YT YouTube2Text

YouTube2Text → Transcripts → He Dug Through 9,249 SEO Case Studies. Here’s What Actually Works

He Dug Through 9,249 SEO Case Studies. Here’s What Actually Works — Transcript

by Edward Sturm · 12,355 words · 2,040 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00We are talking about SEO tests and
0:02interesting results on this episode of
0:05the pod and I'm joined by Krešimir
0:07Trkulja who put out a while ago, but I
0:11made an I made an episode on this a
0:12couple of days ago. So, Krešimir, you
0:14spent the summer going through more than
0:1730,000 SEO documents and narrowing them
0:21down to 9,249
0:24case studies, tests, and papers with
0:27measurable results. And he it Yeah, I
0:30could talk about that. That's crazy. So
0:32much SEO.
0:34>> Yeah. Yeah, it it it was a lot. My first
0:36idea was
0:38uh
0:39I want to solve SEO like finally just
0:41have tactics which actually work because
0:44everyone has uh
0:46their own opinion on things like should
0:49you index pagination or not, should you
0:51put self-referential canonicals or not,
0:54how we deal with local SEO, backlinks,
0:56and so on and so on and so forth. Um
1:00So, the idea was to just gather
1:02everything that was basically ever
1:05published, but it's mostly the last 2
1:08years because I didn't want all of those
1:10really really old case studies
1:13um because they don't work anymore. Some
1:15of them do. Some of them still work, but
1:17it's not that relevant anymore.
1:20Um then I used Ahrefs and Screaming Frog
1:23and some other tools to find
1:26basically any mention of an SEO case
1:29study uh out there on the internet. And
1:33it was like 30,000 documents or so.
1:37Um after I found them, this is just like
1:40a really short overview of how I did it.
1:43Um when I found them, I created a
1:46pipeline uh where
1:49uh
1:49I first had a filter which removed all
1:51of those uh like industry benchmarks
1:54like AIO reviews are eating up like 30%
1:58of your uh click-through rate or clicks
2:00or whatever.
2:02I wasn't interested in that. I was only
2:03interested in tactics.
2:05Uh, then I used two different AI models,
2:09one from OpenAI, one one from Google,
2:12uh,
2:14check
2:15every document two times to see if they
2:18can figure out outcome and a specific
2:21tactic.
2:22And if possible, even things like the
2:25number of pages the test was done on.
2:28Was there a control group or not?
2:31In what market? So, loads of loads of
2:34different things. And every model passed
2:37through it two times and
2:40they had to agree uh,
2:43it it was like 0.85
2:45uh,
2:47agreement rate. So, they had to agree at
2:49least
2:50at that rate, but most of it was at like
2:530.92 or something. Um,
2:57they had to be grounded completely. So,
2:59every number and every tactic and
3:00outcome had to actually be there.
3:03If they couldn't agree, then it
3:05escalated to a way better model.
3:08Um,
3:09and yeah, then I basically clustered
3:12every tactic there is in into specific
3:16like
3:16is it local SEO? Is it backlinks? Is it
3:21uh, AIO,
3:23geo,
3:24or whatever. And also, was it
3:27click-through rate manipulation? Was it
3:29like anchored text optimization or some
3:33some like really really specific tactic.
3:36Uh, and the best ones are in my book.
3:41The SEO meta-study. It's the biggest
3:43piece of
3:44uh, SEO research ever done.
3:46Uh,
3:47I couldn't do it if people didn't
3:50publish their
3:51case studies on on the internet. It's
3:53completely free. I I can't even if I
3:56wanted to I don't I don't want to I
3:58don't
3:58>> can you there's no there's not even an
3:59email sign up to get it. You like um I I
4:02shared it in the last episode where I
4:03shared your research. It's you literally
4:05just go to uh the research page and it's
4:09a PDF and it's it's not like an actual
4:11book. It's a PDF and you click it. You
4:13don't have to give an email.
4:15You literally just click it and download
4:17it right away.
4:18And so credit to you for for that. This
4:20method of marketing is so effective I
4:22had to make sure it wasn't against
4:23Google's rules before I kept doing it.
4:26It's a form of SEO I call compact
4:28keywords. Whereas most SEO focuses on
4:31putting up articles to answer questions
4:33how what when, compact keywords focuses
4:35on putting up dozens of pages that sell
4:37to searchers who are actually looking to
4:40buy. These pages rank on Google and
4:43convert so much better than normal that
4:45when I discovered this years ago I
4:47couldn't believe this was allowed. It's
4:49less work too. The average compact
4:51keywords page is only 415 words. Compact
4:55keywords is a 13-hour deep course on
4:58getting sales with SEO. A customer
5:00recently said, "Each lesson is dense
5:03with information. You are giving years
5:05worth of experience boiled down into
5:0715-to-30-minute lessons with no filler
5:10or fluff. I feel like I'm gaining a new
5:12superpower." Compact keywords is about
5:14setting up an SEO funnel that brings you
5:16sales for years and years and years. It
5:20works with AI, it's less work than
5:22traditional SEO, and it makes way more
5:25money. You can get it now at
5:27compactkeywords.com.
5:29Back to the podcast.
5:30>> Yeah, thanks. I I thought about it
5:32actually. I thought about having like
5:33some email sign up or something like
5:36that, but it seemed just too
5:39weird to I'm the pipe the pipeline is
5:42mine. Uh the money I invested I invested
5:45about two or three K
5:48uh dollars into the whole project. It
5:52it took me the whole summer.
5:54I didn't work on it every day for 8
5:56hours. I was also on my annual leave, so
5:59I worked on it some and then went off
6:02for like two or three days and so on,
6:03but still a lot of work,
6:06a lot of manual checking of everything,
6:09a lot of money.
6:10And I thought about doing it, but it
6:12wouldn't be fair to all of the people
6:13who put up their case studies for free.
6:16Uh even though there could be an
6:18argument for it, but I think these
6:20things don't only help the industry,
6:23they don't only help people like me,
6:26they also help people who are in um
6:30who are our clients basically understand
6:32that SEO and all we do is not
6:36just like random tactics which don't
6:39work, which don't really have results.
6:41They do have results, the outcomes are
6:43actually pretty good, pretty amazing. I
6:46think the the you could say that the
6:48median uh outcome of SEO tactics
6:52generally is like plus 200%.
6:56So, that's
6:57pretty good. That's a pretty good return
6:59on on investment depending on how much
7:01you pay your SEOs. But yeah, that was
7:04also one of the points. Um so, people
7:07can understand that SEO is not like some
7:10occult thing
7:12uh really scammy people do who message
7:14you on LinkedIn. It's actually like a
7:16real discipline which has real results.
7:19I can't believe I'm saying this in 2026,
7:23but people people get burned on on bad
7:28agencies and freelancers and so on, and
7:30then they
7:32keep
7:32just get get to thinking that SEO is a
7:35scam uh generally.
7:38>> Yeah. So,
7:38>> Yeah, yeah.
7:39>> What what tactics stood out to you the
7:41most? So, wait. So, you got and you got
7:43this um you got So, these 9,249 case
7:47studies, you got this down, the final
7:49data set. I think was like 552 cited
7:52sources and 182 SEO tactics.
7:56>> Yeah, those are the best ones. So, the
7:58ones who had which had the most
8:01controlled tests, uh the best
8:03evidence-based
8:05uh because I ranked every
8:07uh every tactic on how good the evidence
8:11is and how actionable it is. So, the
8:14evidence is uh does it have um
8:17like the number of pages it was done on?
8:20Uh does it have a control test or not?
8:24Uh
8:25and things like that. And action
8:27actionability is actually just uh how
8:30easy it is to do it and how big the
8:34outcome is. So, these two uh are
8:38how I calculated the best uh out of the
8:42uh 1,000 plus tactics. I still have a
8:44lot of them on the back burner uh that I
8:47didn't share just because they don't
8:49have uh the great
8:51uh tests behind them. Uh I
8:54I kind of use them myself for different
8:57things.
8:58Uh I wouldn't share them because some of
9:01them uh I wouldn't share all of them
9:03because some of them are obviously not
9:05something you can do without an SEO and
9:08I'm also kind of afraid that people are
9:11just going to plug it in
9:12to their uh website and see what happens
9:16and they can get burned pretty badly
9:18because a a lot of these tactics are
9:20really niche uh
9:23experimental stuff. Some of them uh
9:27well, actually most of them are uh
9:29pretty general stuff. Uh
9:32we which can also be done, but
9:34that that don't have uh that much
9:37evidence behind them.
9:38Um so I kind of you
9:41I kind of
9:43put out the list. I I sent you the list
9:46of the best tactics from the
9:49published base and also the unpublished
9:53base. So this will be a first for
9:55sharing anything from the unpublished
9:57base
9:59anywhere basically.
10:01>> stood out to you?
10:02>> [clears throat]
10:03>> Well, first of all,
10:06um
10:06a lot of these AIO AEO or GEO tactics
10:12are kind of
10:14weird. Some of them
10:16we've been doing for a long time as
10:19SEOs. So things like adding
10:22quotations
10:24for
10:25in our content from experts in the
10:27field. That's something which was always
10:30a best practice, but the
10:34let's say GEO or industry is just now
10:36finding it in a way. But I'd say that
10:40that works great especially in digital
10:44PR. This is what we use with our
10:46clients. So having like expert response
10:49or expert reviews,
10:52it works way better for SEO and GEO.
10:56But also just putting out general
10:58content with
11:00quotations for AIO. Also what works is
11:05adding statistics, tables. So it's
11:08nothing
11:10really crazy.
11:12But there's some stuff in there that
11:15maybe people aren't doing. Although I'm
11:18kind of iffy on saying that these AEO
11:24tactics really work because I'm not sure
11:26how people measure brand citation.
11:29>> what is work What does a mean?
11:31>> Work as in like actually increase
11:36the number of your citations or your
11:39traffic or revenue. Uh but I'm kind of
11:44like I said iffy about it because I'm
11:45not sure how people track these. Uh
11:49people I guess mostly just ask at the
11:52most basic level they just ask like chat
11:54GPT Claude and and Perplexity or whoever
11:58um name me the best brands to buy shoes
12:02from.
12:03But that doesn't really work because
12:05it's incredibly personalized.
12:08Using the API is also completely
12:11different from uh
12:13pulling out the answers directly from
12:16the chats.
12:17And there's also the issue with
12:21basically tracking and attribute
12:23attributing anything from AI's. The only
12:26thing I can see I see you can do is look
12:29at server logs and check if the is
12:32getting
12:33fetched more often or if it's getting
12:35fetched
12:37when it wasn't before or just checking
12:39if you have uh referral traffic from AIs
12:43generally. But I don't know what having
12:47more brand citations means. Uh
12:51when I put up my Reddit thread on
12:54r/techseo
12:57one person
12:58uh said that uh
13:00a lot of these case studies don't really
13:02track over time as well. So they ask AI
13:06at uh
13:07let's say Wednesday and then they say
13:10okay we're
13:11among the five companies presented for
13:14best running shoes, but no one really
13:17checks the day after or a week after and
13:20then tries to see if there's like an
13:23average you can find over a a or over a
13:26month. Everyone just basically uses one
13:29input.
13:30Uh better companies probably use way
13:32more, but it's all really weird to me.
13:35So, I'm really
13:37uh careful about all of these new
13:40uh AEO tactics uh
13:43unless they're basic stuff like we've
13:45all been doing this whole time like
13:48adding statistic statistics quotations
13:52uh representing
13:54uh
13:54the content in in numbers not just
13:58general statements. So, it so it's not
14:01like uh sales fell after we it's sales
14:04fell by 37% because we did this and
14:08that. That's just generally a best
14:10practice. It it has no
14:13uh connection to AEO, let's say. And
14:16LLM's text is uh some report like
14:21it an increase in in citations uh some
14:23report exactly nothing and I would
14:28uh I would stand on the side of nothing.
14:30I don't really believe it makes sense.
14:33>> Uh
14:34LLM's.txt
14:36basically did nothing from the tests
14:38that people ran.
14:40>> Yeah. Yeah. There are some who said uh
14:42that it they did do some things, but the
14:45tests were bundled and not just bundled
14:48uh they were just basic before and after
14:50tests.
14:51>> Explain bundled. Bundled means so that
14:53they did like a lot of things at once
14:55and they can't measure that it came from
14:57right?
14:58>> Yeah.
14:58>> That that it came from LLM's.txt. It
14:59could have come from any one of the many
15:01things that they did. Yeah.
15:02>> Yeah. Most of the tactics are actually
15:05bundled because when people publish case
15:07studies it's uh we believe
15:09>> very few good ones, but there are some
15:12There were a lot of interesting things
15:13that that you had shared in like the
15:15findings.
15:17Um I could share some of the interesting
15:19some things that I thought were
15:20interesting, but what what like to you,
15:21what stood out as most interesting?
15:24>> Uh
15:25well,
15:27the most interesting thing I guess would
15:29be
15:30uh using less less varied text on anchor
15:35text on internal links.
15:37So, the general best practice was always
15:39that you got to like
15:41use like synonyms or different
15:43spellings, not like completely different
15:46spellings like bad grammar, but just
15:48different spellings of uh
15:50uh of the page you're trying to link to,
15:53but
15:54uh the tests actually uh said that using
15:57less varied text, so even just exact
16:00match every time uh worked way better
16:04than using uh varied anchor texts for
16:07internal links. Backlinks are a bit
16:10different thing. Uh that was also
16:13something
16:14that was interesting to me. There was
16:15one
16:16uh
16:17a pretty good study that showed that uh
16:21basically disavowing, although also the
16:24disavow is something which was
16:26interesting to me, but I'm going to get
16:28through that later.
16:29Uh disavowing links where the anchor
16:32text was uh exact match of the keyword
16:36they were targeting, so they disavowed
16:38them and they got uh I don't know, like
16:41a 300% increase in
16:44uh in organic clicks.
16:46>> Wait, wait, wait. So, so exact you're
16:48saying exact match anchor text backlinks
16:51were disavowed?
16:53>> Yeah.
16:53>> And that got a 300% increase
16:57in clicks?
16:58>> Yeah.
16:59Although that could be because uh
17:03Google was probably already either
17:05ignoring them or they could have figured
17:08out that uh
17:10they were actually losing traffic on
17:12those pages. They lost ranks. Maybe they
17:15just bought like loads of backlinks at
17:19mhm enough so Google found them really
17:21suspicious and then just dropped that
17:24page down, but I don't know if it was
17:26like 300% from baseline or if they fell
17:30and then trying tried to figure out the
17:33issue and then disavowed them and then
17:35in it went back up, maybe even better
17:37than before. Not sure if it was the
17:40first or the later one, but still pretty
17:44interesting.
17:46Um
17:48There are also like loads of
17:51different things like using
17:54plain subject-verb-object sentence
17:56sentences. HubSpot did this. This also
18:00works for
18:01>> Example.
18:02>> Uh so
18:04HubSpot used it for
18:07HubSpot is can automate email marketing.
18:11So just a basic sentence which tells you
18:14everything they do right away.
18:17It helps with branding a lot. I found
18:19that also using
18:21using it in digital PR. So if you're
18:25If you have like data sent to us by I
18:29don't know company X
18:33who let's say automate email marketing,
18:36whatever. That works great because then
18:39that sentence is connected to all of the
18:42other entities on every other page
18:44saying that that specific brand or
18:46company is doing that specific thing and
18:49it's really simple for algorithms and
18:52people also to understand and I would
18:55say that
18:56being careful about how branding works
19:00across different channels and that
19:02basically every channel should have like
19:06one sentence
19:08which works like that. So my company X
19:11does Y. Really basic, really simple,
19:15really memorable. So, you just share
19:17your unique selling proposition.
19:20And that's it.
19:21What one connected finding to that was
19:24that there was one company who just went
19:27over every channel and found that
19:30different channels had slightly
19:32different
19:34slightly different tone of voice or
19:37slightly different
19:39like brand name. It's just basic
19:42hygiene. Uh
19:44from the SEO side, but I think a lot of
19:46people disregard this completely and
19:48think there's way more interesting stuff
19:50to do. But even But now that AIs are
19:54becoming more and more important
19:57although they're still like 0.3%
20:00of all internet traffic coming to
20:02websites, but they're becoming
20:04increasingly important.
20:06You have to have like a transparent
20:08brand everywhere. And everything
20:10everywhere has to be the same including
20:14digital PR
20:16basic link building everywhere
20:18everything. So, everything has to be on
20:19the completely the same page.
20:24>> The You shared
20:26I'm going to read some of the things
20:27that that I thought
20:29>> Sure.
20:29>> Like you These were some of the findings
20:31that were that were interesting. Okay,
20:32so writing title tags entirely Yeah,
20:35let's talk about this one.
20:37Hopefully you remember it. Writing title
20:39tags entirely in capital letters
20:42increased organic traffic 17.5%
20:45across two independent tests.
20:48>> Yeah, that was SearchPilot. Most of
20:50those tests
20:52regarding
20:54the titles are actually from
20:56SearchPilot. They were the MVPs of this
20:59whole study in
21:01in a way because they're the
21:04not the only ones, but the most of their
21:08tests are controlled.
21:10Uh no one else does it basically.
21:12There's I think one
21:14uh like in in the top 10,
21:18uh nine are
21:20uh SearchPilot and one is Ahrefs
21:23uh a controlled study. So,
21:26so
21:27SearchPilot is basically most of the
21:29controlled tests, especially regarding
21:31titles, because they're the easiest to
21:33control uh amongst other things. So,
21:36yeah, uh a lot of these things uh
21:38regarding titles were very very
21:41interesting to me.
21:42Um there was also like adding dynamic
21:46price versus adding a static price,
21:48adding uh
21:50removing the year and putting in just
21:53updated.
21:54>> You're say You're saying these things,
21:55but we got to know the results. We got
21:57to know You're saying adding this but
21:58versus adding that, but what's what's
22:00better?
22:01>> Uh so, for capital letters, it was uh an
22:04increase. I think it was like 10 15%.
22:07So, that's pretty good for just changing
22:11up your title.
22:13For uh adding prices, it was I think
22:17minus 7 or 10% for having static prices
22:22and plus 15 also I think for adding
22:26dynamic prices in in the title tags.
22:29Although, I'm kind of
22:31uh sus
22:33also kind of iffy uh about adding
22:37uh prices generally into uh titles. I
22:40mean, the the test found the completely
22:42other
22:44outcome, but I would think that people
22:46would would be less inclined
22:49uh to click if they could al- already
22:52filter out the results by price. And
22:55then
22:56uh if they clicked and there was no
22:58price there, maybe your website could
23:00make them buy it even though maybe it's
23:04over their price range. But yeah, so
23:08that's just basically about
23:10title freshness. So the titles have to
23:12be fresh. Also adding just updated is
23:16better than adding a year into the
23:18title. Probably also because
23:21>> Really? That one's That's interesting.
23:23So yeah, instead of writing, you know,
23:26um
23:26>> the month and year, yeah.
23:28>> Yeah, yeah. Instead of Instead of
23:30Instead of adding like the best email
23:32marketing platforms 2026, you say best
23:35email marketing platforms just updated.
23:37That's so And that's a lot easier to You
23:39don't have to change it then. That's so
23:41funny.
23:42>> Yeah. Although they I don't know if they
23:44tested
23:46having like
23:48a set of pages which are getting updated
23:51regularly.
23:52Maybe it was the issue was that a lot of
23:55those pages had a
23:58had like old titles 2025, 2023, or
24:03a wrong month or something. But it could
24:06just be that just updated means way more
24:10to people than
24:11>> And you stand out in the SERPs more. You
24:13stand out in the search engine results
24:14pages more. And the reality is most
24:16people don't search the year. They just
24:18search best email marketing platforms.
24:20And so And so yeah, you you Instead of
24:23like Instead of being the sea of like
24:252026 2026 2026, one says just updated,
24:29which could mean that it was updated a
24:31month ago or 2 weeks ago.
24:33That's so funny. That's brilliant.
24:37>> the the user thinks maybe it was updated
24:40like
24:41yesterday.
24:42>> Yesterday, yeah. Yeah.
24:43>> Yeah.
24:45>> That's a smart one.
24:46>> Yeah, I actually wanted to share
24:49um
24:49So you're going to have to read through
24:52this one. It was
24:53>> can talk You can talk about it. Read
24:55read it.
24:56>> Yeah, it's really really interesting.
24:58Like it's something people don't really
25:00do or think about.
25:02I thought about doing it, but I thought
25:05that
25:06it can't work. It's completely crazy.
25:09It's really specific thing.
25:11So,
25:12they basically used EXIF geo data and
25:15image object object schema.
25:18So, they added geolocation EXIF data and
25:22entity tags to destination photos. Then
25:24implement implemented object schema
25:27linking each image to the entity depicts
25:30plus 90% on show me photos of landmark
25:34Google Lens queries. So, that's
25:36something for Google images. And I think
25:39it could actually also work if you're
25:41trying to
25:43prove that you're local somewhere. So,
25:46the EXIF data should probably say that
25:50the image of your store or something was
25:52actually taken
25:54at the place. So, it's not
25:56>> test.
25:58>> Sorry?
25:59>> Who did this test?
26:01>> It
26:03was one site.
26:06Data enrich for a travel site in 2026.
26:10>> Interesting.
26:11>> Yeah.
26:12Really really
26:13>> results were that adding EXIF data
26:15actually helped.
26:17>> Yeah. So, the EXIF data and adding
26:20image object schema
26:22>> this is what this is what confuses a lot
26:23of people because Jake
26:25>> Kundli uh who is a friend of the podcast
26:28and a great SEO,
26:30he did a bunch of he did the most um
26:33detailed test on local business schema
26:37and found that local business schema
26:38pretty much did nothing. But he also did
26:41an extremely extensive recent test on
26:44adding EXIF data
26:46and found that it did absolutely
26:48nothing. Yeah, and it was published in
26:50Search Engine Journal.
26:51And this is why I think this is like why
26:53a lot of people in SEO get confused
26:56because then you hear about these these
26:58tests and
27:00a lot of a lot of the tests require you
27:02to think critically about SEO, to
27:04understand the fundamentals of SEO, and
27:07to think critically about it. So like
27:08when you hear like, "Yeah,
27:10replace 2020
27:12replace like the date with adjusted
27:15updated, if you think critically that
27:17makes a lot of sense because we know
27:19that click through rate in the SERPs
27:20matter. Like we we everyone knows that
27:23that's not like a
27:25everyone knows that that that matters.
27:26Google has said now for years that that
27:29matters.
27:30And just if you think about the way that
27:33most people search, they're not putting
27:34the date, they're just putting the thing
27:36and then expecting to get the most
27:38recent freshest data. And so when you
27:41see just updated, like if you're
27:44thinking critically, that tactic makes a
27:46ton of sense.
27:48Wha- wha- if you think critically about
27:49XIF data, you might be like, "Yeah, I
27:51could see how that would make sense."
27:52But the reality is like not all images
27:55are going to have XIF data. Is that
27:57going to make Is XIF data going to make
28:00one image better than another? Not
28:01necessarily.
28:04>> Yeah, there's lo- loads of things where
28:06you have to think critically. There is
28:09also one case study which said
28:11uh
28:12that that was actually one of my
28:14mistakes I did for one one client. Uh it
28:17was The case study was minus 5% for uh
28:22foo- footer links. So basically adding
28:24like category links in the footer.
28:27Uh that was always like my personal best
28:30practice. So if uh a pa- not for like
28:33billions of uh categories or if a if a
28:36client has like uh 100,000 categories or
28:40or like index to faceted uh
28:43filters and so on. But if it has like a
28:47good number
28:48uh which doesn't fit in the navigation
28:51adding it to the footer having like a
28:54hidden part of the footer for the user
28:57where Google and
28:59every bot could find those links
29:02was something that always helped all of
29:05my clients. It was It was like a really
29:07basic thing which increased because you
29:09have internal links pointing from every
29:12page on the website. So, these
29:14categories get get way more internal
29:16links and it's all great and fine and
29:18dandy, but I had an issue with one
29:21client
29:23who was in the health niche. And I put
29:27it for their lab tests. So, they had
29:29loads of lab tests
29:32which were just like internal links with
29:34the anchor text testosterone or diazepam
29:37because it was a test for diazepam
29:40or if someone was using some sort of
29:43medication and so on. And that was an
29:46issue for
29:48the PPC team
29:50because
29:52they kept getting
29:55blown
29:55blown back from Google for those
29:59specific keywords. And the issue was
30:02that it looked like they were selling
30:03testosterone and diazepam on on every
30:06page. It looked really really scammy. I
30:09didn't think it through completely when
30:12I did it.
30:13I changed it since then, but that's just
30:17something you can get out of this is
30:20that
30:21you really need like an SEO who knows
30:23who
30:24what he's doing and not just knows what
30:26he's doing, but that he actually thinks
30:28through everything he does because even
30:31people who've been
30:33in the industry for years, for decades
30:36even, can make mistakes. And it's
30:40So, I wouldn't really just
30:42use like a plug and play
30:44Claude
30:46for websites unless it's just like a
30:48side project, which I also do. That was
30:50one of
30:51also one of the points for my study. I
30:55was thinking about creating like a hub
30:57of every possible tactic and then just
31:00publishing every tactic on
31:04a website and seeing what happens or
31:06just a part just like maybe internal
31:08links and then like navigation stuff,
31:12page layout stuff, content stuff and so
31:14on just to see if there's like some
31:16magic way to get like a really stupid
31:21affiliate site to make money.
31:24So, that that was also one of the
31:25points, but it doesn't really work that
31:27way.
31:29>> I think prominent brand call to action
31:32blocks to product pages increase
31:34conversions 80% across two tests.
31:38>> Yeah, like you said
31:40in the YouTube video you did. It was
31:43probably because it was way more
31:45visible.
31:46A lot a lot of these tests are
31:49we made something from the page layout
31:52way more visible and it helped our SEO
31:55and UX and revenue. It's also I think
31:58one one test was adding
32:01like three or four bullet points about
32:04the product uh
32:06as as far as you can get above the fold.
32:10So, just below the heading so the user
32:14sees right away, okay, this is for X.
32:17You can use it I don't know, shoes. It's
32:19for hiking, running. We have these
32:21colors and so on.
32:24Maybe if you're a big brand like Zara or
32:26something, you can avoid having product
32:28descriptions
32:30and having content which really helps
32:32the user and AI. You can just have
32:35jacket, but if you're not that big of a
32:38brand, you really have to go through
32:41with having a lot of content. That was
32:43also one of the studies.
32:45So,
32:46creating like AI content on product
32:50pages increased
32:53increased clicks, I think, by or revenue
32:56or conversions, not sure, by 85%.
33:00So, it's not an issue that it's AI
33:02content. It helped
33:05the AIs and algorithms and also the user
33:09to find what the product is about, what
33:12it works for, what it works with.
33:16But, it probably wasn't just like random
33:19words which which are just filler. It
33:21was probably like finding, "Okay, this
33:24product is for X and Y and so on." A a
33:28lot of companies, especially ones who
33:30have loads of partnerships with loads of
33:33it's like fast-moving consumer goods and
33:35stuff, they don't really get good
33:38product descriptions from the
33:42from the supplier. Sometimes they do,
33:44sometimes they they don't. But, maybe it
33:46could help people if
33:49if you have like really bad product
33:50descriptions to just like use AI to find
33:54all of the info it can about that
33:57specific product and feeding it into the
33:59page. It helps everyone. I can't think
34:02why it would be
34:03seen as spammy or something. That that's
34:05a really great tactic and shows that AI
34:09can really help us a lot and in writing
34:12content as well as long as it's not like
34:15really stupid blogs which are commodity
34:18content and don't really do anything.
34:21>> It also I mean, it also depends on like
34:23what the AI content is and is it
34:25reviewed? Like is it is it it it
34:28Obviously, AI can write stuff that's
34:30great. It depends on of power the person
34:33is using it. Is the person like one
34:34shotting it with a little prompting or
34:36is the person giving heavy details about
34:39the product and then the AI is you is
34:41just turning that into something that's
34:42concise and nice and easy to read.
34:45Like those are very those are very
34:47critical things. Or is it literally just
34:50like AI slop? Yeah, this is what a shirt
34:52is. Like
34:53>> Yeah, I mean my AI study is completely
34:57AI content. I mean I checked it. I
35:01worked through the pipeline like a
35:03billion times because I had a billion
35:05bucks.
35:06>> is yours is based on interest. Yours
35:08isn't based on actual novel
35:12proprietary data.
35:14And and so it's like yeah, it's like
35:16you're if you're you have like an e-com
35:18store
35:19and you are giving
35:21everything that there is to know about
35:22the product to chat GPT and you're like
35:26structure it this way.
35:27Obviously, that's going to be better
35:30than nothing. Especially if it's
35:32reviewed afterwards.
35:34Yeah, versus like here's a screenshot of
35:36my shirt, just write some stuff.
35:39Yeah, which still still might be okay to
35:41be honest.
35:42Which might actually be it might
35:44actually be okay. I mean it's not going
35:46to be great. And then it also depends
35:47are people doing this in bulk
35:50or are they doing it and but yeah, but
35:53also if it's like if it's like an e-com,
35:55a lot of people are just going to decide
35:57with their eyes, too. They're going to
35:59look at the product and if they like the
36:00product, that's
36:02that yeah.
36:03>> Yeah, that's really important. Like
36:05having good
36:06I don't think I've found any like case
36:09studies for that specifically, but just
36:13like a general rule of thumb for me is
36:15to have products which have completely
36:17different images from the competition.
36:21So if if everyone just has
36:24like basic jacket on white background,
36:27maybe you can have like a person wear it
36:30or have a person wear it outside or
36:32something. For one client, this was
36:35before AI
36:37or the public AI we now have.
36:41For one client, they were a supplement
36:43store and then we basically just put the
36:47ingredients. If it had ginger in it or
36:50citrus or
36:52curcuma or whatever, we just put it not
36:55like photoshopped it, but actually put
36:57it next to the product. And it the
36:59product images looked way better and it
37:02made people want to click on it way
37:04more.
37:05I don't have any proof behind that
37:07claim, but I think that we did increase
37:11their traffic by like 200%. So, it
37:14probably had an effect, but it was
37:17bundled. So, I don't know which thing
37:19actually had an effect. But regarding AI
37:22content, like my study,
37:25even though it's completely checked by
37:27me, it's human verified and so on. And I
37:30still have to verify it all over again
37:33all the time because there's still some
37:35errors. Maybe for addition two, I'm
37:37going to look at some different things.
37:40But uh
37:41uh
37:42I don't know why more companies don't
37:46use their own data with for content.
37:51They still
37:52base their content work mostly on
37:55commodity stuff or on like
37:59people also ask questions. Uh
38:02Which is fine. That's great. I think
38:05especially tools like also asked are
38:08great. And using them actually
38:11also helps a lot. So, answering
38:14question questions around your product
38:17or service, the more granular you get,
38:20the better it is. It doesn't have to be
38:22like one long piece of content, but just
38:24generally
38:26answering those questions all
38:28all around the website. So, maybe one
38:31answer can be in a table, one answer can
38:34be in like some video, one thing can be
38:36in wherever in the navigation. It
38:39doesn't matter. That also works, but I
38:41think the best thing which works the
38:45best, especially if you're doing link
38:47building, digital PR, this works for us,
38:49is just using
38:52basically the data the client already
38:54has and
38:56analyzing it and putting out a novel
38:59piece of research which always gets
39:03shared
39:04a lot everywhere.
39:06I mean, it depends on the client. Some
39:08Sometimes the client is in a really like
39:10boring niche where you can't really pull
39:12out any interesting data.
39:15So, you have to be a bit more more
39:17creative, but there is some industries
39:19where I'm really surprised that people
39:22don't put out content all the time. I
39:24mean, Ahrefs does this wonderfully. They
39:27have loads of proprietary data and they
39:30keep doing They're up front about it.
39:32It's also like AI content, the the thing
39:35they do, but it's
39:37really
39:38well thought out, really well done AI
39:41content which I don't mind. I don't care
39:44that it's done by or analyzed or
39:46whatever by an AI. It's
39:48actually a great like great piece of
39:50content with great data. So, I don't
39:52know why more companies don't do this.
39:55>> It can be dangerous. Like we are
39:57So, for this
39:59SAS that I'm funding, this five-coded
40:01SAS that I'm funding, there are
40:02companies in our space that are sharing
40:04all the mistakes and learnings that they
40:06have. And it's like saving us so much
40:09time because we're literally like able
40:11to read their change logs, see the most
40:14requested features, uh see what people
40:17care about. See what's actually popular
40:20in their apps versus what's not popular.
40:23And like
40:25we are just we're just lurking on this
40:28data and we're like, "Oh great. This is
40:30This is saving us so much time."
40:33There are risks involved, but then the
40:35rewards, you get a lot of backlinks and
40:38it makes it way easier to do SEO. And
40:41especially if your competitors don't
40:42know how to do SEO, like you can just
40:45>> dominate the
40:47You can dominate all the SEO in your
40:49niche.
40:50Because you're getting all this
40:51authority. And the the truth is like
40:54even though you're making yourself
40:55vulnerable like that,
40:57you still end up winning. You get more
40:59top of mind awareness because these
41:01studies are going they're getting shared
41:03in your niche. And then you're getting
41:05all this authority that makes it easier
41:07to do high intent SEO. And or just SEO
41:11in general.
41:12>> And so like the trade-off is worth it.
41:15>> Yeah, you also get loads of brand
41:17mentions.
41:19If you really put out a novel piece of
41:22research.
41:24For instance, for one one client we did
41:28the
41:29counties where there's
41:32the most traffic accidents. Although
41:36that piece of research is not really
41:38well done because just some counties are
41:40just way more they have way more traffic
41:44because people are going through them
41:46from point A to point B. So it's not
41:49like people from that county drive way
41:53worse than in some other one. It's just
41:55because of the amount of people passing
41:57through through. But
42:00we basically did that and now if you
42:02search or look at ChatGPT or whatever,
42:06which
42:08which county has the worst drivers,
42:11maybe our our study comes up. There's
42:14loads of things like this you can do.
42:16And especially if it's around your This
42:18was for car insurance. Especially if
42:20it's around your industry, it can be
42:22really important important to be
42:25mentioned in different
42:28in different ways. Not just for bottom
42:31of the funnel, your specific keyword you
42:34want to rank for.
42:35That and
42:37uh also some a thing a lot of people
42:41don't do is uh linkable assets. There
42:44was also a case study, a great one on
42:48something really basic. It's It's from
42:50this year. Uh so creating a
42:53calculator
42:54uh
42:55for I think real estate prices and it
42:58worked amazing. They had 100
43:01uh thousand plus
43:04uh users and click They combined it with
43:07like uh
43:08uh general PR campaign, but they also
43:11got loads of organic backlinks, loads of
43:15mentions on Reddit or wherever just
43:18because it was like a really interesting
43:21uh calculator. And that That's like
43:23>> What made it What made it interesting?
43:25What made it like more than just a basic
43:27calculator?
43:28>> Uh actually
43:30not sure. It was SEO Design Chicago
43:35for an architecture firm built an
43:37interactive sustainability calculator
43:39showing long-term cost savings and
43:42environmental impact per project. So
43:45it was
43:47basically some a new spin on like
43:50calculators and also uh Programmatic SEO
43:54dot tools built programmatic financial
43:56calculators plus location guide guides
43:59going from 100 pages to over 950. So it
44:03was like uh
44:04programmatic SEO combined with with
44:07those uh
44:09calculators.
44:10>> How long ago was this study?
44:13>> Uh like I said, it was 2026, yeah.
44:15>> Yeah, I'd love to see the site that the
44:17programmatic SEO is on cuz a lot of
44:19people mess up programmatic SEO and they
44:21get they get that mountain pattern.
44:25>> Yeah, there's loads of studies which are
44:27just like we used programmatic SEO and
44:30it worked wonderfully, but I think there
44:33is way more on the other side. People
44:35are not going to say I think only two or
44:38three percent of everything I have,
44:40which is
44:41a negative outcome, comes from Search
44:43Pilot for my whole study because they're
44:46the only ones who report on that. No
44:48one's going to say I built
44:49>> If you're an SEO agency, yeah, you're
44:51and you and you you experimented with
44:53programmatic SEO on a client site and it
44:55went horribly wrong, you're not going to
44:58write a blog post about that.
45:00>> Yeah, you're you're really not. Uh like
45:04I said,
45:05uh stuff like calculators and generally
45:09linkable assets, I think is something
45:11which is disregarded, especially because
45:13maybe people think
45:15uh okay,
45:16uh Google Google is eventually going to
45:19have all of those calculators and
45:22graphics and everything from inside the
45:25SERPs. That already happened like a
45:28while ago. I can't even remember
45:30>> No, I know what you're talking about.
45:31>> Yeah, yeah.
45:32>> They Google announced it at Google I/O.
45:34They're like, "Look how interactive the
45:36SERPs are now." And actually, a lot of
45:38those promises haven't even come to
45:40fruition. Like Google Google made during
45:43Google I/O It was earlier this year. It
45:45was like, "Yeah, we're going to do this
45:46crazy stuff in the SERPs and it's going
45:48to roll out in the next in the next few
45:50months." And that didn't happen.
45:53>> Yeah, we're just going to burn like
45:55billions of tokens for one search,
45:58basically. I mean, it it's not
46:01that much, but still it would be kind of
46:04weird that they would burn
46:06like a huge amount of tokens for
46:09>> They made it They made it look like you
46:11do a search for like some SaaS product
46:13and then literally Google's AI is going
46:16to vibe code it for you right there in
46:18the SERPs and software is done. Like
46:20that's literally that's what their the
46:22demo looked like.
46:24>> Yeah, probably not going to happen but
46:27uh
46:28even if it did happen I think that
46:30if you really have a novel idea
46:34and know where to look for information
46:38to to build it I think that AI still
46:41can't be that or at all creative about
46:46these things.
46:48Basically anyone can can just vibe code
46:50stuff and even AI itself can give you
46:53ideas but all of those ideas are
46:56just really stupid stuff that your
46:59customers and you don't care about so I
47:02think it would make more sense to kind
47:04of
47:05try to get the creative juices flowing
47:07and see if there's something specific
47:10like it doesn't have to be a calculator
47:13it can be any number of things which can
47:15be a linkable asset.
47:17I mean like my meta study is one example
47:20of linkable asset
47:23it can be like
47:24programmatic pages regarding something
47:27statistics
47:29there was also one study where um
47:33people put out an agency put out um
47:37statistics
47:39which were searched for often by
47:42journalists. So journalists look at
47:44statistics like um I don't know traffic
47:48accident incident rate in whatever
47:51Croatia a statistic and then you have a
47:54piece of content about that you you're
47:56the only one who has a piece of content
47:58about that, they're just generally going
48:01to link to you. I
48:03In in some time, Google may just like
48:06put you out a number, but they would
48:08have to
48:10crawl and synthesize
48:12a lot more content than they do now, I
48:15think, and that would be very very
48:18costly.
48:19Uh not sure they're going to do it,
48:22especially when tokens stop being
48:25uh
48:26uh subsidized by investors.
48:29We'll see.
48:32>> So, you would you did this
48:34this uh this study over the summer
48:37uh because you wanted to learn you
48:39wanted to get better at SEO.
48:41So, is there anything that you wanted
48:44you wanted questions answered?
48:46>> Yeah.
48:46>> Is there anything that changed the way
48:48that you do SEO or enforced it?
48:52>> Mostly just enforced.
48:54There were not many things which were
48:57that surprising to me uh
49:00except for some of the ones I've talked
49:02about.
49:04Um
49:05like uh exit data is interesting, not
49:08something I've thought about that much
49:10before.
49:12Uh variation of anchor text on internal
49:14links being like smaller is also
49:17something which is something I I test, I
49:20check.
49:21Um
49:23like doing more visual PR and linkable
49:26assets and caring more about uh content
49:29and answering questions from users,
49:32however detailed they may be, uh are
49:35something which is way more enforced
49:38now. Uh I see that it really really
49:40works, so it's not something that should
49:42be down on the priority line.
49:46But, a lot of these are not uh things
49:49I've
49:51I've been surprised with. um
49:53like the disavow tool one was something
49:57which was also
49:59just enforced because it showed that it
50:03only worked when you have a manual
50:04action. I always thought this
50:07LLM text doesn't really work. I've
50:10always thought this.
50:12So a lot of these things are just
50:15something which I already knew or just
50:19had a hunch on. I thought okay, this
50:21should work like this basically, but now
50:24I have
50:26the proof behind the pudding.
50:28Um
50:30So yeah, it's not it's not all that
50:33all that surprising. There were some
50:34things I'm I'm sad that didn't get
50:37tested all that much.
50:38No one puts out case studies about this
50:42about these things. So things like
50:45let's say pagination. There were a few
50:48case studies which
50:49canonicalized it to the first page. I
50:52was always of the opinion that you
50:54should just put index follow everywhere
50:57because if a product which has search
51:00volume or people look for it and it's on
51:03the second page, it's not really going
51:05to be indexed if it can't be found by
51:09Google and from what I know Google looks
51:12at uh
51:13uh no index follow as basically like a
51:17soft no index no follow. It doesn't
51:21really care. Um so I'm really sad that
51:25there weren't that many pieces of
51:27research on pagination on like
51:30experimenting with site maps,
51:34experimenting with content pruning.
51:36There were
51:38loads of studies talking about
51:40uh how they pruned their uh
51:44less
51:45well performing content which is fine,
51:48but there's not many of them which are
51:52on like really big websites where you
51:54have like a 100 hundreds of thousands of
51:58pieces of content. It's mostly like, I
52:00don't know, they removed 20 blogs, 50
52:04blogs, 100 ones, or
52:06bad performing products is also
52:08something that's never been tested from
52:11what I've seen. So, just removing
52:13completely the products and categories
52:16which don't work.
52:18I've tried it with some of my clients,
52:20but they're not up for it because they
52:23have like huge teams. They have category
52:26managers managers for everything. I
52:28tried to test and and see if we could
52:30put
52:32like really bad performing categories on
52:35no index or something just to see if
52:37that like authority or link juice or
52:40whatever would pass on through some some
52:43other categories which work way better.
52:47So, there were loads of loads of things
52:49which were not tested at all.
52:53>> Yeah. All right. There's something else
52:54that's worth bringing up is
52:57I call it like an SEO change log.
52:59And I've been doing this for a while.
53:01It's
53:02if if you have decent rankings or
53:04starting to have decent rankings is
53:06after the site has been made your
53:08ranking if you think about changing
53:10anything
53:11actually document it in a change log.
53:14Document your current position. Document
53:16what was changed. Document when. Very
53:19importantly, write why you changed it.
53:22And then and then revisit this the page
53:25in 3 months to see if it helped because
53:28a lot of the times you'll have a hunch,
53:30but you're not sure. You'll make a
53:31change. It might be a big change. You
53:34might change you might change the page
53:36title, for example, which could be quite
53:37a big change. You might change something
53:39above the fold. It could be quite a big
53:40change.
53:41And then write it down and write why you
53:44did it and the date and then revisit it
53:46in a couple of months
53:47to see what happened. I think
53:49any SEO who's doing SEO seriously should
53:51be keeping an SEO change log. And the
53:53more detailed you keep it, the better.
53:57>> Yeah, of course. I think so, too. I
53:59think a lot of SEOs and and agencies
54:03I mean me included
54:06in a way just kind of do stuff without
54:09checking if if
54:11the stuff we're doing actually works
54:13month to month. And like I said, if you
54:16bundle loads of tactics over time,
54:20you're going to you're going to win
54:23eventually. Unless you're doing like
54:24really bad SEO, you're going to figure
54:27out something and maybe you're going to
54:29finally build some really good backlinks
54:32or you're going to create a great piece
54:34of content or
54:36change titles with just updated or all
54:39capitals, whatever. And you're going to
54:42have way better rankings, way more
54:44traffic, organic revenue, whatever
54:46you're you're looking at, but you don't
54:48really know what change really did that.
54:51You just know, "Okay, we've been doing
54:53the work."
54:54And
54:55something
54:57is doing better and that's it. But if
55:00you don't know what
55:01the exact change which helped you, then
55:05you're just doing like really random
55:07stuff until you figure something out.
55:10And you
55:11when you could instead have like a list
55:14of really great tactics with really
55:16clear priorities where you could be,
55:19"Okay, let's not do all of this
55:22stuff around everywhere. Let's just do
55:26this and this and this. These are the
55:28best things we've ever done with clients
55:31in your niche or your industry or for
55:34for like an SaaS or or FMCG business or
55:39whatever. It works for your industry.
55:42This is what we've seen, so we don't
55:43have really have to
55:45uh
55:46charge you like I don't know how many
55:49hours of of work. You you can just be
55:52charged by outcome. So, you're going to
55:54have a great outcome. How much is that
55:56worth to you? And I think that would
55:58work way better for us and for our
56:01clients. I don't mean to say that you
56:03shouldn't do like basic technical SEO
56:06hygiene or content hygiene or backlink
56:09hygiene, but we should look more to
56:12tactics which actually do work and move
56:16the needle, not just uh doing like a
56:19shotgun approach where we do everything.
56:21We have a huge team doing everything and
56:24something will work eventually because
56:25we're wasting time we could probably
56:28invest into other projects or other
56:31clients uh
56:33and make more money generally for
56:35ourselves and our clients.
56:39>> If you're If you're an agency and you
56:41want to do um
56:43you want to do SEO tests so that you
56:46actually get shared. So, you want to You
56:48want to do what you you You want to do
56:50some SEO tests
56:52almost as a linkable asset or maybe a
56:54form of PR. You're like, "I want to do
56:56some SEO tests and get shared on the
56:58AdWords show." Or like, "I want to do
56:59some SEO tests and get and go viral on
57:01X."
57:02What makes an SEO test
57:05interesting and worth sharing?
57:09>> I think interesting would be
57:12a novel idea. So, it's not
57:15Okay, we
57:17>> Pagination. Testing the pagination like
57:19you talked about. That's like that's
57:20something There's not enough of that out
57:22there.
57:23>> Yeah, pagination.
57:25>> But you had a problem with how a lot of
57:26companies were doing their SEO tests.
57:30>> Yeah, I mean I kind of understand them,
57:32Uh but the issue with SEO tests
57:35generally even done in-house is that
57:38they're really not controlled. So,
57:42a lot of these tests are just basic
57:44before and after we made change X or Y
57:47and then after some time we saw that
57:51this happened or this didn't happen.
57:54And that's an issue because you have
57:56loads of factors. It can be seasonality,
57:58it can be uh like different users just
58:02coming to your site. It can be loads of
58:04things which
58:05put your outcome in one
58:08direction or the other. You can't really
58:10know if the change had a direct effect
58:13on the outcome. So, I think people
58:16should get more out of Search Pilot's
58:19book. Uh so, just create controlled
58:22tests. Use
58:24a specific number of pages where you
58:26test something on. So, it's just one
58:30uh change. It's not like five different
58:32ones. You take, I don't know, like 50 or
58:35100 or however many pages to test this
58:38on. It can be layout, titles, adding
58:42videos to articles, which also works,
58:45uh or adding bulleted points, tables,
58:49different colors,
58:51uh whatever, animations, removing
58:53JavaScript, and so on.
58:55So, you do this and check
58:58uh if the control group had like a
59:01differing amount of traffic. So, if you
59:03did only one change, let's say, removed
59:06uh JavaScript and just looked at uh that
59:09the page was generated by CSS.
59:12Uh and you had like a positive move,
59:15uh then you can say, "Okay, this is
59:17something which is objectively a
59:19positive move because the control group
59:21where we changed nothing had less
59:23traffic and it's all of the other
59:25factors are the same or
59:28the
59:29as same as they can be. It can't be
59:31completely the same even when you have a
59:33control group because maybe you just
59:36have like a specific set of users coming
59:40randomly to the pages you've changed or
59:42randomly not coming to the ones you
59:44didn't, but you lessen your chance
59:46chances of of having an issue. But, I
59:49understand why there that's a problem
59:51with many companies uh and clients uh
59:54mine included because you basically got
59:57to sell to clients, "Okay, we're just
59:59going to do one
1:00:01uh this one thing and just not going to
1:00:04do anything else and uh track it for
1:00:07like 2 weeks, and it's probably not
1:00:09going to work." So, um I don't know. I
1:00:12don't know many clients who are inclined
1:00:15to get on that, but maybe you can do it
1:00:17with uh personal projects uh especially
1:00:20if you're not doing well uh in SEO. If
1:00:24you have uh nothing to lose, I would
1:00:27test as many things as I can. From
1:00:30titles, navigation, whatever, adding
1:00:33crazy animations, doing everything,
1:00:37creating
1:00:38uh putting myself in directories even
1:00:40though that's not something that people
1:00:43still do except for a few uh specific
1:00:46directories, but um yeah, I would just
1:00:49do whatever. Just do it controlled so I
1:00:52know what actually moved the needle.
1:00:55>> Yeah, and honestly something else is
1:00:58like just turning turning the uh turning
1:01:00the everyday SEO that you do into a
1:01:04story. And I think Joy Hawkins is a good
1:01:07example of this because she wanted to
1:01:10see if nofollow links move the needle.
1:01:13And so, she heard that she heard she
1:01:15turned this into a story. She's like, "I
1:01:16heard Kyle Roof, who's a great SEO, say
1:01:20that if a nofollow link is on a page
1:01:24that is ranking and getting organic
1:01:27clicks,
1:01:28then the no follow link will actually
1:01:31pass authority and and make a
1:01:33difference. And so she's like, so what I
1:01:35did is I had a page on our local SEO
1:01:39forum
1:01:40that was ranking and getting organic
1:01:42clicks.
1:01:43And I added in a no follow link to a
1:01:48page on our agency website.
1:01:52Uh and there was a YouTube video
1:01:55embedded in that page that was no follow
1:01:58link to.
1:01:59And she's like, it was crazy. I found
1:02:01that not only did the page start ranking
1:02:04for its target keyword, and it wasn't
1:02:07ranking before, the YouTube video that
1:02:09was embedded, which also wasn't ranking
1:02:11before, started showing on page one for
1:02:14the target keyword in the YouTube video
1:02:16that was embedded on the page that was
1:02:18no follow link to. And then she's then
1:02:20she's like, and since then, I've tested
1:02:23this many other times.
1:02:25And she didn't say the all the exact
1:02:27test that she did afterwards. She said,
1:02:29I I tested this many other times and it
1:02:31worked.
1:02:32But it was still in like she told in the
1:02:34story, it was interesting, it was clear,
1:02:37and
1:02:39I I made an episode about that, and a
1:02:40lot of other people shared her findings
1:02:42and shared Kyle Roof's findings. And um
1:02:45making it into just a story, making it
1:02:47into something that's concise, easy to
1:02:49understand, and fun to read.
1:02:51>> Yeah, that that's an example of
1:02:52something I I had a hunch on. So maybe
1:02:55just uh listen to your hunch there.
1:02:57>> Yeah.
1:02:58>> I always had the hunch that it didn't
1:03:01really make sense to me that no follow
1:03:02didn't work because obviously a no
1:03:05follow from the New York Times is better
1:03:07than a follow from some car mechanic's
1:03:10website. It's obviously going to pass
1:03:13some sort of authority. Um I didn't uh
1:03:16think it was like one to one maybe, but
1:03:19I always had a hunch that it had to be
1:03:21really really similar. So, I think for
1:03:24for these kinds of tests, uh it's always
1:03:26good to like just go and see if your
1:03:29hunch is
1:03:31is something that actually works or or
1:03:33makes sense. Because we all have them.
1:03:35We all work with different kinds of
1:03:38clients and see things which are
1:03:40really weird. Um, like for instance, I I
1:03:45always thought that
1:03:47from
1:03:48the start of my SEO career, I always
1:03:50thought that open graph images for
1:03:53Facebook had an effect on
1:03:55the images in the SERPs. I always told
1:03:58people this,
1:03:59and people didn't really think it made
1:04:02sense. But,
1:04:03>> that. Talk about that. Talk about this
1:04:05theory. Explain it.
1:04:06>> Uh, so you the open graph tags on for
1:04:11for Facebook on specific page. So, the
1:04:14image which is used
1:04:17in the open OG image is something which
1:04:21is really really often found found in
1:04:25the SERPs. Uh,
1:04:27>> is this is the image like if you share
1:04:29if you share your post on social media,
1:04:32this is the preview image that shows.
1:04:33So, you maybe you have a different OG
1:04:35image than your featured image. A lot of
1:04:37the times if you just share if you share
1:04:39it in iMessage, it'll use the OG image
1:04:41as well.
1:04:43Um, so the the OG image is used as a
1:04:45preview when you share links on your
1:04:47website, on messengers and social media.
1:04:50And there's lots of different tools to
1:04:51debug OG images, make sure that they're
1:04:53working. Um, and so your theory was
1:04:56what? That OG images are used in the in
1:04:58the search engine results pages in
1:04:59Google images or somewhere else?
1:05:02>> Yeah, just the main image in the search
1:05:05engine result pages. So, especially if
1:05:07you have products, and
1:05:10I saw that a lot of the time
1:05:13these products had like 10 images which
1:05:16were mostly identical, maybe some
1:05:19different like rotation of the shoe or
1:05:21whatever else. And it was almost always
1:05:26the OG image specifically which was
1:05:29used. Sometimes it happened that
1:05:32navigation image was used for
1:05:36this happened with a few of my clients.
1:05:38So if they have like uh
1:05:40images which are connected to categories
1:05:44in the navigation, then that specific
1:05:46image could also be found on the serp
1:05:49sometimes because the Google looks at
1:05:52okay, this is the first image on the
1:05:54page. It's probably the most important
1:05:57image. That's just
1:05:58something which sometimes happens. And I
1:06:02always thought about this about the OG
1:06:05image tag and I think two or three or
1:06:09four years ago
1:06:11someone actually proved that this
1:06:14happens. I think if I remember right.
1:06:17And I always wanted to say ah, I told
1:06:18you so
1:06:20but I had no one to tell it to because I
1:06:23mean no one really cared. It's it's such
1:06:25a really specific specific technical SEO
1:06:28thing. Doesn't really matter all that
1:06:30much. But it was interesting to me. I
1:06:33had like a really good hunch and the
1:06:37only thing I'm sorry about is that I
1:06:38don't I don't test these hunches as
1:06:42often as I should. I should test them
1:06:44all the time. The the hunches I have
1:06:47>> other big the other big thing is is when
1:06:50you actually share the results, you got
1:06:51to share the implications of the
1:06:53results. So like why does it matter that
1:06:56you're seeing the OG images? Why does it
1:06:58matter that the OG images are actually
1:07:00making a difference? How can a company
1:07:02use that?
1:07:04>> Well, my idea was that we could then
1:07:08because
1:07:10my colleagues who I worked with at the
1:07:12time always thought that it was
1:07:14basically more or less random
1:07:17if you have like generally identical
1:07:20images. But if the OG image is actually
1:07:24something which is used then you can use
1:07:26it to
1:07:27increase your CTR. So you can actually
1:07:29choose, okay, I want to put this in the
1:07:32SERPs and not this other image which
1:07:34sucks and is not
1:07:37looking at the product really well. And
1:07:40we have this other image which is done
1:07:42really well. It looks great. It's
1:07:45different from the other ones on the
1:07:47SERPs. So if everyone has like a black
1:07:49background, we have a white background
1:07:51then we stand out. We get more CTR. So
1:07:54stuff like this.
1:07:55>> Right. So you can So you can have a
1:07:57featured image that's appropriate to
1:07:59your brand that makes sense within the
1:08:01page or the post.
1:08:03And you're not always going to want to
1:08:05have the OG image as a featured image. A
1:08:07lot of people like putting text into
1:08:09their into their OG images and that
1:08:12might not be appropriate for your
1:08:13featured image. But then for the OG
1:08:15image to maximize click-through rate,
1:08:17you put some text into it and then this
1:08:19shows on Google
1:08:21and so people are more likely to
1:08:23actually click into your content.
1:08:25>> Yeah, exactly. Like I said,
1:08:29loads of things like this with
1:08:31structured data as well using like
1:08:34specific kinds.
1:08:37Removing
1:08:39for a couple of my clients now, I had
1:08:41the same thing happen. So they had like
1:08:45a
1:08:46They had like an aggregate offer
1:08:49structured data on their pages.
1:08:52They were like legit reviews. No
1:08:56problems with them at all. And they
1:08:59couldn't figure out, okay, we lost
1:09:01rankings for no reason. Everything's the
1:09:05same. I checked it out and everything
1:09:08was the same. I checked the HTML before
1:09:11and after the fall. I checked the
1:09:13backlinks, basically everything. And two
1:09:17times now I had the same issue. So
1:09:20the aggregate offer the it just had
1:09:24I don't know how many reviews, stars,
1:09:28whatever. And I told them just to test,
1:09:30okay, try removing that part of your
1:09:34schema. They removed it and they went
1:09:36back up.
1:09:38>> Why why did you suggest why did what was
1:09:40your hunch about why you suggested
1:09:42removing it?
1:09:44>> It was the only thing I could think of
1:09:47which could make Google think it was um
1:09:52any kind of like suspicious activity. So
1:09:55having like fake reviews or anything
1:09:57because
1:09:59I don't those reviews aren't fake, but I
1:10:02think Google is really careful about
1:10:04these things. I actually kind of think
1:10:06this is just my personal opinion, but I
1:10:09think that aggregate offer and rich
1:10:11results generally are going to be
1:10:13completely removed in time, just
1:10:16completely because
1:10:18a lot of people just misuse it.
1:10:21>> it. They game it all the time. It's one
1:10:22of the most abused things in SEO.
1:10:24>> [clears throat]
1:10:25>> Yeah, they they removed how-to schema,
1:10:28they removed FAQ schema. I think that
1:10:30product schema is next especially
1:10:33because schema doesn't even really
1:10:35matter anymore. Maybe in
1:10:37for some like specific things like using
1:10:40same as to connect entities together. I
1:10:43think that's really good. But other than
1:10:46that, I don't really see
1:10:48good use for it except for rich results
1:10:51while they exist.
1:10:53So yeah, basically I thought about it
1:10:56and saw that the only thing that Google
1:10:59could find suspicious was the number of
1:11:02reviews or the number of stars. They
1:11:04removed it and they just magically went
1:11:08back up the next day even though they
1:11:11couldn't find it find an issue with
1:11:13other agencies or with their in-house
1:11:15teams or whatever else.
1:11:18AIs couldn't find the issue because AIs
1:11:21still don't have like the critical
1:11:24thinking mode. They don't have like the
1:11:27complete context and one client asked me
1:11:30or I think both of them asked me, "Okay,
1:11:33but everyone does this all the same.
1:11:36Everyone has reviews and everyone has
1:11:39like five stars. Why did I get algorithm
1:11:43algorithmic mail
1:11:45But why did I get punished by the
1:11:47algorithm and they didn't?" And I don't
1:11:50really have an answer for that. I think
1:11:52it it's just generally you get randomly
1:11:56one-shotted sometimes. Things work for
1:11:59your
1:12:01competition, they don't work for you and
1:12:03I never tell my clients, "Okay, just do
1:12:06the thing your competition is doing
1:12:08because
1:12:09maybe it works for them but you're
1:12:11probably going to get one-shotted."
1:12:14There was also one client where
1:12:17their competition had
1:12:20exact match domains for keywords. It's
1:12:24like a huge hotel business and they had
1:12:27exact match keywords for
1:12:30basically everything regarding
1:12:32that specific keyword they're trying
1:12:34like hotels in wherever. And
1:12:38they had three or four of the positions
1:12:42on the SERPs. So four of the positions
1:12:45were just there random like
1:12:48doorway websites basically.
1:12:51And the client asked about that. I just
1:12:53said it they it works for them for some
1:12:56reason, probably because they didn't
1:12:59change enough stuff, so they get into
1:13:01Google's attention to reevaluate their
1:13:04site quality or their backlink quality
1:13:07and so on.
1:13:08There is it's static enough. They're not
1:13:11lifting up any dust and Google is
1:13:13letting them be, but if we put out a new
1:13:15website right now when Google has
1:13:18probably fixed that up, uh, they're
1:13:20going to see you and just one shot you
1:13:23and
1:13:24maybe even get the manual penalty for
1:13:26that because that's that's like
1:13:28that's just a doorway page.
1:13:30It's only
1:13:32the website. It's a complete website,
1:13:34but it's a doorway page basically.
1:13:38>> You've also thrown SEO conferences for
1:13:41some really big people in the industry,
1:13:43right?
1:13:44>> Yeah.
1:13:45>> Who?
1:13:45>> I mean, I I just throw my SEO conference
1:13:48at SEO conference.
1:13:51Yeah, so it's Croatia SEO summit. We
1:13:54also do Zagreb SEO day. It's
1:13:57a smaller one-day conference/meetup.
1:14:00It's in 7 days actually. And we also do
1:14:03Croatia SEO summit. It's in Šibenik,
1:14:06Croatia Croatia on the coastline
1:14:10in the summer.
1:14:11So, we basically just do a combination
1:14:14of masterminds, workshops, and lectures.
1:14:18Uh,
1:14:18we're
1:14:20180 people is our max. Right now, we
1:14:23changed a couple of things
1:14:25from last year. We were trying to be
1:14:27just like a general conference
1:14:30before, but uh,
1:14:32we kind of pushed back on that and now
1:14:34we're looking at
1:14:36being something more akin to what we
1:14:40would want to see. So, we removed
1:14:43everything which is like general state
1:14:45of search
1:14:47talks. I don't I don't want to hear
1:14:50about those, so I don't care. Um we put
1:14:53in like uh fuck-ups and uh
1:14:57and wins. So, just like sharing, you
1:14:59have 12 minutes, you share a fuck-up,
1:15:01what you did wrong. So, it's not all we
1:15:03have like amazing outcomes. It's also
1:15:06like I I think the fuck-ups format is
1:15:09generally great for conferences.
1:15:11Uh workshops are also really popular for
1:15:15us. And um yeah, we're going after like
1:15:18being a smaller conference with having
1:15:21uh like ge- genuine uh industry leaders
1:15:26sharing like genuine tactics and stuff
1:15:29which works, which is not like just
1:15:32state of search, which is fine.
1:15:34Fine for everyone. It's fine for me also
1:15:37when I go to some other conferences, but
1:15:40for my conference, I want to do
1:15:42something which I want to hear and talk
1:15:44about, not just like ge- general stuff.
1:15:47I want to get into the real core of the
1:15:50issue.
1:15:51>> Who is Who are some guests that you've
1:15:53had at your events?
1:15:56>> Uh
1:15:58you mean uh like lectures or
1:16:00>> names.
1:16:01>> Yeah, sure. Um Lily Lily Ray, Kyle Roof,
1:16:06uh Dan Petrovic. I love Dan Petrovic. He
1:16:08does like really Yeah, really great
1:16:11stuff. The research he recently put out
1:16:14about um like contextual
1:16:17uh uh links in specific like uh pieces
1:16:21of content regarding to link building.
1:16:24Really great uh
1:16:26If whoever is listening to this, please
1:16:28just Google Dan Petrovic and find his uh
1:16:31latest research.
1:16:33Um so, yeah. Uh
1:16:36who else? Mark Williams Cook. Uh
1:16:40Uh
1:16:42Who else? There were so many people I I
1:16:44completely
1:16:46uh froze right now. But yeah, there
1:16:47there were like
1:16:49uh 40 or 50 uh speakers by now. Uh also
1:16:54uh Harry Clarkson Bennett, uh
1:16:58Barry the I don't know. Just like loads
1:17:01of
1:17:01>> Barry Schwartz
1:17:03>> No, no, no, it wasn't it wasn't Barry
1:17:05Schwartz. It was Barry Adams.
1:17:07>> Uh
1:17:08>> Yeah.
1:17:09>> Are there any uh are there any like are
1:17:11there any widely held SEO beliefs you've
1:17:13heard experts privately disagree with
1:17:16behind the scenes?
1:17:19>> Well,
1:17:21hm, not really. Everyone is I think way
1:17:26too polite to um talk about these
1:17:29things. And if it's completely like off
1:17:33the record stuff, like we're uh at some
1:17:36bar somewhere and just drinking,
1:17:39uh we're not really going to be talking
1:17:41about SEO that much. It's it's going to
1:17:44be like politics and religion and all of
1:17:48the fun topics. And maybe some SEO
1:17:50sprinkle sprinkled in, but when you
1:17:53combine uh like having a few drinks with
1:17:57uh with with with people which are who
1:17:59are kind of strangers but kind of are
1:18:02not, you're going to get to the topics
1:18:04which are the most fun eventually.
1:18:07Uh but but yeah, I I I never heard that
1:18:10someone was the directly or indirectly
1:18:14uh
1:18:14saying that they don't agree with
1:18:16someone. May maybe about some small
1:18:19stuff, but it's small enough that I
1:18:21don't really remember.
1:18:25>> Yeah. Cool cool events though. And thank
1:18:27you for doing um this crazy research and
1:18:30and giving this to the community for
1:18:31free.
1:18:33Just uh putting this PDF up. Again, like
1:18:36there's no requirements to get it. You
1:18:38can just you go to Yeah, you can go to
1:18:40it. Just click it and then download it
1:18:42and uh it's super cool that you did
1:18:44this. And thank you again for messaging
1:18:45me and for um
1:18:47and for coming on. And And you you said
1:18:50you prepared a lot for this episode. And
1:18:52I really Yeah, I really appreciate that.
1:18:55>> Yeah. I mean, I I had to go through
1:18:58uh 700 tactics manually to see what I
1:19:02would single out as uh some of the best
1:19:05stuff.
1:19:06>> If you're going to leave If you're going
1:19:07to leave leave people with um one thing
1:19:10that they should do tomorrow,
1:19:13what would you say?
1:19:16>> Um I'd say
1:19:18either
1:19:20uh just like do digital PR. It works
1:19:24great for everyone, basically.
1:19:27Uh instead of like traditional link
1:19:29building, it works way better. Uh that
1:19:32that's my uh opinion as well as uh much
1:19:35of the research.
1:19:37Uh the second thing is um just like be
1:19:40[snorts]
1:19:41properly full with context on your
1:19:45website.
1:19:47Uh so, everything about your brand,
1:19:49everything about your product or
1:19:51service. So, use also asked Use like uh
1:19:54people also ask Ask your customers
1:19:57directly. Ask your sales teams, "Okay,
1:20:00what do customers ask about or complain
1:20:03about?" Uh
1:20:05just like do the work and see everything
1:20:08which could
1:20:09which
1:20:10uh
1:20:11a customer could ask about your product
1:20:13and service and uh reply to it in some
1:20:17way. So, either add like uh
1:20:20pro- like a statistic, a table, add like
1:20:24frequently asked questions, add blog
1:20:27posts about it. Uh whatever YouTube
1:20:30videos, especially YouTube videos, work
1:20:33great generally for SEO, especially
1:20:35right now.
1:20:37Um, so just like being completely
1:20:40up-to-date with what the customer wants
1:20:43and needs
1:20:45uh and the questions he has because I
1:20:47see a loads of companies disregarding
1:20:49this and then
1:20:51uh let's say a customer asks uh Chat GPT
1:20:56uh
1:20:56okay, does
1:20:58hotel X have is it pet-friendly? And
1:21:03uh Chat GPT says no, even though the
1:21:05hotel changed it to be pet-friendly
1:21:08because they never put out any piece of
1:21:10content about it. They didn't Maybe it's
1:21:12like in some booking process or
1:21:15somewhere in some hidden in some
1:21:17JavaScript or in some video or
1:21:20animation, something that Google maybe
1:21:22can't see. You have to be explicit with
1:21:25these things, especially if it's a big
1:21:27change. Like we changed our company name
1:21:30uh from the Canonical SEO agency, which
1:21:33is a mouthful, to Obvious Agency, which
1:21:36means it depends in in Croatian.
1:21:39Um, uh
1:21:40we never really put out like a PR news
1:21:43piece about that because we're not that
1:21:46big. We're like a boutique agency. Uh
1:21:48no one really cares that we changed our
1:21:50name and Google picked up on
1:21:53on it and every other AI, but if you're
1:21:55doing if you're a huge company and if
1:21:57you're doing something which changed
1:21:59something for your customers, you got to
1:22:01do PR about it, put it up on blog posts
1:22:05and YouTube videos, somewhere, somewhere
1:22:07really prominent so you don't have like
1:22:10a lost transaction and lost revenue
1:22:13because someone didn't want your product
1:22:16or service uh because they just stumble
1:22:18up stumbled upon some bad info somewhere
1:22:21out there, especially if it's Reddit or
1:22:24something. People talk about loads of
1:22:26things and not enough companies do
1:22:29online reputation management.
1:22:32>> That is That is true, too.
1:22:34Yeah.
1:22:35Thank you, Krešimir. Where um where
1:22:37should everybody find you?
1:22:39>> Um they can find me on LinkedIn,
1:22:42obviously.
1:22:44Uh I don't really post on on X or
1:22:48I I sometimes write
1:22:51uh content for some uh media, but I I'm
1:22:55not really out there on anything other
1:22:58than LinkedIn or they can contact me via
1:23:01our agency website, but LinkedIn just
1:23:04works fine.
1:23:06>> Thank you again for coming on. This was
1:23:08a lot of fun.
1:23:09>> Yeah, thank you. Thank you very much. Uh
1:23:12I didn't really think you're going to
1:23:14uh pick up pick pick up on this. Uh so
1:23:18really great to talk to you. Thank you
1:23:19again for
1:23:21uh having me on and love your show.
1:23:24Uh going to I I'll
1:23:26honestly never really
1:23:29uh watched it before someone said to me,
1:23:31"Hey, you you were on the Edward Show."
1:23:34and
1:23:36then I checked it out and I see that's
1:23:38really really great. I'm I don't know
1:23:40how I never stumbled upon it. That's
1:23:43really weird, but going to be honest,
1:23:45I'm going to watch it from now on all
1:23:48the time.
1:23:49>> that. I appreciate that, Krešimir.
1:23:51Um all right, this is episode 1,207
1:23:54of the Pod 1,207
1:23:57days in a row doing this podcast.
1:24:01If you watched this on YouTube, thank
1:24:03you so much for watching. If you
1:24:05listened on Spotify or Apple Podcasts,
1:24:07thank you so much for listening.
1:24:09And I will talk to you again tomorrow.
1:24:13Bye now.

About this transcript

This page contains the full transcript of He Dug Through 9,249 SEO Case Studies. Here’s What Actually Works by Edward Sturm, generated from the public captions YouTube serves with the video. The transcript has 12,355 words across 2,040 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 translate the transcript into 100+ languages.

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, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.