we said in the training, vectors get to the meaning of the content. It looks for semantic relevance. And so we can leverage vector embeddings and help us to find better connections or maybe missed connections on our site. Now to pull this off, you're going to need an API and I recommend a Gemini API and you're also going to need Screaming Inside of Screaming Frog, they've made an update that actually will do vector embeddings during a crawl. What you're going to want to do is go to your configuration and your config. And this is going to set up all the different things you need to do. The first step is to make sure that you're storing and you're rendering the HTML. This is essential in order to pull vector embeddings. We need content and that content has to be stored so that the

tool can then vectorize that content. Now if we go down here to the content, you'll see this little thing that says embeddings. Now we're gonna go ahead and turn these on. You're gonna want to just go ahead and turn everything on. I like all that stuff. And you'll notice here I've got a little a warning. It's because I need to configure the embedding prompt from the AI provider. Now to do this, I will go down to my API access and I will click Gemini. I've already got my key in here. You're gonna need to put your key in and you can select your account type paid or free. And then we go to prompt configuration. From here, they've got default prompts you can add from the library. You can write your own if you're so inclined to do so. But if you're new to this, just go ahead and enable. I like to enable all of them when I'm running a crawl because I can do a lot of cool things and hit insert. Now I've got all these set and I will go back to my embeddings and make sure that this is all set. All the green checks means everything's good to go and I hit okay. Now I place a URL that I want to improve internal links on and I hit start. As it's crawling the site, it's going to do a few extra things for us

now because of the API access. After it stores all of the information, it's going to create vector embeddings for each of the pages. Now if you're interested in where this is happening, you can go to the drop down here and just go to AI. And as you can see, as these addresses are being crawled, we're running semantic similarity embeddings, extract embeddings from the pages themselves. You can see what is this? It's just a bunch of numbers, but these numbers have a whole lot of meaning when it comes to the large language models, the generative AI applications you can use to understand semantic meaning and intent. So I chose a small site on purpose, and what I wanna do is from this AI tab, go ahead and export this and name it whatever it is and save it. The next thing you want to do is go up to the top and you have bulk reports and under links do all in links and give another descriptive name to it as well and it's saved. Now that Screaming Frog's done the heavy lifting for you, you can go back to this tool. All you need to do is upload your data. So this top one will be your in links data. So we've got all the in links here. And this

bottom one will be all of our AI data, and we'll put that here as well. Now within this tool, you have a couple options. You can choose how many links you're looking for per each source, One up to twenty. And then you can also choose the similarity threshold. Remember, we talk about embeddings, a one is exactly similar. It's the same thing. A zero is not similar at all. A point seven is kind of somewhere along the top end. Right? They are gonna have some semantic relevancy with each other. Once you're done and they've loaded, hit submit and this tool will process it. And it'll tell you the source page and it'll tell you the page you should be linking to. It'll show how similar that page is to the source page and the type of anchor text that you should be using when linking these two pages together. Now this is a smaller site, so I might wanna look and see which one of these pages, these service pages that have, okay, agriculture and commercial are very highly related. Agriculture and commercial deck railing. It's not that they're the same thing, it's that they have semantic relationships. They sit closer together within this vector space. So I might want to think about building that

internal link. Now again, with any one of these tools that we use, we have to think through the applications and does it make sense for the end goal of the business to do so. But as you can see, on a very small site, I got some ideas of how I could better internally link the the pages with one another, not based on anchor text, not based on keyword relevancy, but based on something much deeper, which is semantic relationship scores. Vector embeddings are transforming SEO beyond content and links. They're powering better site search experiences. They're enabling more sophisticated understanding of user intent for personalized results and personalized interactions across the web, and even improving image and video SEO through multimodal embeddings. Search engines are using these technologies to understand the relationships between text, images, and user behavior patterns in ways that were impossible just a few years ago. Here's the crucial point. Search engines are already heavily using embeddings. Google relies on vector representations to understand queries, documents, and the relationships between entities and their knowledge graph. It's also a key component to LLMs in AI search.

This isn't future technology. This is current reality. The question isn't whether search engines will use embeddings, but whether you'll be optimizing for how they already work today. As we close out this session, I wanna leave you with four critical takeaways. The first is understanding meaning is now essential. AI powered search requires semantic optimization, not just keyword optimization. Two, vector embeddings are the foundation. They're how machines understand and process meaning in text. Practical applications do exist today. Use embeddings for deeper content analysis, identify content gaps based on meaning, and build effective internal linking strategies. And the last one is please take time to continuously learn. This field evolves rapidly. You need to experiment. You need to stay current with new tools and techniques. This is essential for maintaining a competitive advantage. Vector embeddings aren't the the secret weapon. They're not the silver bullet that's gonna magically make you rank. Just like any tool, we have to take that information. We have to put it into practice. We have to see how well it's working and continue to shift and apply new concepts and techniques

to get the best results in the end. This is just one way for us to better understand the depths of our content, how we can better optimize it for our users to give them the information they need and build trust with them. I hope you found this helpful. If you have any questions, I'd love to stay connected and help you in this journey as well.