UX Design to WIN in Business

More and more people just use chat GPT instead of Google.

More and more people just ask their Gemini AI assistant for information on a certain topic
without ever visiting your website.

Is that something that you're concerned about?

I'm picking up today with SEO and marketing expert Pietro Mingotti from Fuel Lab.

He's just come out of creating a very comprehensive white paper.

Actually, I think he spent three or four months of research on how LLMs actually work and
what this means for SEO.

and for really corporate marketing websites.

Welcome back to the podcast UX Design to Win in Business.

My name is Niels from Dinghy Studio and let's get right into it.

Welcome back to the show.

Yeah, we're recording this today.

Because we, as many of our listeners probably know, have a bunch of websites for our
customers.

are a creative agency.

Pietro is an absolute marketing and SEO professional.

Like that's what we talked about last time in the show.

And we were having more and more conversations about how people obtain information.

recently and how behavior might've changed.

And so I don't know about every one of you, but I know that the icons on my home screen of
my phone definitely changed a bit.

So usually I just opened Safari and just started to Google stuff.

And recently there's two new icons on that home screen and one is perplexity and the other

is clawed and there's also a chat GPT icon somewhere on another page and like all of these
things are there em and we started to talk about what do people do like now like what does

this does this sort of behavior to just ask a question and get an answer through an AI
impact website performance

so much and I started to shop around and Pietro told me that he came out of a very intense
research project getting into the weeds about this really.

So I read through it, amazingly thorough work.

I learned a whole bunch about AI, but maybe you could give us a little bit intro of what
you were getting into there and what was the goal and all of these things.

Sure, absolutely.

Well, I've been doing SEO and I've been a search engine and search marketing expert since
I started since 2006, 2007 when we easily manipulated links on SERP and so on.

So SEO and generically, would say search has always been my field of expertise with
marketing.

And of course,

2020, 2021 brought a lot of things, brought enormous shifts in our lives between COVID and
lockdowns, but also AI models released to the public.

And AI is a big buzzword, but it would be already important to state it's not truly about
AI, it's an MLM, which is an artificial intelligence model that does one thing.

It's a large language model.

It knows language.

So...

When these things, when these tools, when these technologies, they started a global race
for AI, of course, we started to observe a shift in human behavior.

So it's an enormous shift.

It's as big as the invention of electricity, or industrial revolution for organic search,
because pretty quickly, a matter of two or three years,

we have all changed the way we access information.

And so in the industry, immediately there are some people who jump on the boat of
reposting.

Or like we did at the beginning with SEO, if you want to learn how to influence a system,
you need to learn first how it works.

If you don't understand how it works, there's no way you can influence it.

So I was not really keen.

to be honest, to start this research project because I am young on one side, I'm 36 years
old, but if you look at our industry, I'm pretty old.

So I was feeling like, let's the new generation handle all of this AI, but then you can't
because as you said, it impacts today and forever the traffic acquisition and the

possibilities of websites online for everybody, including our customers at Fuel Lab.

And so I decided, okay, let's actually learn how an LLM work and therefore does learn what
can we do to influence it to our advantage for marketing.

Yeah.

And so the trigger term or like the thing that has been discussed that came to my
attention.

Yeah.

Right.

Is this term of geo like generative engine optimization.

Like I remember that I, don't even know where I picked it up, but I was reading through
this and just saw this new acronym and thought like, huh.

interesting.

It's not even gonna be SEO.

Like they even made up an entire new thing for it.

And then I did some research and was like wondering if people offer this as a service, you
know, like as you could buy SEO services, is there somebody already offering GEO services?

And there's a whole bunch of agencies doing that already.

um Of course, right?

Most like a lot in the US, a lot in the UK, a couple in Europe, but so generally it's,
it's a thing and it's sort of packaged together with other SEO services.

So I think, I just wanted to ask going into this, like, do you reckon this is even like a
new industry or is this just a new label on a thing that we've all been doing for a while?

What do you think about that?

There's only one industry.

The one industry is called marketing technology, market, digital marketing.

It's the same thing.

Here we're talking about traffic acquisition, brand awareness.

It's the same usual things.

It's a different technology that changes everything.

But GEO or AEO or however you want to call it is, of course, many agencies, our agency as
well, has a page for it.

What do you want to do?

You want to just disappear from the...

Of course, everybody has to do it.

What's wrong about it is how it's sold as something entirely new, specifically done to
manipulate AIs, which you can't do in no possible way.

Zero point, period.

And we'll get to that.

Why?

But it's true that SEO alone, when you think about organic search,

is not sufficiently longer.

And I started this two years, three years back with organic search engineering.

So already we needed to think outside of the simple on-page, on-site and off-site
optimization to leverage to the best organic strategic positioning.

We had to think about search demand and search demand generation, which already, you know,
if you've been doing SEO the right way, you're already

ready to change how we work and to develop your own blueprint.

And your own blueprint depends on the specific company you're working with and the project
and what you're doing.

So GEO, it's not entirely wrong.

It's not entirely just a buzzword that everybody use.

It's about how you use it.

And if you have took the time to study how the system works,

if you took the time to understand and test and try and see what you actually can do.

But if it's data driven and you can provide sustainable results, yeah, call it Geo, call
it Banana, call it however you like.

Yeah.

I think when I was reading your blog post, which is called, you can't rank on chat GPT.

No, that's not entirely right.

Yeah?

Yeah, yeah.

that's the one thing that makes my blood pressure skyrocket.

So yeah, that gets on my nerves.

Rank on chat GPT in three easy steps.

So first of all, don't rank on chat GPT.

So the issue I have with it is I understand that I'm famous for not being a good
politician.

So if you need to bib this or cut this away,

do so.

what gets on my nerves is that an enormous amount of professionals, or supposed to be in
our industry, are just completely showing to the entire world they don't even they don't

understand the term ranking.

Because in order to rank, there must be a gradatory, a list to rank.

a one, two, three, four, five, 10.

I don't know.

What's the, how long is the serve at the moment?

I don't even know.

the thing is that's a search engine.

The search engine does that, has a ranking system, has an enormously complicated internal
heuristics systems based on a number of algorithms.

And yeah, but at the end there is an index of results and there is a ranking within that
index.

LLM is a generative model.

There's no index to rank.

You don't rank.

Okay, so when I read, when I started writing, reading on LinkedIn, all of these, you know,
posts.

Mm-hmm.

How I ranked on chat.gpv?

No.

That's too much.

was sort of the way where you thought, hold on, so now we're getting into this.

First of all, this can't be right because there's no index.

Okay, so I mean, people, please, we all used some sort of generative model.

Like we all use ChatGPT and Gemini and Claude, and we all went into the chat and asked for
stuff and got a text back, which usually doesn't start with a one, two, three, four, five,

six in a ranking.

So...

which also got you into really understanding how it works, right?

Like, because being opposed to saying ranking on chat GPT is probably, well, that's one
thing, sure.

Like that's a, that's a fight to pick, sorta.

but I think what's really important here is sure you can't rank on chat GPT, but the
question, and why also...

So many in the industry are jumping on this entire hype train is, and that's what we see a
lot with in conversations I have with our customers is I think because everything is

moving so fast and because everything is still so young around the AI tools that we have
today, there is, I think enormous amount of hype that also, yeah.

may like spreads uncertainty in all sorts of directions, right?

Like there's so many companies want to implement AI automation workflows now because maybe
they can be more efficient and everybody seems to do it.

And then in the case of websites, there is talk about, well, but even so Google's AI
summary.

is cannibalizing their own business because nobody's clicking on a link anymore.

And there's like all of these discussions swirling around.

Where I think why I was really interested to have the podcast with you today as well is
that I'm with all of these AI topics.

think in each of them, there's probably some truth.

There's some something there that started the discussion.

And I'm trying to

to find the actual gist in things, you know, like what is actually a paradigm shift that
people will have to pay attention to and will have to make a decision in and what is maybe

stuff where we can just wait a couple of weeks and it will blow over.

Right?

And so as you said, like the generation of interest in general, so people go out.

and ask a question on their LLM now maybe that would involve coming up with an answer that
is based on your content, for example, because you have some domain knowledge.

And these sorts of things, I think that's super interesting.

what are, if they call it geo or if they call it whatever else, what would you say is
actually to be taken serious?

about people just going to JadgpT to ask stuff like what what do think for people who
consider their website an actual sales channel or like a you know like they reuse it in

the way that it's intended to be if it's a marketing website how much should they care you
know like

Well, you should care because you should care.

Because everybody should care.

Now, panic is something else.

First of all, if your user has to convert on your website, it still has to come to your
website to convert.

So, LLMs are not going to be eating up your sales channel.

If you're selling something, something I can buy, I need to come to your website to
purchase it.

And if one tomorrow, I mean, Amazon was...

much worse in that sense.

Now, the real killer here is for informational queries.

things like...

who was the king of France in this ages, right?

Informational queries or who did Leonardo DiCaprio kiss in that scoop last week, know,
stuff like that.

So news websites, media websites, all of those properties which leveraged informational
query traffic and they had as a business mechanism selling ads.

So rendering ads on their portals, yeah.

Hmm.

Yeah.

I see.

Because they're really like all of the, like the big number of visitors coming for a small
piece of information, which is now just surfaced in the AI summary or something like that.

They just don't come to your site anymore and therefore don't have the potential to click
on your ad that you were selling as an ad space, for example.

Gotcha.

those ads any longer.

You have traffic.

So you get to see a lot of these charts on LinkedIn.

SEOs are posting the crocodile effect.

Just fucking Christ.

Such a creative.

the terms.

It's all so cool.

It's crazy.

Okay, so you get to see like impressions going up and clicks going down.

So this is not industry-wide.

In fact, I will send you screenshot of all of our search console properties that don't
have this effect.

Why?

Because we mostly, 99 % work with e-commerce and corporate websites that have to convert
leads.

So everything conversion-based is not impacted this way.

This is informational queries.

or also tools, like something that I used a lot of time on a browser is calculate the
percentage change of.

And so if I'm drawing a report where I'm showing the changes in the data between this year
and last year, I will quickly use these websites where they also show a lot of ads all

around.

And you go there and use a tool, it's useful, and deal done.

Well, now an AI.

an MLM can do that really well.

So also these things, tools, an AI incorporates agents.

So also the weather forecast as an example, weather forecasting websites are heavily
impacted.

So it's all these kinds of properties.

However, even if you have an e-commerce, you already know it's a struggle.

You need to think about brand positioning.

You need to think about a way to get them customers to come to your platform and choose
the person.

purchase your product in their place.

And it was always difficult, right?

Now we have another layer of positioning and technology.

And it's a layer where we have very limited control, but which can also leverage
unexpected advantages.

As an example, if you are selling, I'm Italian, let's say I have an e-commerce has been in
Italy so far.

Now I want to do a market penetration in Spain, right?

Well, with SEO, that would have taken years.

Years.

Now, with LLLM, somebody in Spain might ask the system where to find this kind of product,
and it will, it's possible, it's completely probabilistic, but it's possible that it's

going to offer my product and my link to my website immediately, right?

So, it depends, it depends.

There are couple of technical things that I would like to say because otherwise I think
it's important.

uh

So I totally agree with that.

Just to add one more thing before we get into the technical stuff.

So part of the conversations that I saw pop up on socials, on other people's blogs, where
they were saying, the, yes, we have less visitors overall, but we're also seeing...

actually an increase in conversion rate, like if we try to map it out in the new way,
because like bounce rates are less, like people come apparently just more informed, sort

of ready to buy if as if they did all of their research in secret at home before.

And so I think that's very interesting.

And for me, like

It sucks for ad space for sure.

Let's talk about that a little bit later too.

but for me, like if I know that I see people coming to our agency's websites, who
apparently already sort of know what we do and then go around to validate the assumption

and then just reach out and book a date.

That's like, for me, that's almost better than it was before, because there's a lot less
noise.

Like the people are coming.

are sort of actually on a mission to go somewhere on our side.

And don't just open the tab to close it again because it's absolutely not what they were
looking for.

So yeah, so let's get into what you should be doing maybe from a technical perspective.

Cool.

there's a lot you cannot do and the list of the things you can do is pretty short.

before what you can do and what you can't do, let me tell you why.

So people who listen will make their own informed choice and hey you don't need to
download my research, make your own research and find out.

So this is the point.

These services are based on LLMs, which are based on transformers, which is a specific
kind of technology.

The way this technology is built for each specific model is that it's trained on an
enormous training corpora, a corpus of training, and this is the only place where the

model builds its knowledge with quotes.

So normally this corpora includes, well,

hundreds of thousands of books and which have been digitalized but also hundreds of
thousands of websites but also Wikipedia but also all the scientific research on PubMed,

ARC-Civ, all the possible code on things like you know all the repositories the code pen
all of these things that we use so all of that all of what we have produced has a

humankind

so far, as much as possible of that, is scanned and it's not saved in a memory as
knowledge.

Like the transformer cannot go back in its database and find that web page because it
knows it and it can serve it.

No, it has only used it to get more more accurate in understanding which word goes after
this word.

in the context of this sentence to make it as simple as possible.

The only goal of the machine is to predict efficiently the next token.

It doesn't know truth, it doesn't know lies, it doesn't have a concept of real or false,
it doesn't know facts, it doesn't understand you, to be honest.

It understands its job, which is generate an output from an input.

that gives as close as possible to 100 % correct the next word after the next word.

And it works.

It's incredible.

It works.

But this has implications, though.

It has implications.

The first implication that we learn here is that the system doesn't know about you and
cannot formulate words about you or seemingly about you if you were not present in the

training.

So unless you were present in when it learned, it doesn't know.

But wait, there's RAG.

So there's a technology that now it's default on all the models now, which is it uses its
corpora, its training corpus, but then it also uses a query fan out system where it makes

from your prompt a number of long tail keyword searches using a...

Bing API in the case of ChatGPT or Google Search in the case of Gemini and so on.

So now there is search always attached to the model.

So it uses all it knows for formulating language and plus it sources through search the
new facts.

the same.

Okay, these are the two things.

what you're saying is that first of all, like, so this whole thing with the training data
and the corpus of information, we all know that from before when the huge discussion was

that chat GPT can't give you any references or any citation of, right?

Like two years ago, there was the whole thing.

Like you could ask forever.

And it's exactly what you described.

it couldn't give you a reference because it didn't know about it anyway.

Like it didn't store it in a way that it could get to it.

And so now, this is sort of alleviated by attaching a new thing, like where there's this
new model that comes out in instant GPT-5 and it's Claude 4.1, like all of these things,

that's when a training is finished, right?

And they have trained this new thing.

And now there's the rag.

There's like a piece of technology that you attach to this model that has access to a
search index.

So in the whole panic of SEO and geo and whatever, there's search after all.

Right?

Is that what you're saying?

I'm sorry for all the folks out there who like to say it every year, but SEO is not dead.

It's still...

Because if the model is going to use ragged at every single search, SEO is even more
important than it has ever, ever, ever been.

Now, SEO's importance goes even beyond, but there are some issues.

So, this was already fathomable if...

any other SEO professional is listening.

We noticed, no, we knew when five years back or six years back, Google started to change
its ranking features to rank always conversational forums on top, Reddit, Quora, things

like that.

Now, aside from the financial agreements between the parties and so forth, this was to
train the model that needs

needed to have ranked things that were always fresh, always new, always conversational and
to learn how people want to interact.

First.

Second, we noticed with the helpful content update.

Helpful for whom?

Why did the search result ranking high needed to have trustworthiness, authoritativeness?

Being useful being straight to the answer to the point.

Yes Sure We can argue and speculate that Google wants to offer the best results to their
customers and that's always been But maybe it's also offering the best results to its own

training system Because in fact if it is true that we have the corpus and we have the rag
module we also have the synthetic data on top of it because now to train the model

in more efficient way because you need to imagine that training a model costs millions of
dollars and a very long time.

So in the efficiency of this, companies like OpenAI, they build their own synthetic index
of information and results.

so the model might be using its own synthetic data instead of rag all the time.

So it's a constantly evolving situation.

The thing that changes though is that now, OK, so you want to be ranking anyway with SEO,
with search engine optimization, because you want the model to see you.

When it works with RAD query fan out, I need to find your result.

But then there's things like extractability and citability, the fact that you're ranking
first on or third or fifth.

on a search engine that after all is still built for humans doesn't mean that the content
of your page is easily understandable for the technology in choosing how to cite and

source information from websites because that's a different story.

And so if there's the visibility issue, that's important.

You need models to be able to find you and see you when they use RAG.

That's for sure.

Then there is to do some content normalization because you need the content to be in a
format that is plausible to be selected for citation.

All right, and so like for me, this sound, hmm, is this the same that what we've been
doing before for SEO as well, sorta?

right, so, hmm, yeah.

So because people before, like, I don't know, we've always talked, like when we were
talking about SEO for websites, right, like,

the terms that you also just mentioned, that it's clear and to the point and that the
result is sort of the first because people grow impatient and do all of these things.

So like it sounds to me like a lot of what we're doing now is more of the same just in a
different format or something like that.

Yeah, it's not so far, but there is one pivotal difference.

Earlier, you could measure easily to some degree the results of your optimizations.

And now, you see, while working with algorithms is pretty straightforward in a way, once
you understand the salsa, so to say, once you understand the recipe, you can test, deploy,

and analyze results.

Now,

It's the way that these transformers, technology work is by a gradient descent.

It's probabilistic.

There's no fixed law.

And because it's probabilistic and hyper efficient in doing so, it also falls into the
realm of mathematical irreducibility.

So you can't, like laws of math, say

this tactic will work.

You can't.

You can't say it might work.

There is a high probability that it's going to work.

So this is what really changes.

First of all, it's way faster.

These models are evolving so fast that if you plan a whole website restructuring for LLMs,
maybe, like it happened to me with my research, I posted it when it was finally finished

and the next day GPT-5 was announced and I had to re-go through it.

and updated.

So this is one sensible big change.

The other big change is that you can't measure it efficiently.

You can rely on some companies which are asking a lot of money to tell you how much your
brand has been cited inside of LLMs.

So give you some metrics about these zero-click experiences, but three you can't tell.

Hmm.

There are some things you can do on your server, on server side to assess if you're being
visited and used and crawled by LLMs, but it's really hard to tell.

We have to work by guesswork once again.

I see.

Yeah.

And so...

about what is, so I think what still confuses me a little bit about LLMs in general,
right?

So the idea is that as you said before, it's basically a statistical thing.

it learns on every, by the way, still mind boggling fact, That it, with every model that
has been trained now, it sort of absorb like,

absorb the entire knowledge humans ever put out, at least in the public, to train on that,
which is pretty mind-boggling in itself.

But so...

It's also, there's a lot of talk about that since that's the case, since it learned
everything that's on the internet and that's on in books and everything, there's also a

large degree of mediocrity in that sense.

that it's not, since it's trained on everything and it doesn't know anything necessarily,
and it's mostly statistical.

So.

It does amazing things and it feels almost like there's another person on the other side
sometimes, like, right?

Like the illusion is very enticing, I would say.

But still, like it's only statistics.

It tries to come as close as to whatever you're trying to ask for it and predict whatever
the answer should be without actually knowing anything.

So...

doesn't.

It doesn't even know all of what we produced in our entire...

It doesn't know it.

It just used it to perfect the probability of a successful output.

But it doesn't know.

It's just a way to train even further the accuracy of the prediction of word by word.

But essentially, these tools are the conversational part, at least.

It's just a chat bot.

It's a simulator.

Yeah, exactly.

so, when we're thinking about how we should prepare the content of our website to, as you
said in your research, to increase the probability of being cited, right.

Let's not rank on, there's no ranking, but so to increase the probability of being part of
the training or then later,

being part of the citation through the rag after searching after all.

So if it's always striving for mediocrity, sorta, if it's always trying to find the most
probable thing on everything that it learned, and everything is of course not sublime,

right?

Like per definition, everything is everything, and the medium of everything is, well,
mediocre, I would say.

What do you have to do to your content to make it probable to be cited?

like, you know, like what's the, if we have a blog or if we have service detail pages
where we outline what we're doing and all of that stuff, right?

What would you say is the, like also even from a, from a semantic standpoint, like what
are we doing?

Are we trying to...

dumb it down or smarten it up from depending on from wherever you're coming from or you
know, like are we actually writing new content just for the LLM or are we trying to

summarize it in a way?

Like what's the, what are we trying to do?

Well, the marketer in me would answer to this with buy my book kind of thing, but because
it's the million dollar question now, everybody wants to know how to be more successful in

being sighted in Netherlands.

But of course, I'm just joking.

So of course, I'll tell you a few nuggets.

First of all, consider the technical part.

So visibility hygiene.

You need to have your robot's TXT and also the LLM's TXT open explicitly to all the
crawlers.

So whether you are afraid of your content being stolen, then you don't want to be part of
this race.

Okay?

But if you want to be part of this race, you got to feed that content to the machines.

So one, visibility hygiene.

Make sure that you are crawlable, findable, discoverable, and so forth.

Visibility hygiene also means

SEO.

So you are not going to be part of the training if you were not in the past 10 years.

So unless you have content about your brand on Wikipedia, Common Crawl, all of these
corporate sources that I mentioned, you gotta be there on the search engine.

If you gotta be there on the search engine, figure out, preferably long tail search term
based

What are your pages that are more important and reformat them a little to be useful to a
human being in the terms that it answers specific questions.

Don't try to rank for broad terms because an AI doesn't need you or nobody else to know
what Nike shoes means, right?

So you want to be answering a query, something that a user might ask to a chatbot.

So that is something you can do on

content side and visibility side.

On the format of the contents to be more extractable, well, here's the catch.

These things have to answer in a split second.

So they don't have the time to crawl your entire page, they don't have the time to process
your entire page nor to evaluate it.

So they will assume that if you were one of the lucky few that have been clicked by its...

rag system.

It just wants to find quick, easy, extractable information.

So if your page is long, think about a TLDR block, something that summarizes it.

You want your first paragraph to give out the information immediately.

You want to avoid salesy and marketing terminology.

So the best product in 2025, no.

But

brand provides solution to this kind of customers worldwide.

These systems like also a lot bullet points, info windows like also the latest Google
algorithm update, Movera.

So info windows, small snippets of information, quick, easy, concise, and again written to

answerability left to right, no narrative terminology, no romantic way of talking about
things.

I mean, you can do everything.

You can keep the human part of the page romantic and with a nice storytelling, but you
need some snippets here and there to feed all the core information, both SEO and LLM

equally important.

So these are the main things that I would do if I were you.

And I would also probably think about

schema markup and all the rest of the technical SEO implementations which can make your
content faster to understand in the key facts by a system like these.

yeah, and HTML, semantic HTML are still very important.

If these things, these tools use agents which use crawlers.

So Gemini has its own crawler for

that he uses with Gemini, which is not the same one Google bought mobile, but it's very
similar.

And it looks inside of the page and it looks at the heading structure, H1, H2, H3.

So if it sees that your page starts with H4, then has an H1 that has three H3s followed by
a thousand H2s and then you use H6 here and there, it doesn't understand the context, the

meaning, the nothing.

So...

it drops you and you're dropped.

you know, it's about this mostly.

And so it sounds to me like a lot of these things we've been doing for SEO, right?

Like fast, well-structured content.

That's always been what we should be doing with our websites anyway.

Also for accessibility, like these things have always been a very good idea.

So there's nothing new here generally.

The content part in itself sounds interesting.

It's almost like I, least while you were describing how the content should be and the,
that the, answer comes quickly and that it's very factual and it's like well contained in

little blocks and so on.

Like this sounds a lot like people obtain info, like actual people obtain information as
well or like to obtain information by now.

And so there's like, for example, like there's these, you know, do you search for anything
cheesecake?

and then you come onto this first side and the actual recipe for the cheesecake is like
somewhere down there.

And before there's like this gigantic story about cheesecakes in general and like, I don't
know, like all sorts of stuff where most people probably start to read this and are like,

this is not how I make a cheesecake and then start to scroll down and so on.

So it sounds a bit like the...

the LLM crawler things also behave a little bit more like actual humans?

Or is this outrageous to say?

There are similarities for sure, both human beings, but for wrong reasons though.

Like, it makes sense that an LLM needs to have an answer very fast.

There is a user waiting on the other side of the prompt and it needs to be accurate, fast,
efficient.

Also, there is an enormous processing cost that goes on every single time.

Imagine also how power hungry these machines are, right?

So, makes sense.

As human beings, no, it doesn't make sense because we have a serious, serious problem with
how fast we expect our lives to be efficient.

And we're not actually wired that way, right?

So if you need to get crazy because you need it.

So if you need the recipe for cheesecake, it means you've got nothing present to do in
your life.

You want to make a goddamn cheesecake.

So you might as well relax, right?

And learn something.

m

true.

Yeah, this is the hard pill to swallow, you know.

We are kind of creating the doom, the digital doom that we're in.

We're creating it and we have the power to stop it if only we reminded ourselves that
we're human beings and we don't need to be that fast.

So yeah, if I need to make a cheesecake, I want to find a type cheesecake because I'm too
lazy to type how to make a cheesecake.

So I will just type cheesecake, right?

Okay.

So, I'll pack cheesecake.

It hurts also because it's true.

So, you go and you click the first results.

You say this, please keep the ads because you kind of know that somebody will make money
if you click them, although it has no impact for you.

And maybe if you clicked the ad, you would have found what you wanted, but okay.

So we scroll, first organic index result, we click it.

Okay, I want my recipe.

there's a whole story.

There's also an engagement farming part where if I subscribe to the newsletter, I get this
and that.

And my god, there are ads.

There are ads.

So ads and banner blindness kind of got us used to skim content until we find what we
expected.

And so since we do that and we only look for the factual information, not only we miss the
opportunity to learn something new, but

All the content strategists, all the very good copywriters, all the UX designers who have
thought how to create the best possible experience for my user for my very hardly earned

click.

No, my user has to love this gradient that I created.

My user will love so much the transparency effect.

I have to write five jQuery versions to achieve this, but nobody cares.

Mmm.

want the information.

So I think that NLM should work like that.

think that human beings should instead chill, chill a little bit.

And enjoy, enjoy things because as an example, you remember our conversations started with
me complimenting on your agency's website.

I still love what you guys did with the glass morphisms or the transparencies.

Yeah.

To me, you see, if a person is aware of what happens on the screen, of course they
appreciate it deeply.

But the problem is that everybody runs too fast.

And so they also assume that an answer from an LLM is factual.

Exactly.

So, I mean, this is a whole nother territory, right?

Yeah, yeah, absolutely.

so I, well, as I said somewhere in the middle that like what we're striving to do at the
moment is to keep our customers sane from the hype, I would say.

So on the one hand, I sort of agree that there's probably a big shift to...

how the internet works even, I don't know, like as you also pointed out, we've had this
whole, this whole social thing explode in our faces for the last 10 years.

And I think we're not yet completely done trying to, like we didn't really overcome this
yet or don't really know how to manage, like at least in Germany, discussions about that.

cell phone, like smartphones with social apps on it should not be allowed in school and
like that the ability to learn is decreasing and so on.

we had before we started to record, we had this conversation about this MIT study coming
out that already shows that your brain is surprisingly a lot less active when working with

AI applications in that way.

It proves further, it proves that the neural networks based on the principle of
neuroplasticity that are required to develop your brain faculties at a cognitive level,

they don't develop the way they are supposed to when you are in the development phase.

So yeah, literally we can see it.

AI is going to be raising a generation of...

cognitively impaired people.

Yeah.

And I think so from, from this, from this sort of standpoint, I mean, this is probably
fanning out too much from, like for listeners who came for the, for the B2B kind of

business advice on what to do with, with AI in, their context of websites and so on.

But we did have a discussion.

with most of our customers about if they even wanted their website to be part of the
training, for example.

And I think, right, and there's like a whole, there's so many different sorts of stances.

Like there's the, like the very simple one of like, so I sort of want people to come to my
website and not just be sighted in a chat, so to say, because I want this brand

visibility.

I want people to come for whatever reason.

But there's of course the much bigger discussion on how ethical is it even to just gobble
up the entire knowledge of humanity for training purposes and to then put certain people

out of business and this whole thing.

so I think it's super relevant to stop every once in a while and reflect on what you're
doing and why even.

Right?

And so I can only urge people to actually download your white paper on how this works
because it's super eye-opening when you just read, it's disclaimer, it's technical.

All right?

So this is not, I've got to read something.

Very interesting with my morning coffee.

Like you can, but get a second one because...

This is just really, I think it's really mind bending to understand what's happening with
all of this.

mean, again, imagine how much text is the publicly available knowledge of humanity.

It's ridiculous amounts of text.

Right?

And the sheer...

billions of billions of billions of dollars.

It's amazing.

And the sheer fact that in the last three years, this was gobbled up multiple times to
train new models is pretty ridiculous in itself.

I would say like that is pretty mind bending.

but so I think it's important for people to reflect.

if you're looking for an easy way out or for an automation of some sort or

Like for an application where you have some sort of your business where you think this
could be more efficient and maybe I can replace 10 people with AI or something.

I think it's very important to keep an eye out, not only for the factual result of like, I
have this thing in whatever quick time, but for the fact that we have to stay together as

humanity, like the empathetic part.

It's just not here at all.

like we have multiple conversations about user research, for example, something that we do
a lot.

And if we couldn't replace our entire user research with AI tooling, right.

And my short answer to that is always, well, but we're not building the product for an AI,
we're building it for a human.

And humans are like super fuzzy, weirdo things that don't...

act in any reliable way actually at all.

And so for me, it takes a conversation with the person where I can see their facial
expressions of when I, you know, like for me, a test is successful if I just ask a

question and I see them go like this.

No, okay.

That was not the right question.

And they don't have to say a word.

And that's just not something that you can do through chat GPT, I would say.

Right.

Yeah, if I have a suggestion for your, you know, people who are listening that are looking
for business advice, I would also suggest to stay human.

I do understand that the feeling of I need to rush into the boat immediately is really
pressing and you should rush to the boat, but don't think this is your business.

Don't think about it that way because everybody's doing the same thing and as

Nils said when everything is everything there's just mediocrity and you're never going to
win the big companies.

Apple is still going to be Apple, Google is still going to be Google, Nike is still going
to be Nike.

You're never going to win this way.

If anything you're going to be winning now more than ever with your authenticity.

So now is the moment to stay human.

Now is the moment to be human.

You just need to though always remember that marketing is marketing.

And so unless you don't want anybody in the world to see your shop, so to say, then yeah,
then you can hide in a pleasant, stranded desert island and just enjoy life.

I would do that.

I will join you.

But if you want to market your product or your service or your brand, you need to be where
the people are and you need to be in the front row too.

So, so consider how the LLM work, understand

How can you best leverage it to expose you the most?

And then just keep an eye on those branded queries in your Google Search Console reports.

Try to see if you manage to have more branded queries.

Because if you're not doing any new marketing efforts, but trying to work with LLMs, you
see branded search escalating, that might be that you manage to be narrated by an AI, by

an LLM.

And that's for sure an enormous marketing issue.

I see.

Yeah.

Man, we could definitely talk about this for a much longer time.

this is like, it was again, like that, I don't know, we started to have these
conversations and then you reached out and said, look, I've actually just did this for the

last three months, working out this white paper and doing the research.

I thought, awesome.

I'm obviously not the only one.

trying to navigate this entire new space.

that's really cool.

But the way you did it was like, I, again, read the blog post, get the white paper.

And for me, it was very, very educational to understand the tech behind it and to make a
much more informed decision on what we want to do.

Like from Dinghy's perspective, what do we want to do?

for everything that we talked about today.

And for example, we're using AI tools in many different ways.

Like we're using coding tools, we're using AI to transcribe interviews, we're doing all of
these things.

But the content that we're putting out, that's something that we're writing ourselves.

Because our business is to, well...

help people with communication and to empathetically understand each other's standpoint.

Like that's sort of the core of why user research and usability research is a thing that
helps is because you actually understand what the other person wants.

And if we write about our services or if we let write about our services from an LLM,
which is well...

going to be a lot quicker, but mediocre, then we're failing our service.

Like that is exactly where we draw the line, where we use all of this amazing new
technology to maybe give us more time for what's important.

And for us it's important to keep conversations between humans, I would say.

Yeah.

Well, there's also the emotional aspect indeed.

How would you feel if after all our conversation today and all the time that you spent
reading my white paper, if I told you that I did it entirely with AI?

Because I was told that.

I was told by somebody who, you know, who accessed it and I was told, have you done this
with chat GPT?

I was like, well, first of all, I would have used Gemini because I used that, but no, put
a bit here, but like, fuck no.

I felt so violated.

so like, I didn't spend time with my husband or my dog or my friends, like beyond work
time.

I worked the nights on this thing.

I still have the InDesign file.

Like it was all organic and analogic.

But it's just normal to assume that a piece of content is not made by a human being.

But I think the question comes from a good place because this person, I guess, would want
the real thing as in being written by you, right?

And with the knowing that you put in the work and that you actually used your brain to
reflect on all of this and to not have the AI slob, I guess.

So I think there's like a...

It's probably a little bit of rude way to ask it straight out.

But I guess that person was very happy to learn that all of this is your work and that
it's not just generated by chat GPT.

I don't know, hey, just downloaded my paper and disappeared, but I hope you enjoyed it.

I hope you enjoyed it.

Maybe he needed to use it for something else, could be.

It's also a time where information flows that way and we need to accept it.

Now, I don't put it behind the gated content like subscribe to my newsletter or pay me.

I don't know.

I think that, you know, I've did some lifting and somebody else did some other lifting and
we should make...

grounded information circulates.

The one thing that really bothers me is people sharing things that they have reposted
without verifying and they just change the graphics a little bit on LinkedIn for vanity

metrics.

How many comments can I get?

How many?

That part is really bad.

It's really bad because I would say that 80 % of the marketing content is flooded with
this recycled, non-verified of...

non-thought-after content policy.

So I would say let's all do our job and publish our research for free.

And this is how you fight the good fight to kind of keep it human, keep it real, keep it
authentic and so forth.

I agree 100%.

And even, I would even go as far as to say that I, I think people are developing this
longing for original content again, like even faster now, because there's so much AI slob.

And so I would go as far as to say it's even turning into an opportunity to do what we
talked about for years, like be yourself, have original content.

do the thinking, all of that stuff was important for a business always.

Like that's sort of your ticket to existence, I would say, because you're adding something
to our cultural togetherness that nobody else was adding in that way.

And so now that we've been flooded in an ocean of slob, you have the perfect opportunity
to stand out because you're...

one of the few who can resist the urge to just take the shortcut, I guess.

And this is going to be the best tactic for being included in LLMs in terms of marketing
because these machines, the biggest problem that developers have now is that if an

enormous amount of content online is all produced using an AI, then all of the content it
can learn on or fetch, it's all self-produced.

And this is a...

very dangerous loop where the information growth starts to stay on and then decay very
fast.

And so the OpenAI, Google, they're all working on how to make sure that they can detect
human created content because it's fundamental.

So keep it yours, keep it human.

We all use these LLMs tools all the time to make work faster.

So it can get you started.

much quicker for sure.

It can draft things for you.

It's great for brainstorming, good idea when I'm studying.

If I didn't understand very well one block of information, I don't have a teacher who can
tell me so I'll ask it, the system, not only to explain but to put in context and to

amplify.

So if you use these tools the right way, yeah, they're going to explode, you know, your
possibilities.

with studying, with working, with all of these things.

But they're just tools.

Exactly.

And I think they're a much more, it lures you in because it makes the appearance of take,
like it can take decisions for you and stuff.

So right, like before from the typewriter to the computer, yes, it was more amazing and
could do more things.

So you still have to do it.

Now I think it's a bit of a different kind of tool just because it can decide stuff for
you.

And it's a bit.

resist.

You need to resist.

You need to resist the temptation to outsource your thinking because if you outsource your
thinking then you lose a part of yourself that is too big to quickly replace.

So I would advise against that.

Absolutely.

What a nice way to end this episode.

It's right.

Like I think I love that you, that we're in the same direction on this.

that where it really matters, please use your own thinking.

Like that's important, not only for your website, but for humanity as a whole, I would
say.

I think we can all use more.

reflection, thinking and kindness in general.

um Right.

And so, yeah, thank you for doing the research, for coming on to talk about it and for
taking a stance in all of this, I would say.

Thank you so much, Nils for the invitation.

I always love to be having a chat with you and it's been too long since our last time.

That's true.

Cool.

Well then, till the next time, Pietro and everybody out there listening, decide what you
want to do with the new tools that you have available to you.

And if you have any questions, please feel free to reach out to each of us.

I think we're more than happy to have a discussion about what's useful or not.

Awesome.

All right.

See you on the next one.

Bye.