Reciprocal Rank Fusion & AI Visibility with Metehan Yeşilyurt

Full Transcript

0:00If you have a strong SEO fundamental, you can absolutely have a chance for the ChatGPT ranking citation and on all LM because the system is Hey everybody, welcome back to another episode of Jolly SEO's TL;DR. I am really excited this week. Metehan Yeşilyurt, the chief growth officer of AEO Vision is here to explain reciprocal rank fusion and we're going to dive in in just a moment. But first, Metehan, thank you so much for taking the time. Oh, thanks for thanks for the invite and thanks for having me today. I'm super excited to uh tell my finding my experiment.

0:54I I'm excited because you wrote this incredibly interesting paper. I think it was in July of 201. Yeah. Yeah. It it a lot of people took notice of this. Uh you don't know me, but I'm actually not very technical and I really appreciated how your paper made it easy for me to understand the math behind reciprocal rank fusion. Uh so before we dive into that though, really quick, that was the middle of 2025. Now we're getting to the end of 2025. Does this concept of RRF still apply going into 2026?

1:50Uh yes, after this finding actually it's not a new thing for the academ uh side because the rate fusion first academic paper actually published in 2009. So and we know Google is also developing its own search infrastructure and it actually about uh developing algorithm itself in real time. So we already um similar with the uh mathematical formulation uh in the search site. So um and we know OpenAI had an agreement with Microsoft Bing. So when I first discovered this actually it started with um checking console locks on my Chrome browser or Opera. Uh when I see those code line I would like to uh dive into for more information then realize that uh actually uh TPT is also using very similar handout approach in their web search product.

3:09So when I check the old sources in the back end, I mean you can just uh ask a question like what are the best trendy sneakers in 25 check more sorted in the uh request but we only see actually partial citations. I mean there were something like fourth sources but we I can only see citations in the final response. So there is a certain uh reducing mechanism uh for their web search in structure.

4:17So when I start the start to work hard on it then I realized there's a uh link fusion and as I mentioned it's not a new thing uh for the search market but then uh want to publish a very paper academic friendly uh version of tip we will tell everyone uh then we'd like to hear more insights and feedback from our lovely uh search industry and it's went politely viral. Of course, I'm I'm happy with that.

5:27Uh and now we are here and after um rank fusion of course there were many update on the Google AI mod and AI builds and of course on the ChatGPT open AI and Gemini part because they when we ask long form question to Lamb they are splitting uh question uh that long for font prompt into up question search and fetch result each of them then combine in a data pool let's say a data pool then there's a reducing mechanism and we see the final citation with the actual response so it's quite complicated for me still uh but I can see many patterns are applied by rank fusion RF uh still today. That's awesome. Thank you for that overview and kind of validating uh the importance and like you said it did go viral and uh it it was a really fascinating thing to consider. You know, a lot of us 2025 certainly, but going into 2026, more of us SEOs than ever are going to be thinking about answer engine optimization. I'm curious with AEO Vision, obviously your goal is to help users understand their ranking in these AI search engines.

6:21uh but you know maybe not all of us have uh the ability to do what you just described was your process to identify that rank fusion is at play here so so let's say we are monitoring how we are doing in these AI search engines we can see where we are doing well where competitors are destroying us once we have this information where do we even start as we begin to try to optimize Yeah, first I'm not a typical GEO guys. Um I mean I'm not criticizing anyone in our search industry because I respect everybody every opinion because the landscape is uh changing very rapidly. Uh so we need to see what happening in the uh industry.

7:17Of course on the also the customer size because uh I started to work on ChatGPT uh for an optimization part. I get end of 24 uh heavily wanted to invest because also trying to improve my skill. uh and I realized that people or companies just uh started come up with uh question of like how can I show up my brand in chat and right now we also have perplexity copilot cloudy and ok gemini but the we also know the biggest market share belongs to tetto Open AI right now and X AI is coming etc.

8:23uh but when people companies brand came up question like this so there's a new uh pattern over there and I still believe in SEO because I have a very technical SEO background uh over a decade and it like yes there there is an opportunity and I love working with algorithms So I believe in the United States uh part of VC it's some company just wanted to jump in this hype and they want to call something new but we uh also knew that there very shifting model in early 2000 man care and renamed it mobile SEO but right now we are not calling it mobile SEO right now or we had some predictions about uh voice search late 2010s and we were predicting uh year like 20125 uh almost 50% of the search queries will come from uh voice search.

9:41So uh the landscape is changing very rapidly and of course AI it is really different because you can uh if you can use in an effective way you can increase your um outputs outcome. So you can actually create your better stack and I believe you need to start with a site audit.

10:49I I wanted to explain all of these because I believe if you have a strong SEO fundamental, you can absolutely have a chance for the uh ChatGPT ranking citation and on all LMS because the system is very similar at the moment and of course the retrial part is different than uh the traditional Google search and we know Google also is releasing their own uh um AI search product and everyone is basically working on how can we monetize this with ads. Uh so what I'm trying to tell these are very similish uh with early 2010. So the thing is I always start with a site order because LMPS are also using user agent and those both user agent are your website they need to pass your layout they need to read your content.

11:50uh right now they are not boot on the image I mean multimedia part like Google like uh you need to provide them a very clear semant structure because they are uh heavily texted uh retrival tools at the moment especially on the search metahan we didn't plan to cover this but I have a shotgun question here for you do you consider making content just for AI crawlers. I know that some SEOs are considering this and some SEOs are saying no, we stick human first and we make sure it's structured so that AI crawlers can do their job. Where do you sit in this discussion? Um, first respect every idea because I'm still trying to learn more uh and hear more from our industry to develop my own ideas. But this is very clear when you only create content based on the AI system. Uh probably you will need many pattern like you'll do engagement etc because are using a temperature settings.

12:47It allows to create more creative content on their uh back end and set GPT for all API model uh have a 10 per setting but GP5 uh and 5.1 just doesn't have this system. So uh when you only create with API in a let's say bulk mode uh you can create thousand of content but I believe uh if you only use basic prompts uh probably you won't drive uh traffic thousands of millions of traffic from or uh Google and the second ping is eleant are both improving improving the state contact moderation.

13:59So you need to be careful and if you only uh um follow him pattern uh just like early version of Google search algorithm probably you will get panel in the future I don't know maybe 2 years later uh you can still generate uh revenue maybe 6 months later and we also know that I guess it was from prompt watch uh finding ChatGPT I mean open AI right now uh using an entity system and uh it also I get in the profound uh finding those are just uh great study uh so they are also improving uh their system but the may the third thing is Land are not targeting the search vector. They're trying to create something new. But Google expert type is on the search part. The search pipe is also generating the most revenue for uh in Google.

15:09So you talked about LLM are creating something new. uh the let me just briefly try to say my understanding of reciprocal rank fusion and then please correct me so I can get this right for the audience's sake and for my own uh but but what I saw as a takeaway from this is as SEOs we are no longer aiming just for spot number one we need to consider that we want to be in the top 10 maybe even a lower down but in the top as as many as we can and there's waiting to this. Uh yeah. So can you help me understand how this applies to AEO?

16:00Uh ho how much of reciprocal rank fusion we we want to adopt as you said it's been around for over 15 years this idea of rank fusion but how much we need to adapt or adopt our SEO best practices with this new paradigm that it's not just number one. We all know what's been happening even in Google search. But the the AI stuff it it was new. Uh we also saw it in the US 2023 search generation. Um third gen experience uh SGE. uh what I saw an experience uh is the traffic uh were already uh complete live for almost everybody. Of course, if you are in ecom business or um let's say healthcare business, h click need to be made by the users like if you are going to purchase something.

17:09Uh but for the all query uh we will we are already in a periodic traffic and we already got month. So this is is it new uh for this period I believe visibility black visibility is the most important metrics. I mean if you are visible on AI overviews, AI modes and also in Land it will count a positive point for your brand.

18:14In the past you need to trade the traffic but right now we are chaping for is I need that that's the actual diction for this period and the final thing is earned media and old media are very important because only owning your brand domain isn't enough uh for the rank fusion for the visibility you need to be more visible on and other platforms. I mean let your audience please. Yeah. You know there's the classic marketing saying which is a customer needs to see your brand seven times before they make a purchase. Yes. I I found it very fascinating that part of 2025 has been SEOs realizing we are still marketers even though there was some resistance to this idea in SEO prior to this. Uh I I really happen to agree with you. Of course you are speaking our language because we do earned uh linkbuilding only. We don't do paid links.

19:03So we focus a lot on brand as a result getting getting clients brands in the media which to be honest until this year we were nervous. Are we still going to be useful in the future? There was some talk in SEO about links are they still valuable or not. It turns out getting your brand out there in a lot of different places is very useful in this next generation. Uh, but I want to put you in a little bit of a hot seat here, Metehan. I'm going to ask you a little dangerous question. The algorithms are changing all the time, but can you make one prediction for us? As I just admitted, I couldn't have predicted that brand citations would be so useful right now.

19:51But going into 2026, can you make a prediction on where the Open AI ranking algorithm is heading? I believe they're uh really connected with being I believe they will start in their own search infrastructure and when you look for or uh search for some mathematical formulations on the relevant document comparison or document finding site I for the semantics search you can only find some uh actually you can find 10 to 100 patents uh at at the time. So imagining yourself.

21:04Uh I mean this is for everyone who watching our uh podcast right now listening our podcast and my prediction is first blackheads can be the one of the hottest topic in the neck here because LM have many bi at the moment like research Okay. Uh and the second is I believe of course with the atlot browsers the agent browsers um the agent pitch hype can be bigger than uh right now.

22:19So uh just let's see how it go because when the agent prints optimization arrive maybe people want to name it something very different acron uh so I I don't know but we'll see when the agentic optimization uh arrived we see uh and we will try to optimize any web page or any product page on the core ele that will be also a huge difference but let's see yeah I think it's a great prediction you know going back to my question I asked you uh right now maybe no one is really building just for bots right but the agents they are coming uh they're slowly coming. I was just at an event in Chiang Mai uh last week and only two of us out of about 100 people said we had agents working inside of our business. Uh certainly there are people more on the cutting edge and and doing things, but as you said, we we haven't even scratched the surface yet in terms of optimizing for agents because they're not yet really out there operating. I I I like that prediction a lot. Uh Metehan, I loved reading your paper.

23:07I I saw that it went viral. Uh I tried to ask my best questions today, but why I really wanted you here is actually to see what you can teach us. Before we wrap up, is there something I didn't ask you about that you'd like to share with the audience that you're wondering like why why are we closing out? I didn't yet get to share this. Um, yes, I um I'm currently working on reinker uh reinking systems because I can um I can tell you I mentioned these topics in some of my posts but not really uh clearly.

24:04Uh please first please please use academic research paper websites uh often uh just type LLM ChatGPT use your own uh keywords. Uh it has a lot that those platforms are my first and favorite platforms for new findings. uh and I believe we need to lead more especially on the fun part because Lance has many unsold problems on the retrial parts. They're using retrial augmented generation which is uh actually different from Google the traditional search I and I believe Google has the back system because it their job and they echard deep mind over a decade ago. So they are good at it. And uh second is re ranking layers. Uh if you use 10 stat with a visible text on your page, it helps a lot.

25:14Please just update your uh more than six months old content. uh it helps a lot for the uh driving more traffic and citation from LLMs. I'm telling this because re uh in techt reanking system favors wom contents that obies I also publish uh post about this uh reanking system of chips is very ding and has many parameters he site so I'm trying to uh work on it But what I realized and didn't share uh before is the information already validated uh in our search industry but it also uh ties with Chip's own ranking system.

26:32When you use um KDR section in your uh beginning of content, it helps a lot uh for the ele citations and traffic because uh right now Lance have uh lost in the middle problem. their attention window uh for crawling any website is the beginning part and the end part. They are just getting lost in the middle section and they're trying to uh use an houseate sometime on the content. So using bulleted lid with basic HTML table in the middle part helps a lot. Just apply this and see the difference. Metehan, that was amazing. You packed in as many tips as you could in 25 minutes. So I really appreciate you coming on. I I know people can look at aeovision.ai.

27:26Is there anywhere else you would love people to come and follow you and ask some questions? Yeah, at a time on LinkedIn or at TB. Awesome. Awesome. Well, thank you so much, Tashik. Really appreciate it. Metehan and creative. Thank you.