YT YouTube2Text

YouTube2Text → Transcripts → ChatGPT Is Building Its Own Search Index. Here's How to Rank in It

ChatGPT Is Building Its Own Search Index. Here's How to Rank in It — Transcript

by Edward Sturm · 11,142 words · 1,604 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00David, you you said something
0:01interesting which is not something that
0:03you normally say. You said that the LLMs
0:07are now building their own index and in
0:10the past you have said this is not going
0:12to happen or this is unlikely to happen.
0:14Um but so tell tell us what's going on
0:17David. Okay, so OpenAI did say a few
0:21years ago that they are going to start
0:23building an index and it's going to be a
0:24fiveyear um process. And
0:29I've also been looking at how for
0:31example
0:33expensive it is for these LLMs to um
0:36sort of like hit Google up for answers
0:38and also this the extreme inefficiency
0:40of it, right? Like I think a lot of
0:41people have been looking at how often do
0:43my pages get crawled by a certain LLM is
0:46that interest, right? versus is is that
0:49just part of the real-time synthesis
0:50that's happening. So there's a there's
0:52an RA ag there's a a query fan out um
0:57the indexes return back a certain number
0:59of so in that process if if you look at
1:02some of your sites particular if you
1:04look at Bing web master tools you'll see
1:06you've got 57 80,000 citations they're
1:09typically across a very small number of
1:11pages like I have a site where only 19
1:14pages are cited and so there's like
1:1657,000 crawl requests happening for
1:19these same 17 pages is and I don't
1:22update content that much. I don't think
1:23even a news article gets updated that
1:25often, right? Um, and so it seems like
1:28hyper inefficient, especially given that
1:31there's such a lag for every single
1:33site. So if you've got one site that's
1:35super fast and then five other sites
1:36that are super slow and so what I
1:38started to notice is that let's say when
1:41we invent a new um arena because that's
1:44always a fun place to figure out how
1:46LLMs work, right? because there's no
1:48existing content in their um corpus and
1:51they can't really expand their corpus,
1:54right? They're, you know, cuz it's it's
1:56it's filled with vocabulary, what
1:57they've learned during training.
1:59And so, it makes for an interesting way
2:01to figure out how do you influence them.
2:04And I've noticed that sometimes you you
2:07get cited way before what you what
2:11you're be what you're actually saying
2:12gets mentioned. Like I think I mentioned
2:13before when I was um trying to rewrite
2:16what LM was saying about eat and and and
2:19trying to get them to be less emphatic
2:20about it cuz they're they're literally
2:22convincing people that eat is a real
2:24thing that's really happening that
2:25Google's training um their algorithms to
2:28somehow detect eat that somehow if you
2:30mention I have 5 years experience they
2:32go like wow this guy's got five which is
2:34just you know it it it's just silly but
2:37anyway um it makes it makes a lot of
2:40sense that if you if you're going to
2:42cash this that cache becomes a a form of
2:44of of index and I can see that I was
2:47being quoted and attributed to the
2:50opposite of what I was saying. So I
2:52realized that my my page had now pushed
2:56other pages out of the way and it had
2:58met whatever criteria the LLM was
3:00looking for to be included but it was
3:02attached to a statement that I didn't
3:04make. And then in the an hour later that
3:07synthesis changes, right? Um and the
3:11message changes. And so it seems to be a
3:13process where first of all you're not
3:15included, then you're included, then
3:17you're included in the synthesis, and
3:18then the synthesis changes. And you've
3:21been watching the um King of AO project,
3:24which I have with great enthusiasm. And
3:27it's fun because anyone can come in,
3:29right? There's no barrier to entry. Um,
3:31you can build a YouTube video, you can
3:33do a press release, you can build a a
3:35site. And the the great thing about it
3:38is that it's not it's not connected to
3:40anything that's revenue generating,
3:42right? And a lot of people will say
3:44like, oh, it's not a it's not about, you
3:46know, there's not a huge amount of um
3:48competition. There's not a lot of
3:49backlinks involved. And I think that's
3:50what makes it accessible to all, right?
3:52Is that you don't have to come in and
3:53build a million back links, but you can
3:55come in and see what people are
3:57searching for. And I was going to share
3:59a screenshot with you earlier because I
4:00was looking at one of my pages that
4:02ranked. And one of the questions in Bing
4:04was why is AI so easy to manipulate? And
4:08I thought this is really great because
4:09this is somebody who's watching along
4:11and typing things in and thinking for
4:12themselves. And it it goes to show that
4:15a lot of people do believe that LLMs are
4:17research engines, right? And I think the
4:20the response they're giving now shows a
4:22lot of maturity in the last two years
4:24where they're saying this seems to be
4:25designed to manipulate because obviously
4:27there have been a lot of conversations
4:29it can pick up on. It can see these
4:31press releases but as the amount of
4:34content expands it becomes more real,
4:37right? And so back to the index
4:41question, while it's going to take Open
4:43AI five plus years plus way too much
4:46investment, especially when they they're
4:48so hungry for investment and Europe is
4:50almost I was talking to a guy I know
4:52from Limmerick in Ireland who's saying
4:53who's barking at how much money is going
4:55into these. And I'm like, so what if
4:58it's if they're losing 48 billion? Look
5:00at how big the transportation industry
5:02has grown since steam engines, right?
5:04It's a,000 100,000% in growth, right?
5:08What what what was $48 billion debt? If
5:11this company ends up becoming a trillion
5:13dollar company, it's actually cheap
5:15change, right? So I they they're
5:19definitely having to get to a point
5:21where their cash is becoming an index.
5:24They can't get into Google as much to
5:26tweak it. And as James Douly and also
5:29Charles and Kellen um proposed they're
5:32also relying a lot more on open crawl
5:34right or common crawl sorry. So co
5:37common crawl is providing a large index
5:40of pages and people are starting to do
5:43tests at what if they block themselves
5:45out of Google out of Bing. I think one
5:48of the experiments was for penalized
5:50sites but penalized sites remain indexed
5:52through just that's like how Google is
5:54efficient and the penalty or the
5:57classifier acts as a backs stop it
5:58automatically demotes the page from
6:00within the index after the event not
6:03before the event but if you can start
6:05playing with getting into common crawl
6:07and see yourself getting cited in chat
6:10GPT as it moves away from Google and
6:12it's got a lot of problems with this
6:14sort of like being handcuffed to the
6:15Microsoft um system.
6:18Um they are yeah building more and more
6:20of their own index and um I think one of
6:23the interesting things I've seen about
6:25the king of AEO is that ranking first in
6:28Google is not actually impacting
6:30perplexity and chatbing
6:34is so effective I had to make sure it
6:35wasn't against Google's rules before I
6:37kept doing it. It's a form of SEO I call
6:40compact keywords. Whereas most SEO
6:43focuses on putting up articles to answer
6:45questions, how, what, when, compact
6:47keywords focuses on putting up dozens of
6:49pages that sell to searchers who are
6:51actually looking to buy. These pages
6:54rank on Google and convert so much
6:56better than normal that when I
6:58discovered this years ago, I couldn't
7:00believe this was allowed. It's less
7:02work, too. The average compact keywords
7:04page is only 415 words. Compact Keywords
7:08is a 13-hour deep course on getting
7:11sales with SEO. A customer recently
7:13said, "Each lesson is dense with
7:16information. You're giving years worth
7:18of experience boiled down into 15 to 30
7:21minute lessons with no filler or fluff.
7:23I feel like I'm gaining a new
7:25superpower. Compact Keywords is about
7:27setting up an SEO funnel that brings you
7:29sales for years and years and years. It
7:32works with AI. It's less work than
7:34traditional SEO and it makes way more
7:37money. You can get it now at
7:39compactkeywords.com.
7:41Back to the podcast. has OpenAI made a
7:44statement saying we're building
7:46>> because I was I was researching it and I
7:48was um I saw there was uh I saw that
7:52like Nick Turley head of Chachi PT
7:55described the goal as serving about this
7:58was yeah this this was sworn testimony
8:03dur April 2025 Google antitrust remedies
8:06trial Nick Turley head of ChachiPT
8:09described the goal as serving about 80%
8:11of our graphic from our own firstparty
8:14index index and called it a lofty goal
8:17they were nowhere near hitting. He added
8:19that 100% was attainable in the long run
8:22but not an operational target. So that
8:25was April 2025 and this is researching
8:28right now literally with Claude. Then as
8:30of April 2026, OpenAI's help center says
8:34eligible workspace accounts can turn on
8:37offline web search which uses OpenAI's
8:40indexed and cached web content.
8:45>> Yeah. So they're it's like building a
8:47bridge from opposite sides of a canyon.
8:49your your cache is becoming more and
8:51more of an index because real time
8:54fetching and synthesis is you know um
8:56there's also um rag for retrieval and
8:59rag for training so where you can bring
9:02files in and use it to uprain which I
9:05know very little about um a good person
9:08to that is really interested in that on
9:10X is Ryan Jones
9:13and um Ryan is studying a lot um about
9:16search engines and um augmented
9:19retrieval and how indexes work, how
9:22vector databases work. What's
9:23interesting is that a lot of um the
9:26things he talks about they're standard
9:29practice in large search engines, right?
9:31as we move away from databases like SQL
9:33databases um that I for example
9:37one of the database formats I've seen is
9:39that schema helps understand entity and
9:42now I can see that that's where a lot of
9:44initial how do we get into LLM's uh
9:47operators came from and a lot of people
9:49were saying thinking that schema was a
9:51great way to define entities except I'm
9:54almost 99.99% sure that's not driving
9:57Google and and it was through a lucky
9:59accident I've been playing with my own
10:01knowledge panel which I was like
10:02skeptical as as to what could be done
10:05with it and I had to get it renamed to
10:08get it away from Dennis Quaid and I then
10:11started applying for edits like hey
10:13quote this from my Wikipedia page quote
10:15this from this page and I think it took
10:182 months for Google to change it and I
10:20thought that we were doing it by
10:22changing the same as information in the
10:25schema and then I realized I had broken
10:28a field either I'd put in a comma or a a
10:32bracket in the wrong place.
10:34>> Schema was wrong the whole time.
10:36>> My schema was wrong the whole time and
10:37got dropped. And I noticed it took once
10:40Google updated the schema the the
10:42knowledge graph entry it became visible
10:45immediately and all of the edits were
10:48revisible. It took them a long time to
10:50do it right because I I guess it's a
10:51manual process in an automated world.
10:53And I noticed it did that without us
10:56having any schema whatsoever. And I
10:58think that they effectively decided to
11:01not trust schemas because there's no way
11:04to validate schema, right? And we we I
11:08think you said on a previous podcast,
11:10how many people are are in the knowledge
11:12graph that don't even have a website,
11:14right? Like Elvis Presley for example,
11:16right? And so um I I I think it's a
11:20little naive to think that somehow
11:21schema is above editing, right? It's
11:24above um manipulation. It's not. It's
11:26highly manipulable. Um, there's
11:28definitely there's definitely the
11:30ability to say, "Look, this is my
11:32personal branded website. It's part of
11:34this if you want to consider it. This is
11:36my ex. This is my LinkedIn," which has
11:38always been there. But I think the
11:40people that are are pushing it and and
11:42trying to infer or create the idea that
11:45if you have a knowledge graph entry,
11:46you're more trustable, starts to get
11:48into weight. How did how did authors and
11:51writers, which is everybody, right?
11:53there's no legal. It's kind of like the
11:55issue with the press, right? Um, you
11:57know, the US Constitution, like a lot of
11:59other constitutions, protects free
12:01speech and and and the right of the
12:02press, but there's no there's no sort of
12:05official capacity for the press, you
12:07know, just because you're a anyone can
12:08be a journalist, right? Um, and so and
12:12and I don't think being a journalist or
12:14a writer or an author is automatically
12:17trustworthy. And I don't think that um
12:20Google requires everyone to have
12:23expertise and experience at the same
12:24time. And I think all of these things
12:26are just they're start starting to
12:28become blurred where we're blurring um
12:33certain industries, you know, requiring
12:36trust,
12:38the subjective requirements for trust
12:40and the fact that Google can't actually
12:43go and do these. And I think people are
12:45trying to oversimplify,
12:47but I think it's
12:50I I really think schema and entity
12:53schema is a large amount of nonsense in
12:56in terms of like it's fun to play with,
12:59right? It's great to have a knowledge
13:00graph, but it it's only going to show if
13:01people search for you, right? It's not
13:03going to make you rank better. You can't
13:05add a king of AEO schema to your page
13:07and suddenly start ranking because you
13:09are you say you're the king of of AO,
13:11right? which would be the only next
13:12natural step to go with it. Um, so I
13:16think it's it's fun. There's some
13:20ambition there, but yeah, it's not
13:21translating to reality.
13:23>> So with OpenAI building its own index or
13:26any other LLM building its own index,
13:29you know that you sh you you um share
13:31this great slide from Google's antitrust
13:34trial, which is we don't understand
13:37content, we fake it. And you know,
13:40Google has the number one browser in the
13:42world. They have the the best behavioral
13:46data that you can have into what search
13:49results people actually want to interact
13:52with, what search results people
13:53actually like. Um, and that is a great
13:56that and then detecting links. Um, all
13:59of that helps Google give authority to
14:02sites and and figure out who should
14:04actually be prioritized above others.
14:08But OpenAI doesn't have the number one
14:10browser in the world. They don't have a
14:11browser that I I don't even I think they
14:13retried they tried a browser. They
14:15retired it. Maybe they'll bring it back.
14:16I'm not I don't even know. Um how how is
14:20OpenAI going to get behavioral data like
14:22this? How are they going to like Yeah.
14:25How will how will Open AI judge
14:27authority if they're building their own
14:28index so that they can prioritize
14:30documents above one another? A common
14:33crawl uses the same fundamental model
14:36nearest seed, right? So distance from a
14:39seed of trust to as a as a sort of like
14:41>> Wikipedia.
14:42>> Yeah, like a baseline, right? And it's
14:45it's probably not Wikipedia. Um, but
14:49it's probably Wikipedia might be in
14:50there, right? But it could also be like
14:52the New York Times. It could be
14:53Microsoft.com. It could be Right. And so
14:56how far you are and your relevance to
14:58those steps, you have other problems,
15:00right? like these sites can only be
15:03relevant related to so many different
15:04topics and they start to to fall out. Um
15:09I think that if you look go back to the
15:12example of having a p a website with 100
15:13pages and seven pages are getting cited
15:15all the time and you're doing 57,000.
15:20But again, how many how many queries can
15:2219 pages actually impact? And it's
15:24probably
15:26five root maybe 20 different root
15:29queries. Maybe maybe they could each
15:32rank for three, but then you get 50,000
15:35prompts layered above that, right? So
15:38you now you now have a pyramid again
15:40where you only have to go to Google once
15:43a week per prompt to get something of a
15:47quality signal and you just have to
15:49adjust your top 20 pages according to
15:51that. That's still a far cry. that's
15:54still only scratching the top 0.1% of an
15:57entire index because the technical debt
16:00would be too much. The like you said the
16:02access to information which I think
16:04comes more from the actual desktop
16:06search right the problem with um LLM
16:10search is that there that the user isn't
16:13uh plugging in in um feedback based on
16:16and they're missing that loop. And so I
16:19wonder if Google's strategy has been to
16:22sort of like help deteriorate
16:24Open AI's quality and and access to
16:27information and also force them to spend
16:30more on the index issue. And I don't
16:32know if that's a good gambit by Google
16:35because when you synthesize the top 20
16:38pages and one of the pages is say a bad
16:41actor or it's self um supporting or
16:46whatever you want to call it that's
16:48going to be eliminated by you know
16:50statistical um agreement right
16:53consensus.
16:54And so I wonder if
16:58the fact that it just has to pick the
17:00top 20 pages. I mean, if you look at
17:02Bing and Google, most of the differences
17:05are like this page ranks in third place
17:07on Google and fifth place on Bing or
17:09ranks first on Bing and 15th on Google.
17:12So when you sort of like take all 20
17:14articles and synthesize them, that kind
17:16of washes out. And so I wonder if that's
17:19sort of like a really good growth hack
17:21for open AI just to say we can wing this
17:24for a while because our answer sounds so
17:27good. Um and again I think I really
17:30think that people um overestimate it.
17:32Like I one of the problems I see with
17:34people using it as an SEO tool is its
17:37ability to hallucinate.
17:40One of the the things that I I I I think
17:42I brought up when I first joined the
17:44show was if you take an example like
17:46saying to Claude or your favorite LM,
17:49who's the best SEO on Reddit? You you
17:52would think it's driving some process
17:54where it goes through Reddit and looks
17:56at who gets most up votes or who's the
17:58most active and but all it does is it
18:01runs a query
18:02top SEOs on Reddit and some of those
18:04pages that come back aren't even from
18:06Reddit. And so if you manage to rank
18:09like myself or James Douly do and create
18:12your own list, it will just take that
18:14list and regurgitate it back. And so
18:18maybe that's very obvious for that use
18:20case. But when you do something else
18:23like who's the best restaurant in Tokyo,
18:26who's the best this or who's the best
18:28credit card
18:30and it kind of understands some of your
18:32subjective preferences, maybe the answer
18:34is so good that it maybe doesn't matter
18:36as much, right? But as long as common
18:40crawl
18:41starts keeps emulating Google and
18:43they're able to tweak that top 1%, maybe
18:47their ind their cache index is good
18:49enough, right?
18:53Hey, I was having this conversation with
18:56with Harprit yesterday about if they
18:58were building um their own index and how
19:02much of a field day it would be for SEOs
19:05because you know something that
19:06something that you said on the podcast
19:08before is Google has decades experience
19:13dealing with SEO spam and tactics that
19:17SEOs have been using to game search
19:19engines since the dawn of time since the
19:21dawn of humanity.
19:23We have been out there manipulating
19:26Google getting our brands number one and
19:29open AI doesn't have this experience and
19:31it is a brutal world out there the
19:33things that some of us are doing uh and
19:36and so it I feel like if this is
19:39happening it could be a crazy field day
19:41for SEOs within uh large language models
19:44that are making their own index.
19:46>> Like a new dig moment.
19:48>> Yeah, like a new dig moment. That was
19:50hilarious.
19:51>> And Jackie and so Jackie Jackie is
19:54coming on the podcast. Jackie Chu is
19:56coming on the podcast tomorrow with you
19:58and he he was I think the first one to
20:00start talking about oh my gosh dig what
20:03what did he say? Like dig is a new
20:04Reddit. This is crazy. you got and
20:06everyone was creating their own like dig
20:09communities and posts and doing parasite
20:11and it was working really well and and
20:12they couldn't handle it and they shut it
20:14down and restarted it and and now it has
20:16now it's it's entirely a message board
20:19for AI news. That's what dig is now.
20:22>> Oh, okay. I didn't even follow up with
20:23it actually.
20:24>> Yeah. Yeah. Now it's now it's just
20:25entirely they restarted it as just a
20:27message board for AI. And you know
20:29what's really crazy is that when they
20:32relaunched Dig, it was the original
20:35founder of Dig. Um, and then Alexis
20:38O'Haneian, one of the co-founders of
20:40Reddit, who helped her relaunch Dig. And
20:42you would think that they they have all
20:44this experience with parasite SEO and
20:46and people using um Reddit for SEO, but
20:50it was like, oh, it was a
20:52>> weren't ready for it. And Reddit has
20:54actually gotten has only gotten a lot
20:56better in the last 8 weeks. They have
20:59gotten terrifying. I' I've so many
21:01friends on X who've had their accounts
21:03completely suspended. And these guys are
21:05not spammers.
21:06>> Um
21:07>> Wow. And you can you can have a look in
21:10some subreddits where
21:1330 40 50% of the accounts replying are
21:17instantly vaporized, just completely
21:19suspended out. Um, so Reddit seem to be
21:22doing a lot of similar similarities to
21:25Google where they're going in with a
21:26sledgehammer where they used to use a
21:29sort of um cobbler's mallet. They're
21:32just hacking things out.
21:35>> What do you think, David? So if if the
21:37if the LLMs are really making their own
21:39index and are they're really taking this
21:40seriously, what will doing
21:46what kind of tasks do you think will
21:48work to get more visibility within large
21:52language models that are actually making
21:54their own index? Like if people wanted
21:55to start preparing now, like even though
21:58this could be years in the future, what
22:00would people be doing?
22:03>> Help me answer that question. I'm going
22:04to come back to what we were just
22:05talking about, right? Like OpenAI not
22:07having access to Google. What what's
22:09I've always been surprised that this
22:11never came up on a conversation, but
22:13look at Claude, which uh uses Brave
22:15Search. And Brave Search's results are
22:18so similar to Google. It's terrifying,
22:21right? I I actually genuinely thought
22:23that Brave Search was just a white label
22:25on top of Google and just servicing it
22:27serving the same results and making its
22:30own ad revenue, but it does seem to be a
22:32a separate search engine. And I'm
22:34wondering if there's a way that they can
22:37it's kind of like the way Simrush and
22:39Afs have their own um page rank
22:41algorithm that you know for the for
22:43domain authority or site authority. And
22:45I wonder if that by just tweaking a few
22:48things they can get close enough to 90%
22:51of the results and that's good enough.
22:53So my answer would be keep doing it and
22:56then just look at how where you're
22:59ranking and the slight ranking
23:01difference and then see if you can look
23:03at the query fan out and if you're
23:07varying out too far you could you could
23:09always make your page but I would say
23:11continue doing the same thing even
23:13despite these changes because I think
23:15the pressure to copy Google so let's
23:18just go back a year and a half ago I
23:20think um open AI I was almost
23:25going to Google 90% of the time and
23:27doing a query fan out. In fact, a lot of
23:29people had said they'd even given up on
23:31Bing. And it made sense to me because
23:32they were, you know, they were having so
23:34many problems with with Google's with
23:35Microsoft's legal team. Remember, they
23:37wanted to buy that startup and they were
23:39negotiating to buy the the company out
23:41and Google went and offered the
23:42executives the money straight in cash
23:44and just left an empty shell with just
23:46the employees.
23:48Um, which I guess is kind of nasty for
23:49the employees, but um I think that they
23:53that Google then started to build up
23:55this moat starting with go to 100,
23:58starting with other things, suing SER
23:59API which failed unfortunately uh or for
24:02them I mean and I think that but I still
24:05think that the emphasis of trying to
24:08copy Google or using Google as a 1%
24:11quality or last mile quality indicator
24:14is still happening. Right? that's that's
24:16still feasible. It's also it also shows
24:19the the the
24:21change in search, right? I think the the
24:24white paper from um Shopify was
24:26incredibly interesting where they could
24:28see cuz they essentially the platform
24:30for all of the e-commerce shops, right?
24:32What are they hosting? 30% of e-commerce
24:34stores. They were saying that they're
24:36still getting the same traffic from
24:38Google, but they're getting a a new set
24:41of traffic from LLMs. So, it's not
24:44really
24:46it's not the same searches anymore,
24:48right? And that that's really
24:50interesting. Like I I've seen, for
24:52example, um topics that are very popular
24:56in SEO subreddits have a lot of search
25:00in Google and zero search in Bing or
25:02LLMs. So, I I have entire sites that
25:05have like 20,000 clicks in Google and
25:09zero I mean zero LLM um uh citations uh
25:15because those topics are for Google SEOs
25:17only and they're not entering them into
25:19LLMs which I thought is fascinating.
25:21Maybe we'll be able to share one day.
25:23>> Uh I'll share it with you but uh maybe
25:25we can bring it publicly one day.
25:26>> Oh man. Yeah, I want to know. So, are
25:28are you saying basically people should
25:30just focus on doing SEO for Google to
25:33get the most impact?
25:34>> Yeah. Um, and and maybe fine-tune it
25:37with Bing.
25:39And I I I think there'll be a lot of
25:41interest in common crawl in the next few
25:44months. I think a lot more SEOs will
25:46start to see can they can they query the
25:48index? Can they will they be able to
25:51build like their own search engine so
25:52they can check their own rankings? Maybe
25:54Sam Rush and AF should check out can we
25:57see rankings in there.
25:58>> Are you doing this?
25:59>> That'll be interesting. Um I'm not
26:01currently um but I am trying to figure
26:03it out. I think the conversation we had
26:05with James Douly and the last
26:06conversation with Charles Flo and Kalin
26:08so interesting the work that they did
26:10that that absolutely changed my mind
26:12about how you can get cited in the
26:15trained module right and they were
26:17saying it was down to um common call as
26:19well. I will say, you know, um the go-to
26:23the go-to redirects have just thrown the
26:26craziest wrench into rank trackers. And
26:29I was having this conversation on X just
26:31yesterday with Charles Float and Joy
26:33Hawkins and and um Charles said like the
26:36the rank trackers are going crazy
26:38lately. They're just not working at all.
26:41Um, and I was seeing the same thing on
26:43my end and you know, so I've had um I've
26:45had like one of the heads of Moz on the
26:47show and and I so I said it was
26:49screenshot for compact keywords and you
26:51know on Google search console my uh my
26:53page for compact keywords is number one
26:55as it should be. Um, and within uh, MA
27:00compact keywords there wasn't on page
27:02one and page one was about like medical
27:04devices when you for a compact keywords
27:07with their SER preview uh, with their
27:09SER analysis and so I sent a screenshot
27:11of this to to the head. I'm like looks
27:12like something's wrong and he's like yep
27:14there's some something is definitely
27:15wrong and I think this is going on with
27:18all rank trackers uh, is that they're
27:22having so much difficulty now. Um,
27:26>> have you noticed if you've got a SER
27:27report, um, I don't know how many
27:29different SER reports you have. I have
27:32maybe 40 different domains at the
27:33moment. Um, once a site starts to rank
27:37for common thread, right? So, let's
27:39let's go back to our our um SAS payroll
27:41software, right? So, if you imagine all
27:43of the variations of SAS payroll
27:44software,
27:46um, as it's running those queries, the
27:49same domain starts ranking on page one
27:52for everything. And then when you start
27:53manually checking or you look in GSC
27:55afterwards, it's completely fictitious,
27:58right? And so I think it's down to two
28:01things. Google's hyperpersonalization.
28:03So it's it sees you're searching for the
28:04same phrase and it starts to rank
28:07different sites the same but just for
28:09you. And that's actually mirroring in
28:12the rank tracker cuz the rank tracker is
28:13following the same patterns. And so
28:16we're noticing where um we can see sites
28:18ranking and then someone goes to check
28:20in an incognito window or they go to
28:22browser stack and it's like yeah we're
28:24on page five for this right. Um seeing a
28:27lot of oscillation and seeing a lot of
28:29pages where you're ranking for these
28:32keywords these keywords have very low
28:34clickthrough rates. They start to fall.
28:36I I'm seeing a lot of um people posting
28:38search console screenshots where clicks
28:41are flat but impressions are falling in
28:43the background. Um, and I think that's
28:46part of Google's uh sort of getting it's
28:50starting to react faster. Um, and it's
28:52going to be and and I I think as Joy
28:55Hawkins said um the idea of being number
28:57one is changed very much. It's it's very
28:59much about perspective from
29:01>> I think I think you got to use I think
29:02you got to use Google Search Console as
29:04a source of truth for your tracked
29:06keywords. You have to have that that's
29:08the source of truth.
29:09>> Do you use data for SEO or anything?
29:10like I won't be surprised if if more and
29:12more people are starting to use tools
29:14like that and then extract that data
29:15from GSC and build like a reverse um
29:18rank tracker. I think that's going to
29:20become more useful. We we had a we had a
29:22chat right after this talking about is
29:25this the does this represent an
29:27existential threat for SER trackers? And
29:30I think unless they start to adopt that
29:31they're going to start maybe they'll
29:33wait and acquire a startup who does a
29:35good really good job of it. But I think
29:37um they have to get over their issue of
29:40using search console data. But it's
29:43going to be interesting.
29:44>> You're using data for SEO.
29:46>> I'm not. I'm not. Um
29:50I I probably should be. I just I just
29:51have so much on. Um I really wish I
29:54could.
29:55>> Well, honestly, it's not like uh it's
29:58not the biggest blocker. That's the
30:00thing. When you work with a specific
30:02tool and you're very familiar with the
30:04output of a specific tool, you can look
30:06at the output and see that as relative
30:08to other output.
30:09>> And so unless there's really really
30:12tremendous bugs, just because you have
30:15experience with a specific tool, like
30:18you will look at the relative results
30:20and and it still makes sense to you.
30:23>> Exactly. And and we've gotten very used
30:25to like these very narrow search phrases
30:27which we didn't have 20 years ago where
30:31they were a lot more individualistic and
30:32reported uh more individualistically.
30:35But you kind of get a sense that for
30:38this group of phrases with this root
30:39phrase that's generally where you're
30:41ranking and it's going to change
30:44throughout the day. And most SER
30:46trackers can only afford to do one um
30:49sample a day anyway. And so
30:54as long as you're as long as you have
30:55eyes on what your general pattern is, I
30:58it's good enough. Um and I but I wonder
31:01how long that runway has. Right.
31:04>> I think a lot of people are going to
31:06have to start using the Google Search
31:08Console API.
31:10>> Absolutely.
31:10>> And and we'll be building their own
31:12tools.
31:13>> I think so. Um
31:14>> and I think that's in the near future.
31:16>> Mhm.
31:17>> I'm not looking forward to it cuz it's
31:19annoying. It's a pain. Like it think we
31:21had what we had I I would say what we
31:23had five years ago was great
31:26>> like stability was great. I also
31:28something that I something that I wonder
31:29is you know like there it's there's been
31:33this like quote unquote war against SEOs
31:36with with Google versus SEOs for you
31:39know the last several decades. Why
31:41hasn't Google done something like this
31:43sooner?
31:45Um, yeah.
31:47>> It's actually something that I said is
31:48like maybe they maybe like
31:54maybe they didn't want to disrupt things
31:56this much and they had to because of the
31:59large language models.
32:01>> I can remember when we could see the
32:03keyword
32:05for all of the transactions in Google
32:07Analytics
32:09um just like you you have with ads. And
32:12I remember when Google started taking
32:13that away and people were like, "Okay,
32:14SEO is over." I was like, "I don't
32:16understand how SEO is over. I understand
32:17how it's more difficult." But I can also
32:19remember going far back enough in time
32:21where we were flying blind and Google
32:22was only 30% of traffic um as it was
32:25still growing.
32:26>> Um so I'm somewhat comfortable with
32:29flying blind and we're where I think
32:31where I think the problem is that so
32:33many CMOs and CEOs expect that you can
32:35build an ROI table and it's just like we
32:37spend this and we got this and it's so
32:39easy. That's gone. a a company I was
32:42talking to in the last month, they were
32:45talking about, for example, what if we
32:46turned off branded paid search? And I
32:50said, fine, but you're not you're going
32:52to lose so much data. You're now not
32:54going to know who's bidding against you
32:55or how much because you have to be in
32:57the game. So, I don't know. I I think
33:02Google's definitely reeling from giving
33:03away too much information cuz at one
33:05point you could just use search console
33:07to build a website as a map, right? But
33:10that information is also out there in
33:12someone else's RSS feed, right? I think
33:14uh Julian Goldie um and Charles have
33:17talked about that a lot where you can
33:18just take an RSS feed and reverse
33:20engineer a whole website from it because
33:22all of the important data is in the URL
33:24structure.
33:26Um so I I don't know like I said it's
33:29really interesting how their war with
33:32open AI is shaping
33:35SEO tools and the question is like how
33:37much does it matter, right? Um,
33:39>> I think it matters for newer people.
33:42>> From which
33:42>> for newer people?
33:44>> Yes, it does.
33:44>> Mhm.
33:45>> Because like part of the way that we got
33:47good was being able to look at the data
33:49and figure out what's working and what's
33:50not working. And if you if you're a
33:52newer person and you don't have that
33:54accurate data, it's a lot harder.
33:56>> And also, if if you have a world where
33:58it's just AI, you now no longer have
34:00click-through rates, right? And I don't
34:02know that Google can use that much
34:04information from Chrome because of the
34:07European Union, right? Right? I know
34:08it's built in there. I know I know the
34:10capability to harvest that data is
34:12there. I just don't know how much they
34:14use it. Uh cuz if you think about it,
34:18you know, if they're definitely I don't
34:19think they're using it now. And there
34:21there are not Chrome isn't ubiquitous
34:24across all industries, right? And and
34:26all browsers. And so they've managed to
34:29keep the index,
34:32I think, pretty much the same for the
34:35last 5 years with just clickthrough data
34:38on their browser, right? Um,
34:42and so it's it almost it's almost like
34:46OpenAI and um, Anthropic and Gemini need
34:50old Google search in order to have that
34:52quality feedback loop in order to keep
34:55the subindex for LLMs in healthy shape.
34:59Right?
35:01>> Let me ask you and I'm going to I
35:03actually you know what before I ask this
35:05question I want to share why we're doing
35:06this episode. But I want to just talk
35:07about what happened before we before we
35:10did this episode and then I'm going to
35:11ask you. So we actually we had a geo
35:13versus uh SEO podcast right before this,
35:16right? And um you know this is this if
35:20you're a frequent listener of the
35:21podcast you know that the geo versus SEO
35:23debates are a recurring segment on this
35:27show because they're fun because we have
35:28these people on who don't really know
35:30much about how things work and they're
35:32kind of shilling go and and they're
35:34misrepresenting SEO which makes our
35:37industry look bad. Uh and obviously we
35:40don't like our industry being
35:41misrepresented and so we have these
35:43debates geo versus SEO. And so we had on
35:46this woman and I won't say her name and
35:48she was shilling, you know, she was
35:50shilling this and how SEO is so
35:52different. And within the first 10
35:54minutes, she got bodied so bad from
35:57David that she's like, I can't can we
35:59can we stop this? I I give up. I don't
36:02want to use this. We can't do it. And so
36:04and and please don't pub can you not
36:06publish it? I'm like, yeah, you know,
36:08I'm I'm nice to we're nice to guests who
36:09come on the show. And so uh we're like,
36:12oh shoot. So, we got to make an episode
36:14for today and and something that you but
36:17something that you had said um early on
36:20in the geo versus SEO debate that we
36:23can't publish unfortunately because you
36:25won so badly was that uh was that OpenAI
36:29is building their own index. And so, um
36:32I think I want to finish with this
36:34question. David, if large language
36:37models are making their own index, do
36:38you think do you think actually doing
36:42geo or getting visibility in large
36:44language models,
36:46do you think there will be a lot of
36:48loopholes at the beginning, a lot of a
36:51lot of areas for manipulation that the
36:54AI companies are going to miss? And do
36:57you think it'll be just like a field
36:59day? I do think it'll be crazy easy for
37:01people like us, people with experience
37:04um as they as they gradually rely on
37:08more of their own index that they are
37:10building.
37:11>> I think that the the person who was
37:13meant to be on the show, right? Um and
37:14and just to be fair, I asked some very
37:17basic questions. She made some massive
37:19assumptions, right? She came in with the
37:21LLM out there researching every brand
37:23and building this knowledge base about
37:24it. And
37:26then she said, "Well, look, I'm actually
37:28>> But David, you just said that. You just
37:29said
37:30>> Mhm.
37:30>> that they're using their own cash.
37:32>> I I I think they they're tweaking their
37:34own cash, right? They they have their
37:35own cash or that's an automatic
37:37byproduct of what they're doing, right?
37:39Um, and if you I think we started off
37:41with the report by peak that showed that
37:43YouTube and Reddit had had come down and
37:46I was simply making the point that there
37:49or I was trying to set up for the
37:50premise that there is no objective
37:52trustable resource on the internet,
37:55right? And Reddit and YouTube are
37:57certainly not examples of that, right?
37:59Neither is Wikipedia for that matter
38:00because Wikipedia is so dry, right?
38:03Lacks subjective input.
38:07So we saw that one of the things that
38:09that triggered that I realized that um
38:12chatbt had moved more towards its cache
38:14and less on Google was the fact that
38:16Reddit and YouTube had dropped so much.
38:18But YouTube's not actually a useful
38:19place for an LLM to synthesize data
38:21because first of all YouTube only
38:24creates closed captions for or or
38:26content scripts for under 45 minutes.
38:29And somebody asked me on X, how did you
38:32how why is there no AI overview for your
38:34video? and I said because it's over an
38:35hour and a half and so it doesn't get
38:37close captioned. So all of that is an
38:39outlier. Plus it's way too much text to
38:42synthesize in real term. So I think chat
38:45GPT just made the or the decision we
38:48we're not going to include all of this
38:50content. Right. The other problem is
38:52that the way Reddit is structured, we
38:54see it as one page in a browser, but
38:55it's actually a number of pages and you
38:57have to API call fetch them all to
38:59synthesize them. And I don't think it's
39:01a good place either. And so
39:04the problem that people have about how
39:07LLMs work and that there are these that
39:11there is somehow an ability for an LLM
39:14to go and do all re all all this
39:15research and arrive at the truth. That
39:18is a mistaken premise that people just
39:21need to get over. I know it sounds safe
39:23and it sounds lovely. It's not it's not
39:26a realistic outcome and Google said that
39:27in their most recent white paper and I
39:29think people are openly ignoring it.
39:32Right? So I don't see just because Open
39:35AI has dropped Reddit and YouTube
39:39from the veneer. I don't think they were
39:41I don't think they were utilizing them.
39:43I think it was a great marketing sound
39:46bite for people to look like experts.
39:48Go, oh yeah, we analyzed 15 million
39:50prompts and YouTube and Reddit were
39:52always number one. So what? Because the
39:55immediate outcome we got was like this
39:56person who wanted to come on the show
39:57and say, "Well, using very broken logic,
40:01well, YouTube is always quoted.
40:03Therefore, if we get your YouTube video,
40:05you'll always be quoted, right?" And
40:06it's just like that's just not how it
40:08works. The same problem Reddit has. Just
40:10because Reddit's quoted, quite often
40:12they are subreddits from 2, three years
40:14ago, which you're not getting into
40:16anymore because they're archived. Um,
40:18you can't just say because it's cited
40:20that by being publish by publishing on
40:22it, you're going to be relevant, right?
40:25Um,
40:26so
40:28I I I I think all they're doing is
40:30they're removing like the veneer of it.
40:34I don't think that they were really
40:35relying on it.
40:37Do you think if the LLMs are making
40:40their own index, getting AI visibility
40:43will be a lot easier because their own
40:45indexes will be so much easier to gain
40:47than Google's index.
40:49>> They're following Google, right? Because
40:51>> they're following Google, but they don't
40:52have the behavioral data that Google
40:54has. They don't have the years of
40:56experience dealing with spam that Google
40:57has, the decades of experience.
40:59>> It's the same thing. Look at the car
41:01industry. You know, yes, Mercedes might
41:03sometimes be 20 years ahead, but
41:05eventually everybody like, you know, the
41:07the anti-breaking system was built by
41:09Bosch. Bosch sells brake systems to
41:12everybody. The system that looks at how
41:15much water that's um throws infrared
41:18light into your windscreen and based on
41:19how much is leaked, they know how much
41:21water is on your windscreen. That's made
41:22in Limmerick, Ireland, right? So, you
41:25can all of those technologies eventually
41:27become available to everybody, right? So
41:30I think that Google's um
41:33IP and their ability to um
41:37in other words at a surface level if you
41:40take one out of every 100 searches and
41:43look at the search results you get the
41:44benefit of all of that quality
41:46engineering right
41:50even with their go-to 100. You're saying
41:54you're saying that within
41:56you're saying that within uh open AI
41:58within an open AI index you're saying so
42:01you're so you're wait I I'm sorry I I'm
42:04not completely following are you saying
42:05that an open AI index from like is going
42:11from the bat like just from the start is
42:14going to be very very close to Google's
42:16index. I'm saying that currently
42:19what even though they're not running to
42:21YouTube at every for every instance,
42:24right? Again, it comes back to why would
42:25you why would you fetch the same page
42:2757,000 times in a in a 30-day period or
42:31a 90-day period
42:33um for synthesis. You have a cache.
42:37You've already you retrieved the page an
42:38hour ago. Why would you retrieve it
42:4057,000 times? It doesn't make any sense.
42:42It's not going to be fast enough for
42:44real-time synthesis. So cache already
42:46exists. So if you've got 57,000
42:51um citations from the same page, it's
42:54likely that that one page is servicing
42:57um what what's 3,000 searches. So you
43:02only have to get 10 searches from Google
43:05and you can update 100,000 results and
43:07use it as a quality last mile quality
43:10insurer. you don't have to do every
43:12single search. And so that saves money
43:15and saves load. So they can take their
43:18own index and assume that they don't
43:21need to worry about like let's say James
43:25Douly's page is number one today and
43:27then tomorrow it's two and Julian is one
43:29and then Dave Quaid is coming in at
43:30three and then Vus comes in and replaces
43:32them. They're synthesizing all five
43:35pages. It doesn't matter what order they
43:37rank, right? So you don't have to take
43:40every query and go to Google because one
43:44the same index is probably supporting
43:4610,000 prompts and so you just need to
43:49know which are the top 20 and then pick
43:52which ones.
43:53>> So let's let's ask let me ask you this
43:55then. So they okay fine let's say let's
43:57say they have like several dozen pages
44:00that they're looking at because you know
44:02they they like yeah but how you still
44:04have to prioritize information above one
44:08another that you so like if if page A is
44:12saying something and page B is saying
44:14something different and page C is saying
44:15another thing different how how are they
44:18prioritizing the information? Are they
44:20still using a ranking system and then
44:23just getting it wrong? So that's what I
44:25mean. I don't think Google's ranking
44:26system ever helped with that level,
44:28right?
44:30>> It it it helped user like for for
44:33searchers before LLMs for searchers who
44:36weren't necessarily clicking on every
44:38result on a page on a SER. They were
44:41just going to the clicking on the top
44:42one and then getting their information.
44:45That's what ranking did. And and we know
44:47and you you and I we know the
44:48click-through rate between position one,
44:50position two, and position three just
44:52drops off completely. Uh, so
44:54>> I guess what I'm saying is I don't know
44:56that Google putting a page in position
44:58one, two, or three meant that the
45:00message in the page was better than the
45:03others,
45:04>> right?
45:04>> It it that's not Oh, for sure. Like for
45:08sure, but the LLM still needs a way of
45:11deciding which information is most
45:13authorit authoritative.
45:16>> True. But um I what they're clearly
45:19doing is looking at the um 20 documents
45:22that are in the fan out. Right. So
45:25>> So do do you think they'll just do
45:27consensus or what?
45:29>> They they are just doing consensus.
45:30That's all they can ever do. I So I'm
45:32saying that whether the page is first,
45:34second or third I don't think impacts
45:36its input into the uh synthesized.
45:40Basically, basically the way to game AI,
45:43it hasn't really changed it and it
45:45wouldn't even change if they were making
45:47their own index. It it would it's it's
45:50just have as many of your own pages in a
45:54SER as you can because you don't know
45:56you because OpenAI SER might be similar
46:00enough that they're still going to have
46:02a bunch of your pages in and you want to
46:04have the consensus.
46:05If it's if the query is about your
46:07brand, it'll probably take your content
46:10first, right?
46:12>> I think I think it'll be easier though
46:13because I think that there's going to be
46:15I think I think there's going to be just
46:17a lot of loopholes because again, they
46:19don't have the the decades experience
46:21dealing with spam and people like us are
46:23going to uh we're going to discover
46:25things ourselves or we're going to have
46:27people on the show who have discovered
46:29things and
46:30>> EMDs are more important than ever.
46:33>> Yep.
46:33>> 100%. satellite sites, satellite sites
46:36that that don't look like your own site
46:38and say things about you, which might
46:40not.
46:41>> One of the things that they
46:42>> the most thing, but people do it.
46:45>> One of the things I'm old enough to
46:46remember cuz I was a software engineer
46:47in Dal and I would need to look up why
46:49Windows NT4 crashed or how I built this
46:52how I got this driver to work or how,
46:53you know, the plugandplay was barely,
46:56you know, it was a Microsoft concept,
46:57but it wasn't real. And the problem we
46:59had is if you went to to Microsoft or
47:01you went to Yahoo and you did a search
47:03about Windows NT, the first 100 results
47:06would be for Microsoft.com.
47:08One of Google's biggest innovations was
47:10just to reduce every domain to one entry
47:12and one indented entry. So that sounds
47:16small, but that was a massive leap,
47:19right? all 3,500 other search engines
47:22all worked the same until Google came
47:24back and reduced the number of results.
47:27That innovation is still present in um
47:31LMS today where they very rarely site
47:34more than two pages from the same
47:35domain. So while that's been built into
47:38Google and they've harnessed it and
47:40they've learned how to use that, you can
47:42copy that very quickly, right?
47:45And so you can copy other innovations
47:47like that very very quickly. Um the same
47:50way we reverse engineer Google, so can
47:52Open AI, right? They can test for the
47:55same things. And so
47:57>> they'll continue to be the gold standard
47:59whether or not Google lets them copy
48:01their data. And that's what I I think is
48:04interesting. Will will Google's gambit
48:06pay off or can they actually moat out
48:09them sufficiently?
48:11Well, what's interesting is that uh is
48:14that the output from Gemini is so awful
48:16compared to the other LLMs. It's really
48:19and and Gemini Gemini has the it has the
48:23best index. It has the best information.
48:25It by far it has it has all the Google
48:28properties, all the Google platforms,
48:30has everything, but their output is so
48:32much worse than PT and so much worse
48:35than than Claude and
48:37>> so bad. I think even perplexity is
48:39better and and I think that
48:40>> it's crazy. It's crazy how bad it is.
48:43>> And that underlines my point that if
48:45open AI can give a better synthesized
48:48document,
48:50the rank order of the top 20 50 pages
48:53doesn't matter that much. And so that's
48:55why I think this is an interesting
48:56gambit for
48:56>> you know what's a you know what's a
48:57great tactic and this also is this is
48:59probably definitely gay hat and I was
49:02talking about this with um Harpit
49:04yesterday. It's like uh making up your
49:06own awards and then giving yourself an
49:08award and creating a site for like
49:10creating like a the found this madeup
49:13foundation that issues the award and
49:14it's like and you know that's happening
49:16a lot right now with the king of AEO
49:18competition. That's like the perfect
49:19example. And the honestly the the most
49:22straightforward example is how when the
49:24competition kicked off, it was just
49:26James Douly and Jper Nissan. And James
49:29Douly put out a $6 press release using
49:31AB Newswire saying, you know, he's
49:33created this thing called DEEO. And
49:35previously, he's a king of he's a king
49:37of AEO who creates this new thing called
49:38DEEO. Um, and he creates this press
49:41release, six bucks. Then Jasper Nissan
49:43gets a $66 press release with EIN, EIN,
49:47PressWire, Newswire, I don't remember.
49:48Newswire Press Wire, but EIN. He gets a
49:50$66 press release and says that there's
49:54this crowning event at Lee Sports
49:57Village at at a stadium. There's a
49:59actual real crown. He says that there's
50:01and obviously it's made up. Anyone who
50:03anyone who's
50:05anyone who's normal can understand that
50:07it's a madeup a madeup thing. It's not
50:09the crowning event did not happen. There
50:11was they were not at this this sports
50:13village at this stadium, but the AI
50:15overview said that they were at this
50:18stadium at the sports village. There was
50:20a real crowning event. And it's really
50:23really really easy to trick the LLM. Um
50:27and and like you create your own award
50:30your your own uh organizations that
50:33issue these awards and you and the award
50:35is targeting a keyword like the king of
50:38AEO for example or it's like best app
50:41for whatever like you this you get you
50:44get the award for best uh email tracking
50:50best email tracking app the email
50:52trackers of America issued the award to
50:55David's email tracking app. Best email
50:57tracking app. And
51:00>> that's how how the internet's evolved,
51:02right? And the
51:03>> and you create all these websites and
51:05it's like a game of telephone. You
51:06create all these websites around this
51:07one award and everyone's writing about
51:09this one award and then when when the uh
51:12large language model is looking at a
51:14SER, what is it seeing? It's seeing all
51:16these different all these different
51:17awards. It's crazy.
51:20>> It is. And and I think it it also goes
51:22to show that that people who think that
51:24consensus or mentions or information
51:27gain somehow overcomes these problems.
51:30It doesn't,
51:31>> right? It doesn't. It just you have to
51:33just create enough information until it
51:34becomes real unless you were physically
51:37there.
51:37>> Well, then the two things are that's if
51:39that's if you're going after a uh
51:42competitive a competitive SER. The other
51:45option is to find the less competitive
51:47SERs that not everybody is targeting.
51:49Exactly. And with query fan outs, those
51:52get created all the time because like I
51:54said, I I think the the mistake being
51:55made by the LMS is is the amount of
51:58silence they're putting into this one.
51:59It's growing so much misinformation on
52:01the geo side. And that's why I think
52:03it's important to to tackle these people
52:05who come in, you know, not doing
52:07homework and making up these grifter
52:09stories about how LLM's out there trying
52:11to rank your brand and then coming on a
52:13podcast having done zero homework and
52:15then blaming everyone else for it. is
52:18ridiculous.
52:19>> And you're a YouTube expert and you've
52:21got the lowest resolution camera and we
52:23can see that by looking at your YouTube
52:24channel.
52:25>> This is a guest that we had right before
52:27this. It was so bad. It was so bad.
52:30>> And throughout the the blame kept
52:34shifting like, "Well, is that good
52:35enough for you, Edward?" And it's like,
52:36you're you're the one who's presented
52:38yourself as an expert, right? And yet we
52:41look at any of your previous YouTube
52:42videos, you've got 240 DP, you know. Um
52:47>> Yeah. Yeah. For her. Yeah.
52:49>> Um
52:49>> to be fair, my green screen is cutting
52:51out. So So uh I know I'm not perfect
52:54myself.
52:54>> We can hear you, right?
52:56>> Yeah. Yeah. Yeah. Yeah. She had
52:58tremendous audio issues and showed up
53:00late. It was It was not um It was not a
53:03good It was not a good scene. Uh cool.
53:07This was This was fun. Too bad that geo
53:10debate didn't work out. You know,
53:11actually something else. So, this was
53:13two geo debates in a row that didn't
53:15work out because David bodied both
53:18guests. So, we had another geo debate
53:21last week with this guy in Florida and
53:23this guy started he was cursing. He was
53:26insulting David. He was screaming. He
53:29was spitting it. I'm not even I'm not
53:32making this up. He was so angry. His
53:35face was turning red like a tomato. And
53:38the shouting so much.
53:39>> He was shouting so much and these were
53:41>> his audio was cutting out
53:42>> and and he asked not to use it. Yeah,
53:44his audio was cutting out as well
53:45because he didn't bring headphones like
53:46I told him in the emails five times and
53:49this this was two geo debates in a row
53:52where you completely bodied the guests
53:55and then they asked us not to use it.
53:58>> David, you got to take it easy. You're
53:59too good. You're too You're too good.
54:01You're too good. David,
54:02>> you can't come on a show where you're
54:04potentially going to meet meet 10,000
54:06people and think a sales pitch is just
54:08going to wash over, right? Do your
54:10homework, right? You can't just come on
54:12a show and say, "This is how it works
54:13because it suits me." And then if it
54:15doesn't if I don't hear what I want to
54:17hear, I'm just going to melt down like a
54:19child. Right? This is reality, right?
54:22You're you're if you're going to make a
54:24claim, you need to support it. And and
54:27that's just that simple.
54:29>> Yeah. I mean, I will say that we have we
54:31have had some um guests for the debates
54:34who they actually realized that they
54:35were wrong and they they said, "You know
54:37what? I'm not right on this. David,
54:39you're right on this." What should
54:40people do?
54:41>> Respect.
54:42>> Yeah. Respect to those people. Uh
54:45because you're just, you know, David, if
54:47and we've said this to to anyone before,
54:49but if you're listening and you want to
54:51try going up against David and and get
54:52your butt whooped, you're welcome to to
54:55send a pitch.
54:56>> And I'm not an expert in this, right?
54:58It's just
54:59>> Okay.
55:01Yes, you are. Come on. Come on. Uh,
55:05you're you're kind of an expert. We're
55:07But I mean, I don't know. I think we're
55:09kind of experts.
55:11>> I'm not an LLM engineer, right? And I'm
55:13not trying to be one. Um, but I'm and
55:16and that's the problem. If you can't if
55:18you can't meet at our level, but you're
55:20going around telling people that's how
55:22it works, that's problematic. The the
55:24problem is when these people
55:25misrepresent search engine optimization,
55:28the disinformation campaigns that people
55:30will do against actually I got this um I
55:33sent this to Matt Diamonte like an hour
55:35and a half ago. I got this uh
55:37testimonial for compact keywords. Okay.
55:40>> And this guy, he'd been using compact
55:41keywords for a couple of months. Um this
55:43was he started November 2025. Things
55:46were going up and up and up. Every month
55:48was better than the last month. And in
55:50January of this year, he's like, you
55:52know, I'm a business owner. Like, why am
55:54I doing SEO myself? He's like, I learned
55:56it. It's great. Why am I doing SEO
55:57myself? I'm going to go with an agency.
55:59And I I'm going to put this this
56:01testimonial on the landing page
56:02hopefully by the end of today. Um, and
56:05maybe I'll even just include it in this
56:06in this podcast. He's like, I'm going to
56:08go with an agency. And so, he goes with
56:10this agency. The agency is saying, do
56:12everything opposite of what I'm saying.
56:14They're they're like they're like, don't
56:16listen, Edward. We're going to do
56:17everything different. And month one is
56:19abysmal. Month two is even worse. Like
56:22month three is even worse. He just got
56:24went down down and down and down down
56:26and he had to fire them and now he's
56:28building up himself. So I started
56:29following Edward about November 2025. We
56:33had just had our best month because of
56:34Black Friday and applied I started
56:37applying his principles into our SEO. So
56:41pretty much compact keywords what he had
56:43been giving us on social media.
56:46December, good month. January, better
56:49month. February, even better month.
56:51Probably our best month ever. Anyway,
56:54around January, I thought, I'm a
56:56business owner. Let me outsource this.
56:58So, I scouted around for an SEO agency.
57:01They contradicted everything Edward
57:03said. Um, I was naive. Didn't really,
57:08you know, these are guys I'm paying a
57:09lot of money. Edward's just a guy
57:11following on social media. I trust these
57:13this agency. First month sales were down
57:17a third. Next month half half again.
57:20Just in absolute freefall when they
57:23contradicted everything Edward said.
57:26Anyway, back on the bandwagon and we've
57:27been building up again. So Edward,
57:29thanks very much for all your help.
57:31There are all these people that are
57:32misrepresenting our I share this this
57:34story and yes that they're an SEO
57:36agency, but there's all these geo people
57:38who are misrepresenting our industry who
57:40are going to get those negative results
57:42for the people that they are quote
57:43unquote helping and and trying to make
57:46us look bad in the process. And that's
57:48what I think you and I have a problem
57:50with.
57:51>> Yeah. And it's just they won't update
57:52their mental framework, right? They're
57:54just like, "This is what we've learned
57:55and we're just going to keep running
57:56with it."
57:58>> Yeah. Hey, tomorrow's tomorrow's podcast
58:00is going to be great. Jackie,
58:02>> so looking forward to it, Jackie. Wow,
58:03what a legend. I'm so excited for it.
58:06Um, it's it's a bookmarkable one.
58:08>> Yeah, that's going to And I honestly I'm
58:10I still can't believe that we got to do
58:12the podcast last week with Colin and
58:14Charles and you. Like, what an epic
58:17lineup. What an epic team for that
58:19episode.
58:19>> So many drops there. So many drops. I
58:21learned more listening to those guys
58:23than a hundred posts on on LinkedIn.
58:26>> What stood out most? Definitely the way
58:28that they're looking at the math that
58:31drives getting into comic call, how how
58:35much the money economy in the terms of
58:37values of domain names. These are things
58:38about SEO that haven't changed in 20
58:41years, right? Um and these guys are
58:44turning into an industry the same way
58:46property developers are turning property
58:48into money, right? The same thing that's
58:49happening in Florida. what used to be um
58:54the Everglades, swampland being turned
58:56into cities, right? The same things
58:59happening. The valuations on some of
59:01these domains um how they're working out
59:04how money can be made. Um way ahead of
59:08of technical experience. Um but there
59:11are people just making money from doing
59:12SEO. Kind of like people make money in
59:15the stock markets just short selling and
59:17things like that. That's exactly what
59:19they're doing.
59:20>> I want to share one more thing actually
59:22that came to mind. Um, which was you
59:23were saying that LLMs don't like using
59:26information from the from the same site
59:29too much. They they want variance. You
59:31know what is very similar to that that I
59:33found is Google images is is you could
59:36have like a lot of images. You could
59:39have like a lot of perfectly optimized
59:41images for one keyword on your site and
59:45Google is still going to look at other
59:48places to show them from multiple
59:50different sources. So like um I I found
59:53this out because I created a photos page
59:55for um Edward Sterm. It's
59:57edwards.com/phos
59:59because like there's all these other
1:00:00Edward Sterms in media and they had
1:00:02their photos in Google images and I'm
1:00:04like I want more photos of myself here
1:00:06and and I'm like and I can shape the
1:00:08narrative around myself by choosing what
1:00:10photos of me get put up and now the the
1:00:14it's it's like I would say it's working
1:00:15to a decent degree. I think like my
1:00:17photos page was pop is populating Google
1:00:19images like 30 to 40% of the time which
1:00:23is still really great but it should be a
1:00:26lot it it still should be a lot more
1:00:27than that and you see that Google images
1:00:30is picking YouTube thumbnails from
1:00:33various channels it's picking Instagram
1:00:35thumbnails it's taking it it's it like
1:00:38from various channels so like if you
1:00:40want to do better on Google images go on
1:00:43podcast be on different be in different
1:00:46sources And and I've just found that
1:00:48Yeah. Like it wants variance.
1:00:50>> It does. Absolutely. It needs it.
1:00:53>> Yeah. Hey, this was fun. This was a good
1:00:56one. We We made We made uh
1:00:58>> made good a good use of that failed geo
1:01:01debate. All right. Cool. David David
1:01:03will be back everybody tomorrow with
1:01:06Jackie Chu. This is episode 1,183
1:01:09of the podcast. 1,183
1:01:13days in a row. No days missed. And I am
1:01:16so grateful for amazing guests like
1:01:18David G. Quaid of Primary Position for
1:01:21coming on the show. If you watch this,
1:01:25thank you David. If you watch this on
1:01:27YouTube, thank you so much for watching.
1:01:28If you listened on Spotify or Apple
1:01:30Podcasts, thank you so much for
1:01:32listening and I will talk to you again
1:01:35tomorrow. I know.

About this transcript

This page contains the full transcript of ChatGPT Is Building Its Own Search Index. Here's How to Rank in It by Edward Sturm, generated from the public captions YouTube serves with the video. The transcript has 11,142 words across 1,604 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.