Dedicated GEO tools average five cited AI answers each, while established SEO platforms average 131, according to an SEO Rocket study of 6,322 citations across 56 tools measured with Ahrefs Brand Radar. In episode six of The GEO Show, Paris Childress explains why entering the GEO category does not make a brand part of an AI model's category model, and covers ChatGPT's memory-personalized query fan-out, Semrush data on topical depth, and the first public university GEO RFP.
The GEO tool category cannot get itself cited by the engines it exists to measure. SEO Rocket looked at 6,322 citations across 56 SEO and GEO tools using Ahrefs Brand Radar and found established SEO platforms averaging 131 cited AI answers each, while dedicated GEO tools averaged five. The incumbents earned that position honestly, with years of published research and documented methodology behind them. But the gap is the story: a category built on selling AI visibility is largely absent from AI answers about itself.
Paris Childress spends episode six of The GEO Show on that finding and five others, and one thread runs through all of them: category membership is not something a brand declares, it is something a model has to have learned.
Because entering a category is not the same as being part of a model's representation of it. Every tool in the space, GEOforge included, has to demonstrate its own category association before an AI answer will name it.
Merely entering the GEO category does not make a brand part of AI's category model. Every GEO tool that's out there, including our own, GEOforge, really has to demonstrate its own category association and AI visibility to prove why it belongs in the category.
Paris reads the numbers as a structural difference from the SEO tools market. SEO tools were hard to build a decade ago, so that market coalesced around Semrush and Ahrefs, with Moz taking meaningful share. GEO tools are easy to build, so there are now hundreds of them, most focused on AI visibility measurement alone. The category is barely a year to eighteen months old and new entrants keep arriving. His read on who wins: the best marketed and best funded tools, with quality necessary but not sufficient.
One caveat on the study: it is vendor-produced, US only, caps high performers at 100 citations per engine, and Perplexity alone generated more than a third of the sampled citations. Directional, not definitive.
It retires the single canonical prompt as a unit of measurement. OpenAI's updated search documentation confirms ChatGPT may rewrite one prompt into multiple searches, a pattern known as query fan-out, and may use saved memory when constructing those queries.
Personalization now happens before retrieval, not after. Two people typing the same question can send the model down very different retrieval paths before it looks at a single source. Paris is blunt that this makes measurement harder. The practical response is to widen what you test:
Deep. Semrush analyzed 283,215 citation observations across 1,094 ChatGPT categories and found brands were both cited and named in 34% of appearances in closely related categories, against 9% in distant ones. Depth inside a category correlated more strongly with brand mentions than spreading thinly across many.
For a brand competing in several verticals, that means dominating two or three strategic categories before expanding horizontally. Ahrefs' August 20 Deep Dive points the same way on content: it has to contribute something beyond what already exists, whether original data, first-hand experience or a genuinely new perspective. Google has not confirmed its patented information gain score as a live ranking factor, and Paris says so plainly, but his position does not rest on the patent. Synthesized answers will fail over time either way. It is why GEOforge grounds content production in a proprietary knowledge base, BaseForge.
Far more than prompt tracking. South Dakota State University's LLM search optimization RFP went public with an anticipated award date of August 24, an annual ceiling of $80,000, a two-year initial term and roughly 9,000 pages in scope. Its requirements list SEO as the foundation, plus AI answer monitoring, Drupal integration, audit logging, US data residency, and SOC 2 Type II or ISO 27001 credentials.
Paris calls this GEO moving from experimental marketing spend into governed enterprise procurement, which pulls security, permissions, integrations and reporting into scope. He expects more RFPs of this shape from enterprises and large institutions.
It is no longer a yes or no decision. Cloudflare's AI Crawl Control now offers crawler-level monitoring, allow and block policies, robots.txt compliance tracking, and a private-beta pay-per-crawl capability, with the caveat that WAF rules can override crawler policies. There are dozens of AI crawl bots to make decisions about.
Not all of them are what they claim. On one client, Paris's team found a bot posing as OpenAI's user agent that turned out to be spoofing and phishing traffic, which forced them into Cloudflare logs to whitelist and blacklist specific bots by IP address. Bad actors impersonate a legitimate crawler, deep crawl the site and hunt for exploitable vulnerabilities. What follows is ongoing log analysis: which bots crawl, how often, and what needs blocking at the IP level.
The parts of GEO that can be asserted are getting cheap. The parts that have to be earned are getting measurable. Being in the category is an assertion; being cited in it is not.
Hi, everybody. Welcome back to The GEO Show, episode six. I'm your host, Paris Childress, and this episode is brought to you by GEOforge, full self-driving for AI visibility. Let's get into our top stories for today.
First up, ChatGPT search is explicitly personalized before retrieval. This came recently from OpenAI's Help Center, and OpenAI's freshly updated search documentation says that ChatGPT may rewrite one prompt into multiple searches, which is something called query fan-out, which we've talked about before, and can use saved memory when constructing those queries. And that is the really important point. Not only does it do the query fan-out, but it is using saved memory of that particular user's conversation history to personalize those queries.
So that means that two people asking the same question can trigger very different retrieval paths before the model even sees the sources. So query fan-out is now becoming personalized based on memory. So that is making things even harder for us marketers. But it does mean that GEO testing should include buyer personas and contextual variants as much as possible, not just simply one canonical prompt or even one set of canonical fan-out queries.
Next up, information gain is becoming the clearest antidote to AI content sameness. This is coming to us from Ahrefs, whose August 20th Deep Dive argues that useful content needs to contribute information beyond what already exists: original data, first-hand experience, or genuinely new perspectives. It correctly notes that Google has not confirmed its patented information gain score as a live ranking factor.
So whether or not there is a literal patented information gain score that's coming, synthesized answers really are going to be failing over time. And this is a very, very key tenet of our whole philosophy at GEOforge.
We insist on all content production being grounded in a proprietary knowledge base, which is called BaseForge, so that the content that gets produced has high information gain, that it's not derivative AI slop, and that it does bring out the unique original data, first-hand experiences, and new perspectives or brand proprietary knowledge so that it can train the AI models with that new knowledge. And in exchange, it can win citations for that information gain contribution.
Moving along. Topical depth appears more important than broad AI visibility for getting the brand actually named. Semrush analyzed 283,215 citation observations across 1,094 ChatGPT categories. What they found was that in closely related categories, brands were both cited and named in 34% of appearances, versus just 9% in distant categories. So depth inside a category correlated more strongly with brand mentions than spreading thinly across many categories.
So the takeaway here is depth over breadth when it comes to GEO. If you have a brand that competes across several verticals, it is much more important to go deeper and try to dominate two or three of those strategic categories before trying to expand too quickly horizontally into adjacent topics or adjacent categories.
All right. Next up, GEO has officially entered institutional procurement. Now, what do we mean by that? South Dakota State University's LLM search optimization RFP, which has recently gone public, has an anticipated award and negotiation date of August 24th, an annual ceiling of eighty thousand dollars and a two-year initial term. It's roughly nine thousand pages in scope. The requirements for this RFP include SEO as the foundation, plus AI answer monitoring, Drupal integration, audit logging, US data residency, and SOC 2 Type II or ISO 27001 credentials.
So that's quite interesting because this is a major RFP from a major US institution, South Dakota State University. And it's really showing that GEO is moving from experimental marketing spend into a governed enterprise procurement. And that governed enterprise procurement needs additional things like security, governance, permissions, integrations and reporting. So it goes well beyond just prompt tracking.
So we do expect in the near future to see more of these types of GEO RFPs that are coming out from enterprises and large institutions that will additionally require these types of security and governance assurances.
All right, moving on to the next story. The GEO tool category has an AI visibility problem of its own. SEO Rocket measured 6,322 citations across 56 SEO and GEO tools using Ahrefs Brand Radar. Established SEO platforms averaged 131 cited AI answers, versus five answers for dedicated GEO tools. Perplexity generated more than one-third of the sampled citations. And the study is vendor-produced, US only, and caps some high performers at one hundred citations per engine.
So what this is telling me is that the GEO tools category is already much more fragmented and way more competitive than the SEO tools category ever was, which makes perfect sense to me because SEO tools were not that easy to build, let's say, ten years ago. So really that whole market coalesced around really two top players, which were Semrush and Ahrefs. And then there were some others like Moz that picked up some significant market share as well.
But what we see today is literally hundreds of most likely vibe-coded GEO tools, most of them focused on AI visibility measurement only. And this is really causing an AI visibility problem. So merely entering the GEO category does not make a brand part of AI's category model. Every GEO tool that's out there, including our own, GEOforge, really has to demonstrate its own category association and AI visibility to prove why it belongs in the category.
And this is quite interesting because the category of GEO tools is still so young. It's probably barely a year old at this point, maybe a year and a half, and there's still new entrants that are just pouring into this category. So it is going to be mostly about marketing. The best marketed tools, and of course the best quality tools, but the ones that are marketed the best are going to be the ones rising to the top, probably as well as the ones that are getting the most funding.
All right, our last story of the day. AI crawler access is becoming a governance and monetization decision. So let's unpack this. Cloudflare's AI Crawl Control now provides crawler-level monitoring. It allows for allow and block policies, robots.txt compliance tracking, and a private beta pay-per-crawl capability across its platform. Cloudflare notes that WAF rules can override crawler policies.
So the whole crawl game, the AI bot crawl management game, is getting more and more complex. Cloudflare is clearly the leader here. And really this question of should AI bots be allowed to crawl our website, it's no longer just a simple yes or no answer. It's becoming much more nuanced. There are literally dozens of AI crawl bots now.
We've also seen with one particular client that there are spoof or fake AI crawl bots that are named after the popular ones like OpenAI's user bot. We saw one that was posing as OpenAI's user bot, but was in fact a spoofing and a phishing bot. So we have had to now go into Cloudflare, analyzing those logs, and really think about whitelisting and/or blacklisting specific AI bots based on their IP addresses.
So this is becoming a real cat and mouse game. I think now there are some bad actors who can pose as a legitimate AI crawl bot in order to come into your website, do a deep crawl, and hunt around for vulnerabilities that it might be able to exploit.
So brands now have to be very, very vigilant and really analyze their crawl logs, seeing which AI bots are crawling, how frequently, and does anything look suspicious, and would anything need to be blocked at the IP level. So this is now becoming a really serious issue.
All right. That's all I have for today. Hope you've enjoyed this episode, and we will see you all in the next one. So long.
