Koray Tuğberk: Rank #1 with Semantic SEO & Topical Authority!

Corey Tugberg, how are you? I'm good, and you? Yeah, good, thanks. You were just showing me this case study of you taking a single page from zero to 8,000 clicks a day, right?

That is true. A single page, exact mesh domain. This is amazing. And you're focusing on visual semantics.

Yes. How does that work? Yeah. So technically, I believe it was three years ago, Martin Shiplett from Google, he accidentally confessed something.

He said that actually Google is using what they call centerpiece annotation. And he told that centerpiece annotation wasn't a text. It was a visual element that they are trying to see on the page to decide whether we should rank it or not. Then Google's DOJ leak files came out.

And in the DOJ leak files, we also have seen examples of centerpiece annotations. It was also explaining how Google internally was evaluating the above the fold area of a page. Then also in the API leak, the Content Warehouse API leak of Google, there was, and there were, and still there is, the annotations of the semantic label annotations on the page. There was something like 22 different annotations there.

And that domain, it's audio2text.com. It's an exact mesh domain. And people can find it from my YouTube channel. Or if you allow me later, I can also share my screen and show it live from the Search Console.

But in any other case, we've actually used a giant upload button by adding the no-signup annotations. Because when I checked the AI answers or agentic answers of Google or the large language models, being free, being no-signup, there are two of the really important attributes that they are looking for to classify the size. And instead of putting them to a button, like upload now or something like that, I have put a giant element there that shows that actually you can direct the drag and drop your audio or your video. Secondly, I also have put the notification of no-signup.

And then I have written very little amount of text. I have written very little amount of text because if you write too much text, it's going to be diluting the importance of every token on the page. In the first 2% of the tokens on the page, I have put all the important n-grams and entities, like file extensions or like audio2text or video2text or all these from X to Z combinations. And then we started to get the first clicks.

The rest was actually just click test of Google. Technically, the moment that I was able to get, let's say, 50 clicks a day and 8,000 clicks a day, there is zero content change. There is zero visual change on the page, too. I just waited until Google understands it.

And every core update and major update, it keeps increasing the site. I can also in the graph show that sometimes we even lose 50% of the clicks, but we don't change anything there because we know that Google needs time to be confident to rank the site permanently. Now it's ranking with 11 laps and with ION, many global companies, basically. We can tell.

It's amazing. So literally, you created this page. You didn't touch it. Every update, it goes up.

And it's not even about, for example, how much content you put on the page. It's more about how you organize it. That's true. How visually you organize it and how visually you annotate some certain attributes.

Because agentic retrieval is not only about the text on the page. It's also about what they see on the page. If the page needs to be an e-commerce design, you should be putting it like that, even if you are just an affiliate site. Still, you should be doing it in that way because you can imitate e-com design for product pages or category pages.

And you can rank for it. Actually, Indonesia, for iGaming industry, is frequently performed. You can imitate a product page. But when people click to the buy, they just go to the money site to log into the casino.

So Google is actually buying these visual representations by trusting that nobody will be imitating an e-com site or something like that. Technically, it says free audio-to-text tool there. It's mostly free. That is true, but not entirely.

So but Google can't see behind what happens the moment they actually upload the video because they don't upload anything to test the tool. They just believe they click engagement. And they see engagement. And they get confident to rank for it.

The way that you approach SEO is so different to anyone else I know, man. Thank you. I hope it is a good thing. Yeah, of course.

Of course. Where do you get these ideas from? Because you must have that idea months ago. Or you've known about that for years.

And then you've tested it and run that experiment for months. Where do you get these ideas from? The ideas, to be honest, is only coming from actually Google engineers. Because this idea first actually came out from Michael Bandersky and Martin Nyorg.

Both of them are great Google engineers. And they patented something that they call document-aware layout ranking. Technically, it means that actually, or I can also call it like this, layout-aware document ranking. Meaning that actually they were trying to create a design to see your layout and design.

And then they were ranking you according to that. And Nyorg and Bandersky, they are really important engineers in Google. And if they patent something, it's going to be implemented. It was the first moment for me to start to care about the design.

And then I followed some other researchers on Google, did some tests. It was working. And so it has to work. Because right now, anybody can create expert-level text.

Text doesn't directly show originality anymore. It's still a risky asset to trust with. But layout and the design is still requiring human effort. And it can be used as an originality sign.

That's why I believe even if you use AI to design something, adding basically some human signals or human effort signals to the page will be important. And that audio text that comes, the design actually is, again, coming from AI. But I change all the HTML hierarchy or the class names or the CSS and everything there. So I cleaned all the AI signatures there.

Because many lovable sites, for instance, they just use the same class name, same div hierarchies. These signatures that you leave there will be important as in the text, as in the source code, or as in the visual elements. That's it. So you recommend if you're using, for example, Cloud or any sort of AI tool to create your website, clean out the CSS, clean out the HTML?

I would, yes. Because they are definitely leaving lots of signatures in the content. And if I look at Google, and I know that they did in the past, many WordPress and GeneratePress sites, especially, they got penalties. Because they were classified as AI generated so easily.

Because all the content scalers in the early times, everybody was using GeneratePress. And it was an easy signal for Google to consider you as a possible content scaler. And this scale content abuse, they get more sensitive on it. I did also an audit for a company, BrowserStack.

You can also check it. BrowserStack has 23,000 referring domain, 300,000 monthly brand search demands. And they are the biggest company on their field. Still, they lost 70% of their traffic in eight months.

Because they publish 2,000 documents with AI in a single day. So I believe the scaled content decision trees of Google is crushing every other decision. If they see that you are scaling too fast, they just classify you as, let's say, full automated. And it doesn't matter whether you are a really strong brand or whether you have a really strong referring domain network behind you.

You can check that domain. It was also a kind of wow moment for me, too. Because I always considered for the last five and a half years, if you are a brand, if you have brand search demand, if you have a really strong link graph, I mean, Google will tolerate you normally. Today, it's becoming more different, apparently.

How fast is too fast when you're creating content? I mean, I have created hundreds of articles. I've never done 2,000. I think that's pretty wild.

On that area, let me explain like this. For instance, right now, I'm creating a subfolder for a company for just trains in the CBD industry. I believe rather than directly understanding the frequency, we should first understand age of the subfolder or age of the site. If you created a new subfolder, and it suddenly becomes 1,000 URL in seven days, it's not that good.

But if you age it for three or four months, and then you start publishing, let's say, 50, 60, 70 type of an increasing frequency, they will tolerate it better. Because there is click data that is approving you as a satisfying source in that part. Usually, I try to complete my entire network of content around seven to eight months. If I have to publish 5,000 pages at the end of the day, I need to calculate, basically, if I go slow in three months, in the rest of the four months, how many a day I should be publishing.

And I just actually divide it to the number of the days. And then I randomize the numbers. Sometimes, I publish two, but next day, it becomes 27. And then, let's say, seven and eight.

But eventually, it is ending in seven to eight months to complete the entire network. Many companies, they suddenly publish too many high numbers. And since it's a very fresh subfolder without any URL click signals, it is actually risking the health of the entire site, I can tell. It still depends on, basically, your entire topic.

In Citrains, it's normal to publish too many because there are, overall, 6,400 Citrain types. If you are creating a car directory, car model directory, there are thousands of models for every year. And if you publish, like, three a day, it will take years to finish it. And you don't have that time.

So you need to find optimum balance there between risking it and between also going health away for fast. So I believe maturing the subfolder for three months is the most important thing. Once you get the maturity, then you are safer to scale it. The same also goes for Reddit, too.

If you make a subreddit live for three months, it is tolerated better to scale the content under that subreddit, for instance. That's why we open lots of exact match subreddits with the names, like Cruise or, let's say, the eSIM, et cetera. And we make it live for three months. Then we start to post.

It is actually becoming less vulnerable to the Reddit's internal spam detections, basically. Yeah, that makes total sense. Same probably with the main account as well, right? user profile.

That's true. And if you use adspower or any type of tool to create lots of proxies and accounts on Reddit and other places, maturing these accounts in a semi-active way without publishing too many URLs or comments and etc, that will be better. Sometimes just log in and scroll and there are in adspower, there are automation to do these things. We also use them.

And after three months, we try to activate it for our use, basically. It's a kind of, I believe, warm up threshold to trust an account or subreddit or sometimes a subfolder. When it comes to the actual, so, I mean, you've covered a lot there. So for example, like figuring out, okay, here's how many pages we need to create to achieve top quality.

Okay. Dividing that by the number of days in eight months, seven to eight months, and then randomizing it. So some days you publish more, some days you don't. When it comes to the actual topical map itself, how do you come up with that?

Do you automate that now? I semi-automated, that is true. Still semi-automating it because every year the methodology gets new things, especially as visual sections, because now in the content briefs, we have a column that we call component, which explains how these headings will be look or visible on the page and how they are going to be engaging. But when it comes to the, beyond the briefs, when it comes to the topical maps itself, it's like 65, 70% automated, but we created another SOIP that we call query deserves a page, query deserves a heading, or query deserves a section.

So it tries to understand whether we should make a page, heading, or just a sentence. According to the query understanding or the mapping, we basically are able to, let's say semi-automated, but still I need a human to go over these topics while creating content briefs, give the final decision to make it a page or to make it a heading or just the internal link connection to another segment as well. It's possible to semi-automate it definitely, mostly automated actually. And when it comes to completing entire topical map, based on the principles, to be honest, it can go even forever because all the queries are connected to each other.

You can always go deeper. You can always go wider, but usually you don't need to do that. According to your competitors, if you're able to cover overall 25 or 30% more topic and also in a deeper way, you are safe actually, you don't need to like cover three times higher topic in a shorter amount of time from them. If you put like 20% coverage difference, it's usually good to be considered as a replacement candidate during a core update to replace them directly.

So just take all of your, let's say eight main competitors to create their sitemaps, then see what they are covering first, and then create something unified in a better organization to cover 20% higher than all of them and you'll be safe actually permanently. That's interesting. So you combine all your competitors and then you're like, okay, let's cover 20% more as well. Yes.

And that's usually slightly making things harder because business models are not the same. Let's say you're a logistics company for cargo. One of your competitors, let's say they are actually just working for motorcycle cargo, but you might be also doing island to island ship flights as well, which actually makes your topics and business models also different too. But you need to also in a way cover that area as well.

So sometimes if you can't cover something commercially, you need to find a way of covering it as informationally. That type of organization is actually where we do manually. Other sections with AI mostly can be automated, especially if you scrape and crawl their sitemap content, that will be accelerating process in a very good way for you. Actually, just talking about automation as well, I'm interested to know, I mean, for us personally, our team has kind of shrunk at the agency quite a lot with AI automation.

Has your team grown with AI or is it a lot smaller? I actually it's grown. I always try to keep people instead of replacing them with the machines. But besides that, right now, every employee is capable of doing more work.

And since everybody is capable of doing it, I believe AI, yeah, it made things easier, but also it increased expectations from an employee as well. So it doesn't make that much of a human workforce need difference on our end. Because in the old times, I was telling a graphic designer to give me just eight images a day. Now I'm asking him to use this cloud skill and complete the entire site.

And once he completes it, it's a normal thing for me, then I give him another site, which actually didn't decrease the number of hours that they need to work. It just made output numbers very much astronomically higher in that case. So since the demand is also growing too, we still grow the team, but every teammate became like a superhuman beyond human capability, we can tell. Yeah, so you can basically 10x your output.

Yeah, of course. And our expectations from everybody increased too. Because for example, right now, a content writer can be a developer too. If they like designing, we just tell them, okay, take this mail, log into cloud or codecs of OpenAI and start doing what you want to do.

And they are capable of doing these things also by following just the SOPs. So number of designers and developers, that's why it is increasing. Number of writers are slightly decreasing. So slightly switching to the other angles and areas, we can tell.

Yeah, it makes sense. I know you mentioned before as well, like it's so easy to create expert level text now. How do you see that changing things in the future? There are two things there.

I'm a little concerned about the legal side of these things, and probably still we will be safe. But right now, for instance, when you write a book with the AI, there is a chance that you might be losing copyright claims or copyright ownership. If you use AI for inside university, they basically don't accept your PhD anymore. And in the search demand too, it can be seen, AI writer or AI content writer, search demand is decreasing in the trends.

But AI humanize and humanizer AI is increasing instead of that. So it looks like the demand changed towards the humanization and originality signs of the content. But mathematically, it looks like humanizing a content or recognizing the AI content will be becoming easier and easier every year. So I'm expecting a level of basically legal regulations on this field, because the scientists are complaining, novelists are complaining.

And I believe, for example, if you made a movie or television series like The Friends, still do you get money, right? When it is aired on the TV, you get the money. But when you write a book, and if an AI company is using it for training on that, you get nothing. And legally, this is a big problem.

I believe they will regulate this globally on the planet. I feel like in that way, at least, it's like probably there will be an analytics tool to track when your book or your data has been used by the AI to answer something. And there might be a kind of monetization model based on it. And I believe there is a demand for this already.

But from a SEO point of view, I believe whoever is using AI in the most possible creative way, still, you'll be safe in that area. If you created blacklisted words, blacklisted sentence structures, blacklisted word combinations, even if you use the AI, mathematically, they won't be able to recognize that it's actually coming from AI. And usually, AI detection, in my opinion, it works like Turkish police. Because Turkish police people or men, they don't have to prove that you are the guilty.

They just look at you. And if they think that you are guilty, they just handcuff and take you in. And you have to prove that you are innocent. And I believe AI detection systems, they work in that way too.

If it is over a certain threshold, they assume that you are possibly AI, but they can't be certain mathematically. There is no way of proving it. And in the early days of the AI detection, I was putting Bible to an AI detection system. It was saying that it's coming from AI, but it's not.

It's Bible. I mean, God, I hope possibly has written it. And when it comes to this process too, I believe using AI for content creation will require a level of mastery by sensing also AI detection systems at the same time. And that's why recently, I'm also working on the visualization techniques.

Because if I use fewer number of the words, every word will be contributing for relevance better. And if you have fewer number of tokens on the page, it's harder to tell it is AI or not. AI detections works better precision if the content is long. But if your content is short, there is not that much field to make mathematical calculations to see it's AI or not.

But if I design the page in a proper way with visual annotations and place my little amount of content there with small amount of verbalization, I will be safer from both ends. That's why audio text that come too, it doesn't use that much text. And if you run any AI detection tool there, it's invalid detection actually, because it's just three or four sentences. It's not going to be proving any point there.

So that's why I'm trying to also decrease number of tokens on the pages to stay under the radar as well. Yeah, that's a smart strategy for sure. The other thing I was thinking was like, if you're looking at the way SERPs are going and the way, for example, Google AI overviews gives you an answer, something like that website before the audio to text one, it's like that's not going to be replaced by AI anytime soon, maybe in the future. But I think it's probably one of the last formats.

Sure. That's why every click that we lose from informational side, we try to add it back to the free tools. Even if you're a lawyer, you can still have lots of free tools there, like free AI evidence evaluation, or free, let's say, case compensation calculator, or free, let's say, the accident Photoshop evaluation. You can upload your accident photos and for free, that tool can actually tell you who is in the fault possibly, and also put a CTA call and get a free human case evaluation as well.

In every industry, you can create free tools today, and you can actually transform your click engagements to the tools instead of informational section. I didn't run a system yet, but I believe people should do this too. Search Console started to give you AI overview impression data. I believe people should start also compare their regular web impressions to the AI impressions, along with also their crawl data too.

And from there, they can also understand what type of content gets high. percentage of impressions from AI and how frequently they are being crawled by what type of Google user agents. Answering this also might show them some strategical insights to shape their topical maps or to change their content style to get better approvals for informational content as well. It can be also tested because in some projects we also see that they choose certain subfolders or certain sentence beginnings too.

I believe that data can provide good insights as well. Yeah, that makes sense. I think if you combine that, it'd be quite easy to collate that data and then plug it through something like Cloud, get it to analyze it for you. 100%.

If you get AI impression data per page and regular impression data per page, and if you just get your crawl data per page with user agent data, throw it to the Cloud and just ask some questions. I'm sure it will show you lots of graphics, designs, and correlations as well. Yeah. How are you seeing link building play into all of this as well?

A lot. Way much more than before. That's why we are very deep in the link building as well. And we started to call it, actually, James Dooley called it external topical map because as we create a topical map for internal content publication, we do a second one for external publications.

And there too, for us, important thing is actually getting some attribute assignments. For instance, for audiotext.com, I mentioned that no signup tool and free tool. These are two important attributes. If I'm a lawyer, and if I want to get attributes like the highest compensation or highest success rate, biggest compensations, or biggest law firm in Texas, if I want these type of attribute assignments to my entity, I need mentions and declarations.

So from a link publication, I can get actually three things. I can get page rank, I can get a mention, and I can get a claim. The claim meaning that you can mention your brand with a claim, like biggest law firm in Texas, the founder of the brand says so. I keep mentioning myself with certain co-occurrences and certain assignments to myself, which helps us to get also rankings in AI.

If somebody searches for, let's say, who is the most successful lawyer in Texas for brain traumatic injuries, for instance, since I am having lots of mentions of myself with these ways, the link publications also help me to get rankings and citations with my brand name as well. So when we do the link graph building with these external publications, it's mainly for manipulating AI answers by creating a consensus. It's also getting for page rank, and we try to choose the topics according to the domain that we are going to use. Let's say we use a domain like skillsunit.com, then I can link a low side, something like skills to be a lawyer.

And I can mention myself in that article by telling that, let's say, the founder of X company in Atlanta, Georgia, says that to be a lawyer or to be a successful accident lawyer. I can create this type of annotational text to mention myself on that topic. If I'm talking about, let's say, truckcalendar.com, I can actually go for truck accident subject to link myself and my service. So we always use the domain name and domain ranking power to get a link from that subject with certain annotations as well.

And I believe link building is at the center of everything here, because especially for consensus, you need external sources to mention you rather than your internal sources, we can say. Yeah, I've seen a lot of people using guest blogs on other websites with listicles, and that kind of builds, number one, recognition that you're the top expert, and number two, those pages actually rank inside Google AI reviews as well. True, it definitely works. We sometimes create also secondary domains too with exact match domain terms, let's say, Best Lawyers for Filipinos, or let's say, eSIM for Africa, etc.

And when we create this type of domains, we put content there, it also affects AI again, thanks to its exact match domain terms. Sometimes we also create subdomains under Mintlify or under DocuService, Notion, and other platforms, and we use these subdomains to feed the AI with our content directly as well. And sometimes we also turn our blog posts into the Reddit posts, into the video content, and we distribute these videos into the many places on the internet for socials. So it's a kind of content pumping or content seeding strategy.

You always reformat the content as audio, text, and video, and you syndicate it to everywhere else on the web, so that you increase the surface of your content. So actually, all these LLMs and crawlers, they can see it with higher frequency, and they can listen to what you say instead of the competitor. It's a kind of echo. That's certainly something that's worked well for us.

We get the video, and then we'll just blast it across like 30 different platforms in different formats. Yes, especially IMDb, for instance. Every video, long video that we have, we also turn it into a podcast. And as a show, we add it to the IMDb, and IMDb, we are getting a professional profile for the person, which we can write their biography, and it affects the LLM in a great way.

Then we also, in thanks to that, we also turn our podcast into an entity. And that podcast can have multiple episodes that you can distribute to the Spotify, Apple Podcasts, and many other platforms. Then you can turn podcasts into an ebook, and you can submit it to the Goodreads, and you can submit it to the Google Scholar, and Amazon has a free book, which also, again, makes you an author, like a showrunner and author, which actually impacts AI one more time. Because you always use these creative and also alternative platforms to tell what you want to tell.

And it's a kind of good marketing strategy too, actually, because we do this for AI, but also sometimes people, when they search your name, if they see the books, podcasts, and all these knowledge panel things too, it's a good representation as well. So it helps for every angle. Yeah, that makes total sense. And also, how many of your competitors are going to be bothered to do all of that?

Not that many. Many people need a professional support to do these things properly, definitely, because we also open sometimes separate exact match domains, and book names also become exact match too. Let's say, Truck Accident, Love Your Texas. It's the name of the book at the same time.

And it ranks actually on Google as a book site as well, because when you have a different type of site, Google also gives it a chance. And you can turn every chapter of the book as a sub page as well, which again fits LLM one more time as well. So it's a kind of work, but it definitely works in the mid and the long term in a good way as well. Makes sense.

And then what sort of techniques you use to make sure when you're creating content, it follows that semantic structure that you prefer? There are four criteria there. One of them is factual content and opinionated content. According to the query, sometimes I need to use opinionated sentences, like very casual language, I think, or you think, or this is the best, like more sentimental review language.

If the query requires me to be very expert level and factual, then I never include any opinionated sentence there. The second group is about structured content, or structured or unstructured content. Some queries require me to have listicles and tables, and some of them just requires me to have a simple paragraph, basically. And for every one of these combinations, we have different set of sentence structures and different set of sentence orders too.

If it's factual, I always start with an is a sentence. The second sentence becomes with a predicate. The third sentence goes with an attribute. Let's give an example.

We can tell that, let's say, the negligence is XXX. We can give definition. Then negligence means. So second sentence goes with a predicate.

First sentence goes with an is a definition. The third one goes with an attribute. It can be something like, let's say, the importance of negligence or the compensation of negligence. We always use X of Y group in that area.

These are some rules for factual content. When it comes to opinionated one, the rule is actually becoming chaotic, because in opinionated sentences, you don't have to follow an order at all. It has to be very natural. That's the rule actually there.

Writing some opinionated sections. For instance, if you go to the eHarmony Senior Dating page, and if you go to the most bottom part, you are going to see that we repeat some of the questions in the factual section at the most bottom part, like a forum. Like, what's the best place to meet with, let's say, seniors over 60 years old. We ask it.

And then there is a relationship expert there, answering the question in a very casual way, based on her expertise. The purpose there is actually combining factual and user-generated content together on the page. Because sometimes they wait UGC, sometimes they wait factual content. And we try to have both of them on the same page with proper structured data and visual segmenting, so actually we can get our chance from both angles.

This is also existing in their AI overview patent as well. They score both of the language modalities and they give a combined aggregated score at the end. That's why we are trying to have all of them. And there are many other rules to structure them as well.

I believe we explained some of them in our course, but the others also are scattered to our case studies or example sites as well. Awesome. Yeah, that's some good, some great tips today, man. Thank you so much.

I hope it is helpful. Definitely, definitely. And lots of interesting ideas that I've never thought about before. Happy to hear that.

Thank you. It's always a pleasure to see you as well, by the way, and you are doing also a great work in many places. I always follow you, but I guess you closed your Instagram. I was also following you there.

Did you close it or is it? Yeah, I mean, Instagrams, we've now got like four different Instagram profiles, because it just, for some reason, they just take an account down for no reason. And then it's like, okay, well, we'll just have multiple accounts. I guess there's filtering algorithms might not be that smart.

It looks like. Yeah, 100%. Yeah, it's very temperamental. But it's actually one thing that's been surprising this year is we grew an Instagram account from like zero to over 30,000 followers in Wow, that's great actually.

Do you have a video or tutorial for that? I should create one, but it was literally like create three reels a day, three second hook, that's like, you might be, you might, you just got to grab attention. So it'd be like, okay, orange hat, shout China or something like that in the first sentence. Like you got to get people's attention quick and dumb it down as much as you can.

It sounds interesting. I will check that account definitely because we need something like that. It will be a very good practical example for us. Yeah, yeah, for sure.

I know Vaibhav as well. He's doing pretty well with Instagram as well. But yeah, that's it. And then so it's like hook first three seconds.

And then actually it's quite easy to monetize Instagram because you can grab people's attention and then you can be like, okay guys, and if you want this full system for topical authority or visual semantics, comment semantics, and I'll send you the guide. And then you just send them to your funnel. And it is very easy to get traffic. If you look at school.com, the number one source of traffic to pretty much everyone's about pages is Instagram.

And then YouTube and Instagram are right up there. I will definitely check that because I'm not that active on socials, not in this way, but I need to learn better for that. I'll be checking it definitely. Definitely.

I'll send you a link. I will be. Just don't laugh at me. It's good catching up, mate.

Is there anything we can link to inside the description? I believe our course will be good and also our LinkedIn account or Gmail and they can reach out. And I believe if they search my name, they can find us in everywhere. Slack, Discord, Telegram, Facebook.

We are mostly everywhere, but I am not able to keep it active. So I will learn that part from you. Awesome. Well, I know you've just released that video as well with the case study on YouTube about going from zero to 8,000.

Yeah, we'll link to that. It's the last video currently, but after Saturday and Monday, it will be behind that. But they can search like one page exact match domain case study and they can find it. Perfect.

All right. Thank you. Thank you so much. And see you very soon.

See you.