How to Influence ChatGPT & LLMs in 2026 - The SEO Playbook Nobody Is Talking About

sharing 11 ways to influence Chachi PT

and other LLMs things that you have been

reverse engineering obsessively for the

past year and I'm dying to know I'm

dying to know Metahan what are some of

the ways to influence LLMs?

Okay, this is the um hot topic at the

moment. You know, uh everyone is chasing

uh more traffic and conversion from LLMs

at the moment and of course uh CH GPT is

the hottest one. Uh so and I believe

Gemini is following it. But as we know

uh on the traffic site every time when I

uh open Google Analytics I always see

Jet GPT on the top on the uh LLM site

and I created some uh custom reporting

for it. So um how to influence LLMs at

the moment? What can I say at the first?

Um

it's uh it's like

I guess I will start unfortunately but

it's also nature uh

in the llams at the moment building a

consensus

it's it's really works and um everyone

is also uh sharing many uh case studies

around it. uh some some of them maybe

gray hat I don't know uh but

especially if you create a listical or

if you include your brand name or your

uh personal brand name in the uh best or

top listicles at the moment it works

really great and you can also see the

effects almost in 24 hours in the first

24 hours after indexing. So yes,

building a consensus

works at the moment. Second, content

freshness.

Content fence and llams uh have a

recensive bias at the moment. And if you

create the uh most upto-date and

comprehensive uh document uh in a very

short time period, it also works and

it's actually I believe it boosts your

AI visibility at the moment. And

the third one I can say uh LMS also have

middle in the lost problem. So when you

share a uh when you share a longer

context they lost the uh actual context

middle of the post middle of the

document. So when you create HTML tables

uh listings with semantic tags, it also

works

uh and it increase

your um citation rates at the moment.

Uh for the rest of it, I will talk

some other details at Dubai. uh but

today I will share my uh knowledge as

well.

>> This method of marketing is so effective

I had to make sure it wasn't against

Google's rules before I kept doing it.

It's a form of SEO I call compact

keywords. Whereas most SEO focuses on

putting up articles to answer questions

how, what, when. Compact keywords

focuses on putting up dozens of pages

that sell to searchers who are actually

looking to buy. These pages rank on

Google and convert so much better than

normal that when I discovered this years

ago, I couldn't believe this was

allowed. It's less work, too. The

average compact keywords page is only

415 words. Compact Keywords is a 13-hour

deep course on getting sales with SEO. A

customer recently said, "Each lesson is

dense with information. You're giving

years worth of experience boiled down

into 15 to 30 minute lessons with no

filler or fluff. I feel like I'm gaining

a new superpower. Compact Keywords is

about setting up an SEO funnel that

brings you sales for years and years and

years. It works with AI. It's less work

than traditional SEO and it makes way

more money. You can get it now at

compactkeywords.com.

Back to the podcast. Number three, um

can you actually can you talk more about

that? So, you said uh there's like a

context issue.

>> Yes. Yes. Yes. Of course, um, as we

know, LLMs have some context limits and

they can't process uh thousand of pages

of PDF files, HTML files. So if you uh

at the beginning if you provide a clear

uh HTML text or let's say server side

rendering uh and we know JavaScript is

uh affecting in a negative way for the

uh AI uh crawlers they can't process uh

heavily JavaScript websites. So let's

say if you provide a clear semantic

structured um pages to AI crawlers LM

crawlers and uh if you

create uh not too much longer I mean

it's it's not optimal to share a word

count average word count uh because it

depends uh for um every case almost

every case But uh think from a user's

perspective. If you are writing a uh

long prompt let's say it can be

structure you can uh put some steps uh

in it uh even you as a person uh you can

feel you sometimes you lost in the

middle of the context. So at the

beginning and at the end uh you can also

uh check it in your chat GPT or other

LMS uh prompt history on your browser.

Uh you can realize you give the best

context at the beginning and at the end.

So LLMs also have this uh bias at the

moment. So if you uh provide them um

tables and listings at the uh middle of

the content, they can uh process it in a

better way and it also increase your uh

citation

uh

share of voice uh percentage in the AI

uh visibility tools at the moment. you

can you can see.

>> So, are you saying basically like um an

LLM isn't looking at everything on a

page? It's looking at the beginning, the

middle, and the end essentially

>> at the beginning and at the end is the

most uh attention

uh aware parts of your contents at the

moment. I believe it's uh it will be

better in this year because many um AI

research teams um have very uh efficient

research at the moment for understanding

uh the longer context but as all we know

uh LLMs also have some token limits so

they can't process it all and they have

a layout parser at the moment a user

agent

So uh beginning of your article and end

of your article is the most

attentionaware

uh parts of your uh page or your

content.

>> So how come how come you're saying put

the um the comparison in the middle of

the content?

>> Yes. Based on my experiments it's uh

affecting in a positive way. So I can uh

see the uh positive outcomes especially

on the traffic and visibility part.

I want to ask about your experiments in

a moment but is there a certain way to

format you you were saying comparison

tables? So like can you give an example

of what like a well done comparison

table for an LLM would look like?

>> Yes. Um it I don't like mentioning or

saying it depends but it depends. So uh

it I believe it's tying with the user

intent. So if the user intent of the

page is about uh some transactional uh

prompt or topic comparison

tables works well. Uh but if it's

informational let's say uh you can also

uh create some um

listings bullet points and uh also

comparison table works uh well but what

I'm uh trying to create uh most of the

cases yes I can say comparison tables

this

>> is there um well can do you ever see

instances where comparison tables are

difficult for LLMs to parse the

information in them where they mix up

information because maybe maybe

something is uh is like five columns

away from the y-axis for example if you

know what I mean

>> and you have like a lot of rows and

something's yeah something's a lot of

columns away you have a lot of

Uh yes, I I don't have a specific

formula for that. Uh but I just

>> that's a great question. Thanks. I know

>> from the from the instead of the y axis

from the left side from the so you have

yeah it's like you normally comparison

tables go from left to right. So you

have uh so yeah, you have something

that's many columns away from the left

side and several rows deep. And because

this is actually something that I've

been thinking a lot about in terms of

like very detailed comparison tables. Is

it better to do that or is it better to

do bullets?

>> Both works, but I believe uh tables are

the most efficient way. Yeah, they're

well, they're certainly better for

users.

>> Unless I suppose unless I suppose uh

>> it's so much there's so many columns and

rows, then it might actually not be as

great.

>> It's funny kind of the same thing for

Amazon for users.

>> Yes, I I remember a good case at the

moment. I'm running a personal VPN

affiliate website at the moment and you

know some um especially on the affiliate

site if you are running any affiliate

niche um many comparison tables are in

the uh contents and money pages and it's

uh it's because of the nature of the

ecom uh site. So if you remember um or

if you check any top performing VPN or

hosting comparison affiliate websites

those tables uh formats as um are

actually uh good at the moment. What can

I say just for an example?

>> That's great. Yeah actually this is this

was something that I was talking a lot

with Gagen Gotra about and I I think

you've talked to him. He gave me a bunch

of questions to ask for you which which

I'll ask but

>> I

>> we were trying we're we were trying to

figure out though if um yeah if

comparison tables were better or worse

for LLMs. So it's interesting to hear

because you have this this is really

something like uh actually I think now

would be a good time to have you share

your background and how just how much

research you have done into all of this.

Yes, for my background I have uh

experience in SEO more than 10 years a

decade now. Uh but since from the um

2023 actually uh the mid of the year I

guess I started working with the uh

first version of chat GPT the llams and

you know every marketer I believe uh

have has this uh emotion and purpose. If

I knew how to code, I can uh create more

efficient uh websites and maybe I can

buy my new uh Lambo or Ferrari any

moment

>> because I have I'm struggling to uh talk

and um complete new tax uh tasks with

the developers in the com company. So if

I know how to code uh I can build my own

next million dollar idea. So I also uh

had this opinion uh for a long time and

within the LLMs I finally uh have a

chance to create my own experiments or

websites. I don't support AI slope by

the way. This is very clear, but you can

uh find ways to

uh build your own legacy

uh with using AI uh and you can actually

produce uh more outcomes uh for your

personal brand, for your company or uh

affiliate website or any niche uh etc.

So uh

and I realized that uh I was working at

a company we were also targeting the US

market and within the search generation

um SGE exper experience. uh we realized

our traffic uh started to crash just for

the testing base and uh after that I

realized that by the way AI overviews

and uh many uh AI features isn't roll

out in Turkey right now unfortunately.

So we but the on the good side we are

still uh seeing the 10 blue links with

the feature snippets at the moment but

for the other markets I realized that we

are uh we were losing traffic but our

impressions uh impression numbers uh

were going high. So I realized what was

happening. So I met with the

um bunch of AI futures and um on the

that moment. So I I also love working in

a reverse engineering way. So uh here we

are today.

>> Yeah. Yeah. And for people who haven't

heard of Metahan, a lot of uh a lot of

very impressive

SEO people uh like Sean Anderson who was

on the podcast twice now. He's he's

mentioned you a good deal. Like I said,

Gagan has mentioned you a good deal and

you're followed on on X by some of the

biggest names in the industry. You are

very legit and I am so happy to have you

on the show. I have a cool question

because you so you have done a lot of uh

experiments for AI visibility, right?

>> Yes. Yes.

>> So what what are some of the most

surprising experiments that you have

done?

uh starting with an e-commerce website 6

months ago almost 6 months ago and we

almost had no SEO visibility then uh we

start created content and um

and I need to say I uh I believe in SEO

and uh I respect everyone who is saying

and AO and go is completely different

than SEO sometimes

you know it's uh just about money or

market positioning or etc. So I respect

them. Uh for my opinion, it's uh these

terms are uh tying with SEO very close

way. So uh and at the end of the day, if

you don't have any index pages on Google

or Bing, you are almost uh not visible

in other laps at the moment. And we also

know uh SER API case at the moment.

Google uh open just fill the case

against them. uh so it connects with the

uh SEO and I feel like uh for the AI

visibility part yes it's uh closely it's

very close with the semantics at the

moment and I I also believe Google has

the best technology ever on the layout

parsing part on the semantic part for

every part but the gener the generative

part is very interesting because it's

um produces

probabilistic and nondeterministic

results every time. Even you ask the

same question context almost stays the

same but you can see citation are

shifting for every time. So and it's the

bad part for SEOs and

>> it's even more volatile than Google.

>> Yes. Yes. Unfortunately. Yeah. So I deep

dive on this uh experiments and I uh

proposed

a new

uh term. It's not new term uh actually

but uh reciprocal rank fusion let's say

rank fusion it's very um it's actually

similar with the quer fanouts and I

believe query fanouts uh will be another

hot topic in this year we already

started to talk uh about it and some

tactics um works until if you uh are the

only one who's uh working on it let's

say querify us you found the all uh

contact content and topic gates gaps uh

at the moment and you started to fill it

and I

it's really interesting the uh early

companies who started investing in LLM

and AI visibility part they started with

the uh with filling those content gaps

and many cases are around. But I believe

uh as I mentioned at uh beginning hour

of um talk building a consensus

yes filling the content gaps and the

backlink part is uh still essential

because uh yes we have a new term named

citations but uh very similar almost

same approach with the uh outreach and

link building is also essential for the

uh AI visibility.

>> You you mentioned an e-commerce

experiment.

>> Uh yes sir. Uh almost uh zero visibility

on the SEO part and uh we started to

fill those content gaps but we started

creating with topic clusters uh at the

beginning because if I say Edward uh is

the best saleserson in the world uh

>> oh that's not true.

Yes, that's not true. But but but

uh

>> I can tell you that.

>> Yes. Let's say I created a new uh hot

blog post. It's trending and everyone is

will search for your name and they will

uh need to find some uh supporting

documents or information. If it's not,

you are not the best uh salesperson in

the world. So it's very similar and uh

we also

uh released many uh PR releases

uh then it worked.

>> What are you using for for press

releases?

>> Uh several options

uh but I will be honest. So um the top

PR release companies services at the

moment I tried uh almost all of them and

it helps for your visibility. It also uh

helps it also helps for connecting

entities

uh for your brand. Let's say if you are

running a bookstore and if you're

selling uh historical books, there are

some entities around it and with with

using uh PR releases, you can uh

actually connect all the uh entities

around your brand and you can see the

effects because it rolls out uh

thousands of uh web URLs. It helps a lot

to build a consensus.

>> That's great. Uh, you know, there's so

many different press release services

varying in price and people want to save

money. So, which which press release

distribution services have you tried

which are like a good price point but

also effective for entity SEO and

influencing LLMs?

Um yes there the there are actually

almost uh same websites in those

packages but I can say AP news and

Reuters

um

affects a lot at the moment for the AI

visibility.

>> Where where are you where are you buying

where are you buying your press release

distribution from? Are you using PR log?

Are you using PR newswire? Are you using

Are you using AB AB Newswire? I think

that's Yeah, AB Newswire.

>> Uh I can say a press newswire. Yes.

>> Which one is it?

>> A uh a press wire. Although

>> a press wire. Oh, EIN press wire.

>> Einh

Okay, cool. And you've used that a lot.

>> Yes. Yes. I can also recommend brand

push.co.

>> brandpush.co.

>> Yes.

>> Cool.

>> It's

>> build instant authority and rank on

Google. We write and publish your story

on major news sites to build trust and

rank your brand on Google and AI. So you

you've done this. I'm going to view the

packages. How much is it?

>> Yes. The the brand push is uh releasing

a created package. So uh less if you

compare less number of links but uh you

can see I I also started using uh some

uh link platforms like white press and

press uh so

>> oh press yeah Charles float was on the

was on the show last week

>> yes I chose uh so it's also uh effective

the most time consuming part is uh you

need to okay I will share my honest

method I will uh search manually

uh almost I guess 100 times in average

for different terms and I try to compile

the listicles uh from SERs sometimes I'm

using uh data for SEO API

um and

I create a list of the articles and I

search those platforms uh in white press

or press to find it. If I find it, I

send them a request for the link

placement or uh adding a new paragraph

for my brand. It it works

>> and it and this is influencing LLMs as

well. Yes, it's uh

>> it's helping with consensus.

>> Consensus. Yeah. Yes.

>> That's cool. What's your um actually now

now I want to ask like what's your

what's your stack like your most trusted

tools for doing SEO and for doing AI

SEO? Actually

nowadays I'm always trying to create my

own specialized

um mini tool with using cursor and

cloud and Gemini mix of them. Sometimes

JPT codeex uh is also fine but mostly

cursor and uh beside of these AI

platforms I'm using Samrush Hrefs a

vision art and uh

I also started using SE uh SE ranking SE

ranking so

>> yes yeah cool uh Gagen wanted me to ask

me that So, Gagan had a bunch of

questions. He wanted me to ask about

backlinks impact on AI SEO, which we've

already talked about. He said, "Content

optimization strategy to improve AI SEO

performance like content chunking. Is it

worth it or not?" And I have my own

opinion on this, but I I want to hear

what you have to say.

>> So, this is a debate right now,

>> which is crazy to me. I'm It's crazy to

me that this is a debate, but Okay.

Yeah. Go on.

>> No, no, no worries. I love Goen. And uh

what can I say? I

uh I published some posts in the last

year about chunking. Okay. Um

from a user perspective

um you can see bunch of text won't help

your user experience. So

>> you mean a big block of text is not

going to help your user experience.

Yeah.

>> Yes. And if you uh create some text

parts with some tables with some bullet

points, it also uh helps to increase

your average time on your uh web pages

which is okay for Google I believe. So

it uh also uh it will help to rank

higher on some uh search terms let's

say. And if you rank higher on the

search engines, you can be included in

the reranking process of the LLMs and it

also almost doubles your citation uh

chance in a probabilistic system design.

Uh so chunking is a great way to uh

provide better user experience from my

perspective. But

are there any direct studies mentioning

chunking? Okay, we tried chunking

chunking and we doubled our AI traffic.

I never see one. Maybe we can see in

this year I'm not sure. uh but chunking

is helping to

I believe actually to AI user agents.

Chpt has different bots, Gemini has

different bots even for deep research

process. So each of them uh actually

have different layout parsing mechanism.

So if you create uh some great chunks on

your page, uh it will also help to

process your um maybe

long block of uh text parts. Uh I

believe it can be processed faster with

using chunking.

>> My my take is content chunking has

always been a best practice for

copywriting and SEO. It's just it's

something that like if you there's a

million studies done on on how it is it

improves engagement. Readers like it.

Nobody likes huge block paragraphs. When

I see a huge block paragraph, it is very

hard for me to continue reading. Is

very, very challenging. For LLMs,

you can literally just copy an entire

landing page text, give it to the LLM,

and it will do a very very very very

good job parsing the information on it

despite it could it could be one entire

line of text, the whole page, and it

chachi BT will do just fine. I think

content chunking makes sense in terms of

chunking the topics because that makes

things easier to understand for an LLM,

but in terms of like visibly

chunking the content for specifically an

LLM, it doesn't make a difference. But

like you said, it makes a difference for

SEO. And when your SEO is improving,

you're going to have more visibility in

AI. That's

>> Yeah.

>> Yeah.

>> Yes. And what most people actually uh

for what they forget about it, Google is

a multimodel search engine. And I also

uh worked on some uh console cloud

experiments like you can also create

your own website specific uh very

similar to AI mode or AI overview search

engine for your enterprise company.

Let's say you have thousands of pages of

documents uh and you need a internal

search engine let's say for your company

for um anywhere in the world for your

all branch offices uh etc. Um Google all

Google is also selling its own uh maybe

lightweight technology of the search

stack and um in the discovery engine

product uh you can see some settings for

the chunking they have a uh chunking

limit I'm talking uh for the Google's

discovery engine product and uh headings

are also traveling with it chunk so

Google has the um technology it can also

connect with your headings with your

chunks which is very effective.

>> What does that mean?

>> Can you can you go deeper into that?

Yes, as we all know uh there are some

heading structures almost for uh every

web page URL and um Google has a

discovery engine product which provides

very similar um

with AI mode and AI overviews for your

any enterprise oper search uh needed

operations and there is a settings in

discovery engine on Google cloud console

uh which is uh each headings can travel

with each chunk. So which means

uh Google can understand any uh heading

structure if it's tying with uh the

chunk next after heading. So it's a

small feature but what I can uh learn

from this Google can also connect

headings with chunks if it's needed.

So so we can even see some uh these

little options uh on this product. So uh

Google has a really great AI search

infrastructure

at the moment and I believe check GPT is

uh following the next or perplexity

uh but most of us uh are just ignoring

Google is multimodel so they can

understand if any embedded video on the

page and if it's connecting with topic

or entity or they can understand any

images uh on the page but LLMs I mean

chat GPT perplexity uh and others not

Google ones um they mostly follow uh the

text semantics of the page

so I believe that's why Google and John

Miller is uh saying that time to time

chunking is not a uh is not an optimal

way to optimize your pages for lams for

go a uh of course I don't believe every

claim or statement from Google

>> yeah you can't you can't believe

everything

>> yes yes

>> uh what are okay if um you you talked

about this a bit but I I'm I'd like you

to go deeper just because uh like like

you said you said that you spent the

last year reverse engineering

>> I guess. Yeah. So,

>> uh I I like I'm just so curious like if

you were to share with someone just some

of the most surprising findings that

would make their jaw drop like like

where where you were sitting at your

computer looking at the results of an

experiment and you went what I can't

believe that like what would you share?

>> Okay, for the perplexed part I guess I

can mention it. uh for the perplexity in

the uh their search inf structure and

they don't have their own uh llm model

at the moment they are using chat GPT

gemini anthropic uh deepseek and others

um they are using very high embedding

similarity threshold for the queries and

for the user intent

uh because it can it's very dynamic in

Google and CHP but for the perplexity

and I I guess I there was a time

perplexity

was providing better results uh with

using ranking in the last year. I I

guess

uh there was there were some uh posts

and reviews around it and what I

revealed when I saw the perplexity

configuration

uh with listening browser events. uh

they actually they are using very high

embedding similarity

uh to catch and satisfy user with

finding the uh best results and

reranking them. Um

the second one uh I guess the biggest

black box is for me the reranking part

because every time just ask same

question and if it's not a cached query

or prompt uh for that time you can try

uh with a five minutes delay asking same

question you can see citations are very

dynamic and changing almost uh every

time for the ranking for the mentioning

block in the answer etc. So it's very uh

dynamic and blackbox and even you rank

higher at Google or Bing you can see you

are not mentioning it the chat GP that

was uh one of the most popular releases

in the last year. Yes, I I talked with a

VP of a very uh big enterprise company.

They they they were actually dominating

the Google and results in the first

page. But when we compare the uh AI

visibility uh of the domain, it was

relatively low. So that was very um

>> and and do you have any ideas uh why

they weren't being mentioned more in

LLMs

>> sometimes?

Um

we we tried many things in the last

summer, beginning of the uh summer and

uh what can I say? Some sometimes they

just uh

needed a part for the creating trust on

their pages. um sometimes case studies

or sometimes just running a PR release

campaign about stats statistics and uh

they were able to create and release any

statistic at the moment. So we tried

this approach and it worked well. You

know, at the end of the day,

doing proper SEO or AI SEO, it's about

getting sales. And it's it's about

converting. And a lot of SEOs forget

that SEO is so much more than just the

ranking. Because if you have a page that

ranks

>> and it doesn't and it's not converting

>> and it's it's not helping your top of-

mind awareness or it's not building your

trust so that people are converting

later on. It's not doing any of that

stuff. It's not I mean maybe you could

say well it's increasing your topical

authority but

you know that's like so far away. It's

uh you want to be doing things that are

going to get you sales. So, basically,

you're saying that you put out press

releases and did off-site SEO and

that off-site activity

actually increased your AI visibility

without your without really changing

much of your existing rankings or did it

also affect the existing rankings? I

actually imagine that it would affect

your existing rankings too.

Yes, it also affects the existing

rankings in a positive way. uh but

building a uh very strong SEO

fundamentals at the beginning is the

most crucial part because it also ties

with the crawling uh accessibility

creating um

great and satisfying user intent topic

clusters on your website for your

service pages, money pages, blog

contents even for the glossery part is

uh also helps for the increasing

visibility.

Um

so yes

uh the the offsite um SEO link building

or AI citation building and we we didn't

even mention Reddit by the way. So

Reddit was the one of the hottest

websites for the AI visibility and

unfortunately we uh saw many spam cases

in the last year. I don't remember an

exact one but you can

>> you can smell it. Yeah. Yeah.

>> You can you can you can feel it uh when

you look at the first site.

>> So Reddit is very essential. Now we are

talking about Kora Medium and I still

believe YouTube is very um underrated at

the moment.

>> Really YouTube?

>> Yes.

>> Actually, so this was this was a

question from Gagen. He said content

platform investment strategy to get

better results for AI SEO like YouTube

videos or spending more on creating

highquality blog text content. But yeah,

I'm curious about the uh the platforms

the yeah the platform investment

strategy.

This is this is a great question. Here

comes my exact answer. If you are just

running your own website, you are only

running one single domain. And for query

fan arts for rank fusion in chat GPT for

uh very similar to let's say knowledge

grab entities etc you are only trying to

compete with one single domain but if

you own uh if you manage your own uh

content stack and platforms

you are trying to compete with multiple

domains and I believe this is the best

way to uh show your ex um

experience, show your uh trust, building

your trust and you can also show your uh

or tell your case studies from different

uh content types, video, podcasts,

uh blogs, infographics, etc. So

>> how about how about LinkedIn posts,

exosts, threads, Facebook,

>> social. Yes, absolutely.

>> These these are also being picked up by

LLMs.

>> Yes, I saw many uh X threads at the

moment. It I can see on my uh feed at

the moment uh Facebook groups are

gaining more popularity for the parasite

SEO

>> uh

>> efforts. So these are these are changing

time to time but it's also very

effective if not everyone is working on

it. Let's say if you are uh in a very

small group only working on Facebook

groups to create create some parasite uh

SEO effect it will work but everyone is

if everyone is doing it it won't work it

also will be very competitive at the

moment

>> but but not everyone is doing it I mean

even even like it's just so few it's

it's 2026 and still so few people know

how effectively do parasite SEO. People

don't know to take your keyword and to

put it at the beginning of your post,

put it at the beginning of your video

description, video title, like all these

things. People don't know pe like Yeah,

it's so I think it's going to be a while

before people are even like I I don't

even know if people will really ever

>> SEOs will know, but like the common

person, the the common business owner

will not know.

>> Yes. in the last year

um Lily Ray conducted many experiments

on the uh LinkedIn LinkedIn articles. So

she published a new article and she ran

in the first 24 hours.

>> So it was very effective and I believe

it's uh still works. But what can I say?

If you uh identify those sources like

Facebook groups or medium or some

specific Reddit threads, subreddits uh

on HFS and Samrush and it's uh driving

more SEO visibility, it will probably

help your parasite SEO uh efforts and

you can um as I mentioned you can try to

combat with multiple domains with using

your brand name you can use YouTube and

YouTube is very effective and right now

uh we also know Instagram meta allot

Google bot crawl the whole entire uh

platform. So using SEO friendly uh

descriptions text uh lists is also

effective for

um AI visibility too.

If um if you're I have two great

questions, but I think I'll ask this one

first. If you are consulting somebody on

how to actually get more conversions

with AI visibility,

>> like specifically not just a AI

visibility for the sake of AI

visibility, someone wants to get more

sales or uh or calls or users with AI

visibility.

>> Okay. What are you telling them?

>> Still my tactic. I won't share the um

exact

uh way.

>> Come on, give us a sauce, Metahan,

please.

>> Yes. Yes.

I'll um

I always start with gathering more and

more questions as I can um compile

>> with with what more questions. What was

the word that used with getting?

>> Yes. Yes.

Okay, I

uh I need to work on my um introduction

for stealing my uh tactic right now.

Sorry. So, I I always start with the

questions, real user questions. Where

can I get them? and I try to find uh

hundreds of real user questions

from my uh audience where they are

active at the moment. And if you use

those questions

uh to create some synthetic prompts or

you can also target them. Uh it's your

wish or choice however you want. If you

uh pull up more and more questions from

the open web, it helps a lot for the

conversion and driving more uh attribute

sales from AI platforms

at

>> what what are examples of questions?

uh

any questions but what I can say um I

also group them uh exec in SEO

>> but are these are these like purchase

int are these purchase intent questions

specifically are these questions

involving the brand name are these yeah

>> non-branded questions so for example uh

I spilled m I spilled red wine on the

floor and I'm I'm afraid that it's going

to stain. What product should I use to

remove it? Like is it a question like

that? Or is it like um is uh is the

comet cleaner? I don't even know what

the common comet window cleaner uh good

for also cleaning mirrors like is it is

it stuff is it all of those questions or

is it like a specific type? Okay. I

always um

use almost more than thousand questions

for every case based on

>> and where and then how are you

formatting? Are you putting these

questions um just as like H2s on a p on

a page and you're doing multiple pages?

Are you doing blog posts? Are you doing

it in YouTube descriptions?

Uh first I create topic PL topic

clusters around them then supporting

content then you can use it in your

YouTube description your um in your

headings

even it can be your uh the next uh video

idea on Instagram res or Tik Tok. Tik

Tok is very popular in the uh US right

now and I see some patterns around some

e-commerce topics, Tik Tok is actually

very uh effective to increase your AI

visibility. What I believe is sometimes

we as SEOs or AI visibility uh people we

don't understand well the social uh

nature social media platforms sometimes

uh there are topics exploding in social

then we can see the uh search volumes in

the search engines or LLM. So if you can

understand the dynamic of these social

media platforms, we can uh even use that

platforms to uh catch some new questions

from real users. It helps a lot because

uh you will realize that

uh if you review hundreds of real user

questions from the open web from social

media you can understand as same as in

chat GPT or other LLMs context actually

isn't change so much but it's uh also

needed you need to niche down as you can

so even a small question uh can close a

sale.

>> Niche down as in make the questions more

longtail.

>> Yes. Yes.

>> More specific. Yeah. When um so you you

do audits like AI visibility audits for

companies, right?

>> Yes.

>> When you are doing an AI visibility

audit for a company, what do you start

with?

>> Um great question. First I start

crawling the website with using mostly

screaming frog. Uh so it's very uh

flexible for me. I use custom

JavaScripts with using AI APIs.

So I start with the crawling uh the

current health ad situation of the

websites. uh sometimes broken links or

just identifying the missing content

gaps the uh topic clusters or if uh a

bot is struggling to crawl the uh next

pages. Uh

>> how are you how sorry how are you

identifying topic clusters or like

missing missing topic clusters?

>> I'm I'm using custom JavaScript for

that. I didn't share in the uh public

but first I um

>> export the crawl results uh then put it

on Google drive and uh trying to uh

process the all custom JavaScript output

from AI APIs to identify topic cluster

gaps.

Okay,

>> I saw it the uh crawl.

Then I the best part and the most time

consuming part is creating the topics

because every embedding model of LLM has

different topics and understanding

context at the moment. Gemini has

different embedic model chat GPT

entropic and others uh different embedic

model and you can think they're working

like a corpus at the moment. So

understanding of each entity and each

topic is sometimes different than

others. So uh I start asking some

questions about brand and it's mostly

structure uh prompt templates and trying

to understand uh how their um visibility

in the uh core uh embedded model. Then I

uh compare with the topic clusters and

of course I continue with the uh

reviewing real user questions from the

web. Uh and I trying to niche down as I

can. Uh then compare the results and

create start creating a strategy uh

around it. It can be about query fan

outs uh missing content gaps or

identifying some opportunities for the

uh link building or AI citation building

part or Reddit part.

Do you think um do you think AI

visibility strategies

are changing rapidly or do you think uh

do you think there's really just like a

set strategy that has been working since

ChachiPT was released December 2022?

>> Uh it's it's changing very fast

sometimes. I just can't uh I I I just

want to argue with them how changing is

uh fast in a really dynamic way. Uh I

mean the OpenAI team, Google team etc.

But right now uh OpenAI announced a new

health app u in chat GPT. So it's like

your money your life uh situation at the

moment. I didn't experiment with play it

uh with well for the citations or

medical suggestions. So besides of this

right now we have an we have some new

agentic protocols at the moment there's

a shopping research in chat GPD and

Google announced UCP yesterday which

looks

>> how much have you looked into into the

universal commerce protocol?

it looks great but many of ACOs are uh

also mentioned that um I believe it's

also

uh aligning with the merchant feed at

the moment. So if you're uh if you're

technically uh crawable and if you

provide your product information with

the most detailed way uh in a structure

way probably it will help a lot for the

uh universal uh protocol at the moment

but I I didn't I didn't read the all

details

>> okay

>> when I uh read the uh general news and

some uh

official blog posts for it and we know

Google now has a new approach they

release a new uh blog article or video

uh for their newest features. So based

on my ex uh understanding the technical

part uh is very important and I didn't

like it. Yes, it's useful for users for

me, but it also I believe it will

eliminate the website part in the future

because why do we need to have websites

uh for agentic protocols if we can only

provide some

um agentic commands, checkouts and

product feeds? I don't know. We'll see.

I think it'll be I I have I've been

talking about this. I was talking about

it with Gogan on our holiday show

actually. But I think it at first it's

going to be a pretty steep learning

curve and then it's just all going to

happen at once is is cuz like a lot of

people still are not going to trust an

agent to do their shopping with their

credit card and uh like that's that's

just the truth. And then what'll happen

is like oh it'll it'll start working

really well. A friend will be like this

is great. Show the friend you see it

from another friend. Okay, I got to try.

Wow, this really is good. And now you're

telling other people and now everyone is

doing it because it's working so well.

But that could be a while from now. And

then also like some people just like

browsing. Some people like clicking

around and looking. And it's like it's

the difference between going to a

supermarket and and just like scrolling

in your phone and and picking out items.

Like there is a different feeling going

to a supermarket and and like looking at

these like shiny vegetables or packaged

meats and being like, "Oh my gosh, if I

buy this, I could combine it with this

and then I could make this and that' be

like I personally I actually like going

shopping. I like going grocery shopping.

I like cooking." And and there's like a

but yeah, I don't know. It's it's it's I

think I still think we're like a ways

away from agents like dominating this.

Um

yeah, and also it's important to note

like this, uh the UCP was developed with

Shopify. So I also think a lot of merch

merchants, they're not really going to

have to do anything extra. It might be

like a a setting that you tick on, but

it's like

>> if if you are using one of the most

common solutions for e-commerce, you are

not really going to need to do anything

extra because like the developers are

going to handle all of that.

>> Yes. Yes. And um Google has another

advantage here because um they know how

to parse a layout, how to click an add

to cart button, how to make a uh how to

complete a transaction because we are

sending these data as we ecom uh website

owners to Google directly with using

Google ads and more than 20 years I Yes.

So, and

um if you see the market situation at

the moment, yes, it's very exciting. But

please, we need to also remember these

generative AI part uh is disruptive.

Yes, that's correct. But it's also

killed news websites, recipe websites.

So, we don't know what's next for the

ecom. Uh, I don't actually I'm going to

dis I'm going to disagree with you on

that. I don't think I don't It partly

killed news websites, but Tik Tok killed

news websites, too. So did YouTube. So

did Instagram.

>> Yes.

>> A lot of people like to blame the

decline in clicks for news websites on

just AI, which is so false. Like so many

people get their news from YouTube and

from Tik Tok and from other and from X

and from threads.

>> Yes, that's that's great. It's also

connecting with the where your audience

also live in. So if they're

um just watching new videos always in

Tik Tok, it's fine. So I believe you

also need to open an account and show

your brand there as well as your

website. But for the Google parts

I I agree with your opinions um it's all

correct but for the most of the part it

Google is uh behaving

like a very uh monopolic way so which I

don't like it of course I love Google I

love using it tech uh but sometimes they

are pushing more on the independent

uh website owners or web masters or in

the developers. That's why I trying to

tell.

>> Have you done any purchases with agents?

>> Not yet.

>> Are you

>> I'm in Turkey. Sorry. Uh I mean I'm in

Turkey and I need to use VPN uh to

access some new

>> features because uh for the local feeds

it still needs to be developed. Uh yeah

a lot of people are a lot of people are

afraid of of agents taking like people

aren't going to go to websites anymore

but and I've asked this to a few people

are you shopping with agents and they

all say no.

>> So

>> yeah. Yeah. Which to my point now, I'm

not saying it's not going to come. I'm

not saying that. I'm just I'm saying

that it's going to be I think it's going

to be steep. It's going to kind of be

like a hockey a hockey stick curve where

it's like everyone's not going to do it

and then suddenly ow and then suddenly I

hurt my I hurt my wrist in the gym and

and I uh I had a bandage on yesterday

and I

>> so funny

>> snapped my fingers. Yeah. But but then

it's all going to come at once just just

like that. Are you using agents for

anything else? for like research for

SEO. Yes.

>> For content.

>> Oh yeah,

>> a lot.

>> Great.

>> For almost every every every point of

SEO research.

>> Um I can also recommend the Alibaba deep

research uh tool is open source. I will

share the link with you and we can put

in video description.

I sometimes I'm um

not regular basis but I'm trying to use

deep research time to time to uh collect

some new user pain points about SEO

about AI visibility check GP rankings

rank trackers etc. Um I'm using deep

search for uh SEO purposes a lot

directly from chat GPT Gemini cloud

perlexity

um all others and my favorite tool so

far is cursor

>> playing with APIs

>> is really nice and the most uh

challenging part is scraping

>> it's like we are now living early 2010s

at the moment scraping

is becoming very essential potential

surprisingly and not surprisingly. Uh

but scraping I never experienced before

in this uh level so far. Uh it's

actually very challenging. They are

blocking everywhere.

>> So are you able to are you able to get

around it?

Uh yes, I am rotating top and well-known

user agents in my uh crawlers. So if

anyone

of that is getting blocked by Reddit,

I'm trying to use uh something else. I'm

rotating them but still getting blocked.

So

all right, this is my last question.

Metahan, and you've been great. Thank

you again for coming on the show. My

last question is, so we talked about a

lot of things on this episode for

improving AI visibility, but if you were

to narrow it down to one thing that

people should do after this episode to

improve their visibility in LLMs,

what would you what would you advise

people?

>> Okay, great.

Um the best way

uh sorry the the quickest way is

checking your content freshness. First

check your content freshness and

especially for the last uh 180 days I

mean 6 months is actually uh think like

a booster score

u if you

uh update your content and republish it

without changing your URL you can see

the effect and try to use more questions

end of your content.

Try to use more questions at the end of

your content.

>> Yes, FAQs.

>> Okay. Are you doing um is are you doing

FAQs with schema? Are you doing FAQs

with H2s? Are you doing like the typical

accordion FAQ?

>> Mostly H3 and H4 and in a um text

format, not in

>> schema. Very good answer. Yeah, very

good, very good answer. Actually, um,

while we're on that topic, I know I said

that that previous question was my last

question, but I did want to ask you

because you were talking about uh how

harmful JavaScript can be for AI

visibility and there are web developers

who listen to this podcast um or people

who are just who are just making their

own sites with uh client side rendering

and um

>> yes.

>> Yeah. So, I I would love for you to talk

about this.

Um I'm

working hard on this AI stuff more than

a year. I guess it's almost uh 3 years

at the moment. Of course, I'm still

learning. Uh but what I realiz yeah what

I learned and exper experienced I start

playing with some uh uh GitHub scripts.

So please check your CMS or uh stack on

GitHub for look for some SEO related

packages and test them implement them

with your marketing team. So there are

some great uh React components, VUGS

components for uh also increasing the

server side rendering

uh as possible. Uh so you can uh also

check some services like pre render and

uh maybe you can also uh think about

creating your uh possible pages uh to

HTML static formats. So start research

uh on your research on GitHub then look

for other suitable solutions so you can

increase your crawability uh for AI

engines

>> actually. So the question that I asked

before you you advised on page stuff you

advised an FAQ and um what was a what

was the thing that you advised before

the FAQ?

before the FAQ at the end I uh also

suggested uh comparison tables

>> and just before

>> no when I when I asked when I asked what

um what was the one thing that everybody

should do so you said

>> yes yes content refresh

>> yeah content content refresh so updating

the updating actually just like

>> making sure the content is up to date

and then updating the date published and

submitting it to Google search console

>> yes I I published it uh blog post uh for

this topic and some researchers from IS

Japan they tried to rank some documents

uh with using AI as a reranker and they

didn't change anything but the dates. So

the uh fresh dates and uh close dates

ranked higher for uh most of the cases.

Yeah.

>> So they have a recency bias at the

moment and use the last 6 months

advantage.

>> You do see that. You see that in the

query fanouts as well. They're appending

the year in the query. The LLMs a lot of

the time will append the year in query

fanouts. You'll say what is like the

best. Yeah.

>> It it also affects grounding. So if you

have what is the best X in the 2026,

it will probably will be grounded. I

mean the LLMs will use web search and

you can see the results. So there are

some also preconfigured patterns uh for

the content freshness part. You can even

identify in quer fan out. I think one

huge takeaway from this episode as well

is just the importance of reputation

management and thinking about how your

brand appears when you are doing link

building or when just or when you're

being mentioned in any publication like

what language is being used to because

like we've been talking about consensus

is important. So what language is being

used to discuss your brand and your

products and your services across the

web?

>> Yes. And I I need to also mention that

there are actually big opportunities for

the um global scale companies because I

I live in Turkey. I'm sometimes asking

some Turkish Turkish questions and

millions of people Turkey is leading for

the Chat GP usage in the world right

now. So check GPT isn't successful like

in in English uh I mean the sentence

types the user satisfaction and you can

even realize some there are uh uh

e-commerce websites from United States

are recommending by chat GPT which is

not possible uh for Turkey. So uh for

other languages uh global websites have

also another advantage and we need to

see the improvement of these systems for

other languages as well uh for this year

hopefully.

>> Atan thanks again for coming on the pod.

Where should everybody uh find you and

follow you and and consume your awesome

stuff?

>> My personal blog metan.ai AI but uh feel

free to reach me out anytime on X and

LinkedIn.

>> Yeah. Yep. We're very active.

>> We can drop we can drop the links.

>> Absolutely. Your links will be 100% your

links will be in the description for

this and anything else that you want to

share too. Um if uh if you watch this

episode on YouTube, thank you so much

for watching. If you listened on Spotify

or Apple Podcast, thank you so much for

listening. This is episode 923

of the Edward Show. 923 days in a row

doing this podcast. This is going to be

an exciting week actually. Um, tomorrow

I am supposed to have Jackie Chu and

Lars Lofrren coming on the podcast to

together which is going to be crazy. And

then on Thursday I have David Quaid and

Jake Hundley coming on the podcast as

well to talk about schema and whether or

not it's important. It's going to be an

exciting week. So, thank you all so much

and thanks again to Metahan and I will

talk to you again tomorrow.