The GEO Show

Welcome to Episode 3 of The GEO Show, where we break down the biggest developments in Generative Engine Optimization (GEO), AI search, and the rapidly changing world of SEO.

In this episode, Paris Childress, founder of Hop AI and co-founder of GEO Forge, runs through nine stories reshaping AI search this week, including one citation collapse nobody has explained yet.

In this episode:

๐Ÿ” ChatGPT is doing a lot more research for B2B software queries Nective tested 4,000 prompts and found average ChatGPT fan-out queries rose from 2.17 to 7.61, with software queries alone averaging 10.7 searches per prompt. B2B recommendations are becoming a research workflow inside LLMs, not a single query.

๐Ÿ“Š Your GA4 numbers may be lying about AI Overviews traffic A nine-month dataset covering 51,200 tracked events found that 22.4% of detected AI Overview events were attributed to direct traffic instead of organic. In a zero-click world, don't take Search Console or GA4 at face value.

๐Ÿง  Google's own research: LLMs know facts but still can't retrieve them Gemini 3 Pro and ChatGPT-5 encoded 95-90% of tested facts correctly, but failed to directly recall 26-34% of those same facts. Training-data accuracy and real-time discoverability are two separate problems, and you need to win at both.

๐Ÿ› ๏ธ Google AI Overviews can now generate interactive interfaces, not just answers Search Engine Journal reports Google has started rolling out Generative UI, including custom layouts, simulations, and interactive tools. If your lead-gen strategy depends on a gated calculator, Google's AI answer can now reproduce it with no click required.

๐Ÿ“ฃ ChatGPT Ads landed in Europe today OpenAI is expanding into 31 European markets with conversion optimization, geo-targeting, custom audiences, Pixel, and Conversion API measurement, the full performance-marketing stack.

๐Ÿ“‰ Reddit's share of ChatGPT citations collapsed 86% in three days According to Promptwatch, Reddit's average citation share fell from 3.83% to 0.52% between August 14 and 17. Reddit hasn't ruled out a collection issue. Either way, a citation strategy concentrated in one off-site channel just got a lot riskier.

๐Ÿ“น YouTube citations are now measurable at the creator level Profound shipped a citation breakdown that identifies cited channels, individual videos, and video categories. YouTube remains the number one citation source across Google's AI properties, and it's time to measure it like owned media.

โš ๏ธ Even "stable" brand content is feeding LLMs bad facts Profound analyzed 158,000+ evaluated AI claims and found engine-level inaccuracy rates of 4.3-6.8%. Pricing and billing errors made up 24% of inaccurate claims despite being just 12% of claims evaluated, so keep your pricing current and visible.

๐Ÿ’ต Commercial intent inside ChatGPT is rising fast Profound classified a year-long sample of 7.5 million conversations and found commercial intent share rose from 13.9% to 19.2% between June 2025 and June 2026, roughly 533 million weekly commercial conversations. It's still early, and that's the opportunity.

Subscribe to The GEO Show for ongoing analysis of the news, experiments, strategies, and emerging tactics shaping visibility across ChatGPT, Google AI, Perplexity, Claude, and the rest of the AI search ecosystem.

๐Ÿ’ฌ Question: If a channel you'd built your whole citation strategy around collapsed overnight, would you even notice before it showed up in your leads?

What is The GEO Show?

All things Generative Engine Optimization (GEO). A breakdown of all that's happening in the world of AI search.

Paris: Hi everyone, and
welcome back to The GEO Show.

We are on episode three , and I'm
recording this on, uh, Monday, August

twenty-fourth, and I've got a bunch
of new headlines to run through

today, so let's get right into it.

The first is that ChatGPT
is doing dramatically more

research for software queries.

So a company called Nective tested four
thousand prompts, and they found that

average ChatGPT fan-out queries rose
from two point one seven to seven point

six one, and software averaged even more
at ten point seven searches per prompt.

And the site appeared, uh, sixty-four
percent of its sampled fan-outs.

So, uh, what is this telling us?

Well, B2B recommendations are
really becoming increasingly, uh,

research workflows within LLMs, not
just single query, query retrieval.

So We're seeing here query fanouts for
more complex B2B software questions and

prompts on the order of about five times
higher than average, from two to 10.

So that's pretty remarkable, and I
think that means that the surface

area of that retrieval, that search
retrieval process, particularly for

B2B software, is much, much greater.

And I think that actually creates more
opportunity for B2B SaaS marketers to

capture that larger surface area through
things like video repurposing or just

simply scaling out long-tail content.

All right, next story is
about Google AI Overviews.

Google AI Overviews traffic may be
materially misclassified in GA4.

It's been observed that a nine-month
data set covering fifty-one thousand

two hundred tracked events and one
thousand six hundred and sixty-one cited

snippets found that twenty-two point four
percent of detected AI Overview events

attributed to direct rather than organic.

So that is not an insignificant chunk.

Uh, twenty-two percent is pretty massive.

So again, this is a reminder to always,
uh, not-- refrain from taking, uh,

Google Search Console, Google Analytics
data at face value, particularly now

in the zero-click world that we're
living in, because what we see is

that, uh, even a big chunk of AI
Overview traffic is being misclassified

as direct rather than organic.

Um, aside from all this, uh, you all
may know that anytime someone clicks

through from an app, let's say they're
using the Chat-ChatGPT app, if they

click through a citation, that will
also show up as direct and, uh, that

will lose the ChatGPT referral string.

So, uh, this makes the attribution
very difficult, um, but, uh, we still

need to, to press on regardless.

Let's move on to the next story.

Uh, Google Research found that
LLMs can often know facts but

still cannot retrieve those facts.

So this is coming from
Google's own research.

Gemini 3 Pro and ChatGPT-5 encoded
ninety-five to ninety percent of tested

facts, but they failed to directly
recall twenty-six to thirty-four

percent of those same facts.

So rare facts and reversed relationships
were particularly difficult.

So This is interesting because I think
that one of our goals with GEO, of

course, is to give facts about your
brand, accurate facts about your brand

to the LLMs, and also position your
pages and your claims, passages, and all

of your content for optimal retrieval.

Because as we know, sometimes the LLMs
will produce an instant answer from

pre-training data, and other times
they will do a real-time search, and we

need to be in both of those channels.

Um, but what is interesting about this
study is that sometimes you can train

LLMs on the correct facts about your
brand, but if the LLMs are going to do a

real-time search and, uh, what's called
a RAG, a retrieval-augmented generation,

uh, they could still get it wrong.

So I think the goal here is still
to, to do both, um, and, and because

that RAG system does rely on search
rankings, that does mean that you

still need to be getting your pages
indexed for SEO and be discoverable when

Google does a real-time search or when
ChatGPT maybe does a real-time search

Let's move on to the next story now
Google AI Overviews can now generate

interactive interfaces, not just answers.

This has come to us from
Search Engine Journal.

Google has started rolling out
Generative UI features into AI Overviews,

including custom layouts, simulations,
and interact- interactive tools.

So these are things like interactive
charts and graphs, calculators.

Uh, this is already global in English
in AI mode with broader AI Overview

availability expected over the coming
weeks as per Search Engine Journal.

So what does this mean?

Well, a lot of us are building these
interactive tools and interactive

widgets and elements into landing
pages and into lead capture,

particularly lead capture assets.

So these are things like calculators,
comparison tools, um, other

interactive widgets that are often
now, uh, one of the things that

gate a U-- uh, an MQL acquisition.

So in order to see the results,
let's say, of a, of a-- some sort

of a calculator, you need to, uh,
provide your name and email address.

So a lot of these conversion, uh, these
conversion-optimized landing pages

driven by these interactive tools are
now really becoming or are facing threat

from the zero, zero-click threat because
Google's own AI mode and AI Overviews

can reproduce those same tools and
widgets right inside of its answer.

So that means that we're all gonna
collectively have to take our landing page

game up another notch and think of other
ways to, to try to capture a, a lead when

Google can reproduce that same interactive
co- content inside of its own answer.

All right, next story is about ChatGPT,
and this is pretty big news because now

ChatGPT Ads are set to arrive in Europe
on August twenty-fourth, which is today.

So OpenAI is expanding into
thirty-one European markets, initially

through its ad solutions team,
agencies, and technology partners.

Self-service access will follow later.

So OpenAI says that the
platform now supports conversion

optimization, GEO targetingรขย€ยฆ

Uh, sorry, that is geo-targeting,
not GEO in, in the sense of the

GEO Show, but geo-targeting.

Custom audiences, Pixel, and
conversion API measurement.

So that is the full boat of
features from ChatGPT Ads.

So this is really exciting news for
any of us who, like myself, are based

in the EU And have been anxiously
awaiting the rollout of OpenAI's

ChatGPT ads for European targeting.

Our agency, Hop AI, have been using
ChatGPT ads and running them for clients

for probably the last couple of months,
and that targeting has been limited so

far primarily to the US market, but we are
really eager to unlock that for Europe.

And starting from today, it has arrived.

Great news.

All right.

The number six story is probably the
biggest one on the docket today, and

it's one that has been circulating
very heavily over the last few days.

Reddit.

Reddit's share of ChatGPT citations
has reportedly collapsed eighty-six

percent just in the last several days.

According to Promptwatch, Reddit
fell from a three point eight three

average citation share to zero point
five two percent between August

fourteenth and August seventeenth across
millions of monitored AI responses.

And the company, Reddit, says
that the cause is unknown and has

not ruled out a collection issue.

So the interpretation here is that, uh,
this is still very, very early in the GEO

game, and a lot of optimizers have been
really focusing their off-site efforts on

Reddit threads and Reddit optimization.

And what we're seeing, at least for
the time being, is that Reddit as

a citation source has effectively
collapsed and shrunk dramatically.

And I, I don't think any of us saw
this coming, and I think the, the key

takeaway here is the importance of
diversifying your citation building,

your off-site citation building efforts,
not focusing only on Reddit or even

only on, uh, UGC forums, but focusing
on other, other things like, uh, the,

uh, blogs, editorial sites, um, review
sites if you're, uh, if you have a

product or a software, review sites are
really important, and other types of PR

outreach where you can get featured, uh,
around the web in trustworthy places.

Not to mention also content
repurposing, starting with video,

'cause as we've seen, YouTube is
still a major, major citation source.

Uh, in fact, it's the number one
citation source throughout Google's AI

properties, and that hasn't changed.

So, uh, everyone is buzzing about the
Reddit citation share collapse, and we'll

keep an eye on this to see if it starts
to recover or if it stays as low as it

has been over the last ten days or so.

On to the next story,
which is about Profound.

Uh, Profound has made YouTube citations
measurable at the creator level.

They've confirmed that their new
YouTube citation breakdown identifies

cited channels, individual videos,
and video categories, including each

channel's share of YouTube citations.

All right, this kind of follows on
what we had just talked about with the

collapse of Reddit and the stability
of YouTube as a citation source.

Uh, I think that this is very, very
important now to start measuring activity

in YouTube for share of voice metrics
in the same way that you measure a

website, because this is owned media.

Uh, your owned media consists of
your website, but it also consists

of all of your social media channels.

So if you- your YouTube channel or,
or YouTube video within your YouTube

channel shows up as a citation and people
watch that video, then that arguably

has just as much of a positive impact on
your brand favorability than if someone

had seen a link to your website and
clicked through and read a blog post.

They're effectively still
consuming your brand's content.

So I think this is a really important
and a really good move by Profound to

bring in YouTube citation measurement in
at the creator level, and it seems that

they're doing this not only for the brand
channels themselves, but also, uh, at the

category level and at the creator level.

So they are clearly putting a big
emphasis on YouTube, and I do think

this is a strong signal for everyone
looking to up their GEO game.

I think that video, particularly
YouTube, is one of the most important

tactics within that strategy.

All right.

The next story is also coming from
Profound, and that is an AI factual

errors are often coming from stable
brand and earned media content.

So let's unpack this.

Profound has analyzed 158,000 plus
evaluated AI claims and found that

engine-level inaccuracy rates of 4.3

to 6.8%

persist.

And that is not an
insigni-insignificant number.

Um, pricing and billing inaccuracies
represent 24% of those inaccurate

claims, despite being just
12% of the evaluated claims.

51% of brands had at least one inaccurate
claim tied to earned media, while 54%

had one tied to their own content.

And this is research coming from Profound,
and that is quite profound, I must say,

that th-these inaccuracies persist.

Uh, and they persist at a couple
of levels with owned media It's

showing that, uh, let's see.

Yeah, fifty-four percent of the
inaccuracies were tied to a brand's

own content, and another fifty-one
percent of those inaccuracies were

tied to earned media, so that's
your off-site media mentioning you.

Uh, but, uh, that's, you know, this
is kind of troubling, and I think

that pricing and billing, uh, because
pricing tends to, uh, m-maybe if-- In

particular, a lot of companies don't
have pricing pages that display prices,

or if they do, sometimes they still
hide the pricing and try to get you to

click on the button to talk to sales.

And what this does is it leads to the
third party, uh, or at least it leads

to the off-site ecosystem sharing
your pricing through other places.

And if you have a sales team that
negotiates, uh, bespoke pricing

for each customer, which is pretty
common, especially with, let's say,

enterprise-level SaaS, then you could have
a lot of pricing inaccuracies showing up.

So I think that I typically have
always believed that it's a good idea,

in particular, if you're doing any
kind of PLG motion with self-serve,

to always have very clear and
transparent pricing on your website

and to keep that up to, up to date.

In fact, even now, to have it
displayed on the homepage itself so

that it's even easier to access and to
update regularly by AI bot crawlers.

So, uh, stale pricing, packaging,
integrations, certification,

product capabilities, these things
all need continuous monitoring,

both on your website and off your
website, to avoid these inaccuracies.

Because a simple inaccurate
claim by an LLM could actually

cost you a lead All right.

Now we will finish up with the
last story of the day, which is

that commercial intent inside
of ChatGPT is rising rapidly.

And I think that's good news for everyone
in marketing 'cause we're always trying

to hunt down that commercial intent.

So, uh, Profound, again, uh,
third time Profound is cited

here as a research source.

Profound classified a year-long
sample of seven point five million

conversations and says that commercial
intent share rose from thirteen

point nine percent to nineteen point
two percent between June twenty

twenty-five and June twenty twenty-six.

So about five percentage points,
um, which is, I'd say, pretty

significant, and it is still rising.

Uh, this is extrapolated-- This is an
extrapolated estimate, and it puts weekly

commercial conversions at roughly five
hundred and thirty-thir-- five hundred and

thirty-three million versus two hundred
and forty-three million a year ago.

So that is all good news for
everyone who is already optimizing

for ChatGPT as an intent channel
because commercial intent is there.

We've long suspected it, and now
we have some real hard evidence

that it is there, and it is rising.

So your efforts to invest in ChatGPT
as a performance channel to tap

into commercial intent is likely
still a very, very good idea, and I

still believe it's very, very early.

All right.

That's a wrap for today.

Thank you for joining us for episode
four, and we'll see you next time.