Subscribe
Share
Share
Embed
All things Generative Engine Optimization (GEO). A breakdown of all that's happening in the world of AI search.
Hi, everybody.
Welcome back to another episode of The
GEO Show, brought to you by GEO Forge,
full self-driving for AI visibility.
I'm your host, Paris Childress, and let's
get into today's stories for episode 13.
Let's start with research
from Brand Ghost.
Different AI engines are really
reading different interests.
Sorry, different internets, not interests.
Brand Ghost's August 28th run
analyzed seven thousand five
hundred and fourteen citations.
Cross-engine source overlap across
ChatGPT, Claude, Gemini, and
Perplexity was only thirteen point
eight percent Four-- forty-six
percent of comparable recommendation
sets shared no brands at all.
What is this telling us?
AI visibility as a single
blended metric really hides huge
differences across the LLM engines.
So the, the correct move here is
to build engine-specific source and
recommendation strategies, especially
for, let's say, high-value SaaS
categories, if that's what you're in.
Uh, what we have seen clearly is that
ChatGPT historically has favored sites
that it has content partnerships with.
The first of which is Reddit, which
recently did see a large decline.
Uh, but there are statistics that it
heavily favors its content partners.
And then, of course, Gemini, Google
AI Overviews, and Google AI Mode, uh,
particularly AI Mode, strongly favors
YouTube because it's its own property.
So you need to always be cognizant
of, uh, which particular LLM engine
that you're optimizing for and look at
visibility across each engine separately.
All right, next up.
New academic work introduces
competitor-aware GEO.
An August twenty-seventh preprint
argues that GEO tactics cannot be
evaluated in isolation because their
effectiveness changes as competitors
adopt similar optimizations.
Its system outperformed existing
heuristics in controlled benchmarks.
So of course, this is a real
cat and mouse game right now.
Once everyone starts writing
citation-friendly pages,
that advantage erodes.
So as practitioners here in GEO,
we always have to test relative
differentiation versus the competitors,
um, not just simply compliance
with generic GEO checklists.
All right.
Next up.
Being recommended and getting the
citation are different battles.
Sheryl Commerce tested sixty
buying categories across Google
AI Mode, ChatGPT, and Perplexity.
Of the one thousand eight hundred and
fifty-one cited sources, only two point
eight percent were brand-owned pages.
Even when the brand was recommended,
its own site received the citation
only thirty-one percent of the time.
So again, we're seeing more evidence
First of all, that most of the
citation sources come from third-party
sites, not from brand-owned pages.
So that really implies that the
earned media strategy should really
be much further, uh, out in front
than the owned media strategy.
Uh, and also that even when the brand
page is cited, only thirty-one percent
of the time is that brand actually
mentioned in the answer, which I think is
the real win, is to get AI to recommend
you and mention you in its answer.
So third-party authority often
supplies the evidence behind the
recommendation, but it's really
important that we measure brand
recommendation and citation ownership
separately, and we invest in, uh, review,
editorial, and analyst ecosystems.
So that third-party outreach,
that third-party PR is probably
one of the most important things
we all should be doing in GEO.
All right, next story.
LinkedIn remains unusually important
for B2B AI discovery from Organic KPI.
In August twenty-fifth to twenty-eighth,
Recheck found LinkedIn appearing
in twenty-three of one hundred
professional Google AI mode answers
versus just four for ChatGPT answers.
Earlier, Semrush research across
three hundred and twenty-five thousand
prompts found LinkedIn in eleven
percent of AI responses overall
Quite impressive.
So LinkedIn is increasingly an
earned knowledge surface, um,
not just merely a social channel.
And one of the things that we are
doing quite heavily with GEOforge
is we are repurposing all of our
written content, including this
podcast, into LinkedIn articles as
part of the GEOforge daily newsletter.
And part of the reason that we're
doing that, of course, is the
audience distribution, because
LinkedIn newsletters actually get
great access to email inboxes.
But each of those, um, each of those
episode-- they're, well, not episodes,
but every newsletter, uh, article stands
alone as its own potential citation.
And, uh, that's why we have
seen s-some success with that.
So I think it's important, uh, also
to look at your subject matter expert
posts, your executive posts, and
really try to, to, um, structure that
into factual and quotable B2B content
Next story.
Podcasts just became much more
machine, machine-readable.
On August 25th, reported by
TechCrunch Particle has launched
something called Radar, indexing and
transcribing more than one hundred
and thirty thousand podcasts and about
twenty thousand new episodes daily.
Content can be queried through
an API and MCP by AI agents.
So podcast appearances can
now increasingly become more
retrievable evidence for AI systems.
This is great news.
I actually went into the platform
this morning and tested it, and
it is very, very impressive.
You can actually search for someone's
name, and it will pull up all of
their recent podcast appearances.
You can search for keywords, and
you can find all of the instances
where that keyword was mentioned
anywhere in the script of a podcast.
So it's very, very powerful.
I presume this is gonna be a very-- a
fantastic source for AI crawl bots to
come and index podcast content and making
all that knowledge much more accessible.
So this is a major deal, I think,
probably the most important story
of the day for this episode.
Uh, podcast appearances can
increasingly become, become
retrievable evidence for AI systems.
So we should be treating now podcasting
as part of our GEO strategy with
strong expert statements, proprietary
numbers, and clear category
positioning, uh, because now they
have a longer machine-searchable life.
And in fact, that is what I
am attempting to do with this
very podcast, The GEO Show.
I am reading through some stories.
I'm trying to add my own editorial, uh,
value on top, my proprietary knowledge,
and I am hoping that everything that I
speak here will be discoverable by AI.
And, uh, one of the things that we're
doing is publishing show notes and
transcripts on our website after every
single episode, specifically for that
purpose, so that AI bots can quickly
access the content of each of those
episodes, including the full transcripts.
Let's move to our next story.
Search infrastructure is
optimizing itself for agents.
On August twenty-seventh, SERP API added
markdown output across one hundred plus
search APIs, claiming roughly fifty
percent lower token usage on average
and up to ninety percent on some APIs
versus equivalent structured data.
All right, so let's
unpack this a little bit.
It's, it's pretty technical,
uh, markdown output.
But really what this is telling us
is that search data is increasingly
being packaged for machines rather than
humans And, uh, really developer-focused
SaaS clients should think beyond
browser user experience and more
along the lines of clean, compact,
machine-consumable documentation and data.
And I do think that the web as we
know it is starting to split, really.
There is the, the web for humans,
and that, that is a, a web, a
web where UX does matter and
the, uh, the interface matters.
But there's also now increasingly a web
for AI agents, and there what's more
important is not a slick user experience,
but rather the efficiency of data transfer
to those agents and, uh, making their
ability to consume that data and that
information as efficient as possible
All right.
The next story is traditional SEO
suites keep absorbing GEO monitoring.
Scrapy introduced its AI visibility
beta on August 26th, then added
citation exports, source inspection,
country language prompts, and prompt
suggestions through August 29th.
So again, we are seeing evidence, and
this is a repeati- repeating theme on
this show, that a standalone AI visibility
dashboard as a single functionality
is becoming a commodity feature
inside broader SEO and GEO platforms.
So what is the a- added value layer for
these tools and also for agencies that
are offering GEO services is charging for
diagnosis and execution, but not just for
access to another share of voice chart.
And this is something that I think the
be- the best platforms are moving into
really quickly is the strategic analysis,
the diagnosis, and the execution part.
So with GEOForge, our core modules
are around generating content, high
information gain content that is grounded
in proprietary brand knowledge, and also
conducting targeted outreach for citation
building, because as we've seen, citations
are probably the most influential
factor in getting brand visibility for
GEO Next story is about Perplexity.
Perplexity, a Brazilian newspaper
called Folha de São Paulo, is suing
Perplexity, alleging unauthorized
scraping, copyright infringement,
and circumvention of its paywall,
including subscriber-only material.
It is seeking damages and an injunction.
The allegations have not
yet been adjudicated.
All right.
So AI engines, uh, have been using--
AI engines' usable source universe is
partly a legal and licensing question.
I also have seen that-- I, I read
yesterday that now, uh, many of the major
music labels have gotten together to sue
Anthropic for copyright infringement.
Um, so these issues are really gonna
come to a head at some point, and it's
important that we as GEO practitioners
are just aware of this, and we always
are cognizant that we need to diversify
citation strategies and, um, not just
assume that every publisher will remain
equally retrievable by every model.
All right.
The last story of the day is from
Ahrefs, and it is about one of
our favorite topics, AI slop.
A-Ahrefs published a useful
operating model for avoiding AI slop.
Ahrefs says every article it publishes
now uses AI, but its workflow requires
human decisions before drafting evidence
gates, proprietary source of truth
material, and named editorial ownership.
Its AI pipeline could create articles
in six to twelve minutes, but
Ahrefs explicitly rejects scaling
output simply because it can.
All right.
Uh, I think Ahrefs has been one
of the strongest and most vocal
proponents for AI-generated content.
They have consistently produced evidence
that, uh, AI-generated content does not
underperform in search results or in GEO.
So, uh, what they're saying here
is that the, the advantage really
with leaning on AI workflows is
not to produce simply more content.
Uh, it's shifting from producing
more content to producing more
evidence per piece of content.
That means using automation to expand
research, expand access to data,
tools, and subject matter expert
extraction, not just simply word count.
And this is a very important point.
And here I, I wanna describe
our RAG pipeline with GEOforge.
A RAG pipeline is-- it stands for
Retrieval-Augmented Generation.
And what we do, and this is only
enabled, of course, by AI, is that we,
we build with each customer brand, we
build a knowledge base of proprietary
knowledge, and, uh, that is actually
stored in what's called a vector database.
So that knowledge gets, gets chunked.
Each chunk gets scored for
information gain, and those chunks
get embedded in what are called
embeddings in the vector database.
And then that allows the content
engine, which is the AI content
engine, which is producing the
content, to retrieve the, these
knowledge chunks and ground its content
generation in that vector database.
And so this is an example of what
Ahrefs is talking about here.
Uh, we are u- we are leveraging AI to,
uh, produce content that has, uh, deeper
research, a, a stronger connection to a
brand's proprietary knowledge, its tools,
its subject matter expert, uh, interviews,
and sales call transcripts, et cetera.
So we are not just using AI to, to
churn out huge volumes of content,
but we're trying to churn out better
content with higher information gain.
All right.
That's it for today's stories.
Thank you all for listening,
and see you on the next episode.