Citation volume is an exposure signal, not a traffic KPI: Wito reported 17,400 Microsoft Copilot citations across 127 pages over three months that produced only 167 Bing clicks, about 104 citations per click. Paris Childress, recording episode 12 of The GEO Show on August 30, 2026, treats that ratio as the week's throughline and argues GEO reporting has to connect mentions and citations to branded search, assisted conversions, demos, and pipeline. The rest of the briefing covers Searchable's Prompt Universe, volatile AI crawler mix, ksmmodel.ai's three-variable framework, earned-authority PR, cyber risk, DeepCogito, generational engine preference, and Korean citation sources.
Citation volume is an exposure signal, not a traffic KPI. Wito reported 17,400 Microsoft Copilot citations across 127 pages over three months. Bing generated only 167 clicks. That is about 104 citations for every click.
Paris Childress, hosting a solo episode 12 of The GEO Show on August 30, 2026, treats that ratio as the week's argument. Mentions and citations still matter as visibility. They do not, on their own, prove traffic, demos, or pipeline.
Citation volume is not a traffic KPI because a large count of Copilot citations can sit next to almost no Bing clicks. Wito's three-month log is the proof: 17,400 citations, 127 pages, 167 clicks. Paris's line on tape is that GEO success should never be positioned as citation growth, because a massive citation volume still does not deliver traffic.
Citation volume here is clearly not correlated, or it's really not a traffic KPI.
Looking at citation volume alone is not even a good predictor of click volume. Copilot can cite a page as evidence while the searcher never leaves Bing. The citation is real. The visit is optional.
GEO reporting should connect mentions and citations to downstream branded search, assisted conversions (including AI-assisted conversions via self-reported attribution), demos, and pipeline. Paris calls that the full journey.
Searchable's US move points at the same problem from the software side. The UK AI-visibility startup opened a Salt Lake City office, hired two former Scrunch employees, and said it had grown from nearly 1,000 customers in May into the thousands (self-reported). Its planned September Prompt Universe product is about deciding which prompts actually matter. Paris calls that prompt discovery, and treats it as the harder IP in a crowded GEO category. SignalForge uses an agent that reads the brand site, competitor sites, and Ahrefs demand, then translates intent into prompts, including bottom-of-funnel comparisons such as "competitor X alternatives."
Monitor crawler mix by engine because AI crawl traffic can swing without a site change, and the swing is not the same across ChatGPT, Claude, and Google. AI Tools Recap's 36-day server logs, reported August 29, showed the ChatGPT crawl index falling from 100 to 31 while Claude rose to 242 and Google stayed near baseline. Nothing material changed on the site.
Paris's comparison is to classic SEO: a top-three ranking could be defended for years. Citation appearances are more volatile. Do not treat a sudden ChatGPT decline as a sitewide technical failure until other crawlers follow and other technical signals agree.
The same week, ksmmodel.ai launched an assessment built on structured extractability, entity salience, and citation authority. The methodology originates with its founder and has not been independently validated. Paris's test for any proprietary 0-to-100 score is whether it ties recommendations to outcomes. GEOforge's named score on tape is information gain: a 0-to-100 measure of how much a knowledge chunk uploaded into BaseForge contributes net-new knowledge to LLM crawlers.
There is no universal Reddit playbook because citation source ecosystems differ by market, language, query class, and LLM engine. RankIt AI tested 33 Korean purchase-intent responses containing 492 citations and found zero from Reddit, Naver Blog, Tistory, or NamuWiki. Roughly 80% came from vendor and official sources. A US Reddit and digital-PR playbook does not translate.
Audience composition cuts the same way. YouGov data reported by Search Engine Journal shows ChatGPT preferred in every age group, while Claude preference reaches 10.6% among Gen Z versus 1.4% among Boomers+. Copilot shows the reverse. Engine mix should follow the people buying the product.
Search Engine Land's 2027 link-building argument pulls those threads into one earned-authority program: earned stories, original data, journalist relationships, brand mentions, and placements that can influence people and LLMs. Paris says the old wall between SEO link building and PR is collapsing into GEO agencies.
Two other stories sit on risk and knowledge. More than 100 companies, including OpenAI, Anthropic, Google, Microsoft, Cloudflare, CrowdStrike, Okta, and Fortinet, signed a joint call (TechCrunch) warning that AI-enabled attacks will get more widespread. Paris puts website security inside GEO risk management, because compromised pages can inject spam, redirects, and false information that damage retrieval and trust. DeepCogito, founded by former Google AI Search leaders who led Gemini post-training, raised a $43 million Series A (more than $56 million total), with Zscaler as customer and strategic investor. Enterprise AI is moving toward post-trained models on proprietary company knowledge, which is also the GEOforge premise: do not generate public GEO content until the knowledge base is robust enough to keep the output from becoming slop.
00:30. Searchable opens in Salt Lake City, hires former Scrunch talent, and plans Prompt Universe. Paris calls prompt discovery the harder IP in GEO software.
03:04. Wito: 17,400 Microsoft Copilot citations across 127 pages, 167 Bing clicks, about 104 citations per click. Citation volume is not a traffic KPI.
04:31. AI Tools Recap, 36 days of logs: ChatGPT crawl index 100 to 31, Claude to 242, Google near baseline, no site change. Monitor crawler mix by engine.
13:12. YouGov via Search Engine Journal: ChatGPT leads every age group; Claude 10.6% among Gen Z versus 1.4% among Boomers+; Copilot stronger with older users.
14:37. RankIt AI: 33 Korean purchase-intent answers, 492 citations, zero from Reddit, Naver Blog, Tistory, or NamuWiki; roughly 80% vendor and official sources.
The throughline is the Wito ratio. Visibility without a path to branded search, assisted conversions, demos, and pipeline is a number on a dashboard, not a GEO program.
Hi, everybody. Welcome back to another edition of The GEO Show, brought to you by GEOforge, full self-driving for AI visibility. And I'm your host, Paris Childress. Let's get into today's top stories for episode 12. Recording on August 30th.
First up, Searchable is making an aggressive US move with former Scrunch talent. UK AI visibility startup Searchable opened its first US office in Salt Lake City and hired two former Scrunch employees, including Scrunch's former sales lead. Searchable says it has grown from nearly one thousand customers in May into the thousands and plans a September product called Prompt Universe to identify which prompts brands should actually monitor. Company growth numbers are self-reported.
Well, this is interesting from a couple of angles, but most notable for me is this product called Prompt Universe, because I do think that GEO software, which is already a hyper-crowded category, the fight in this category is shifting from just tracking prompts towards deciding which prompts actually matter, which is a much harder problem. And we refer to this internally as prompt discovery. And I think that this is gonna be really the secret sauce of a lot of tools going forward, which is this prompt discovery or prompt selection methodology. I think this is gonna become the critical IP for GEO software providers.
The way that we handle prompt discovery at GEOforge within SignalForge is that we have an AI agent that goes out and does a lot of research. It looks at the brand's website, it looks through its competitor sites. It also goes to Ahrefs and does independent keyword research and tries to understand what are the highest demand keywords and how that intent translates into prompts. And then, of course, it's looking at competitors because a lot of these prompts are also prompts that include things like competitor X alternatives or competitor A versus competitor B.
These are very common bottom-of-the-funnel prompt types that are typically very high intent, that have an above average click-through rate from mentions and citations, and a much higher conversion rate because it indicates people are really at the end of the journey where they're comparing known products head to head. But in any case, prompt discovery is, I think, one of the hardest things in AI visibility right now, and the companies that get this right are really gonna be the winners, I think.
On to the next story. Seventeen thousand four hundred AI citations produced just one hundred and sixty-seven clicks. So this is from Wito, who reported Microsoft Copilot cited it seventeen thousand four hundred times across one hundred and twenty-seven pages over three months, while Bing generated only one hundred and sixty-seven clicks. So that's basically one hundred and four citations per one single click.
So citation volume here is clearly not correlated, or it's really not a traffic KPI. This is an important lesson because GEO success should never really be positioned as citation growth or citation volume because a massive citation volume still does not deliver traffic. So what's important is to tie visibility, such as mentions and citations, to downstream branded search, assisted conversions, AI-assisted conversions via self-reported attribution, and then finally demos and pipeline that result from that. So really that full journey. But looking at citation volume alone is not even a good predictor of any sort of click volume.
Next up, AI crawler traffic can swing violently without anything changing on the site. This is from a single-site field study on August 29th. Across 36 days of server logs, AI Tools Recap saw its ChatGPT crawl index fall from one hundred to thirty-one, while Claude rose to two hundred and forty-two. Google remained near baseline. The site says that nothing materially changed technically during that period.
What is this telling us? AI visibility is extremely volatile. This is not at all like SEO search rankings that tend not to move around too much. In the SEO days, if you got a top three ranking, you were able to defend that and stay there in some cases for years. But AI visibility, particularly with citation appearances, is extremely more volatile. So it's important to monitor the crawler mix by engine because the volatility also is very different across ChatGPT, Claude, Google. So it's important to monitor the crawler mix by engine, and avoid interpreting a sudden ChatGPT decline as a big problem until you see that other crawlers follow suit and that you check other technical signals.
A new GEO entrant is trying to standardize the category around three variables. This new entrant is called ksmmodel.ai. They launched an assessment framework built around three variables. Number one, structured extractability. Number two, entity salience. And number three, citation authority. With metrics such as share of model and entity recognition rate, the methodology originates with its founder and has not yet been independently validated.
So what's happening here? GEO vendors are increasingly competing on framework ownership, not merely data collection. So these different frameworks are emerging, and a defensible agency framework really needs to have evidence that ties its recommendations and its philosophy to outcomes, not just having another proprietary zero to 100 score for something.
Now, that said, we have some proprietary scoring, first of which is called information gain score, which is a score from zero to 100 that measures the degree to which a piece of knowledge that is uploaded into our knowledge base, BaseForge, contributes net new knowledge to LLM crawlers. But beyond that, we are starting to develop our own framework, which is based on this whole principle of information gain. And I think the framework that makes us most unique in our space is that we insist on grounding all generated content from a knowledge base so that it delivers high information gain to AI crawlers. So this is really our framework, and I do think it's important for the leaders in this space to have a unique opinion and a unique framework that they build their software around.
Link building, digital PR, and GEO are collapsing into one earned authority motion. Search Engine Land argues that 2027 link building should move away from link volume toward earned stories, original data, journalist relationships, brand mentions, and placements that are capable of influencing both people and LLMs. So the historic organizational wall between SEO link building and PR really makes less and less sense in an AI search environment. That wall is starting to collapse, and it really makes a lot of sense here, because digital PR is one of the best ways to get high quality citations, third-party citations. And I think now it's important to really think of this all as one activity, one combined earned authority program where you're simultaneously optimizing for links, for third-party mentions, and to try to reach referral audiences and AI citations. So I do think that particularly digital PR agencies and link-building agencies are going to kind of merge together into what are now gonna be GEO agencies.
Next up, more than one hundred major companies are warning that AI-enabled cyberattacks are about to accelerate. OpenAI, Anthropic, Google, Microsoft, Cloudflare, CrowdStrike, Okta, Fortinet, and others have signed a joint call for stronger cyber defense, warning that AI-enabled attacks will become more widespread and sophisticated in coming months. This is coming from TechCrunch.
The interpretation here is that website security increasingly belongs inside of GEO risk management because compromised sites can create spam and redirects, they can poison and create false information that directly damages retrieval and trust. So for cyber clients especially, I think that a comprehensive GEO audit or on-site GEO health monitoring should include technical SEO, classic technical SEO, like crawl accessibility and information architecture, plus AI crawler monitoring, which we've talked about in previous episodes, making sure that the right bots are able to access the site, but the malicious bots are excluded. Plus website security, which is mentioned here, and brand integrity monitoring. Because if a site gets hacked, of course, that definitely is gonna impact its brand integrity, which flows downstream into how an AI might answer questions about that brand.
Next up, former Google AI Search leaders have raised forty-three million dollars with Zscaler as a customer and investor. DeepCogito, I guess it's pronounced Cogito, C-O-G-I-T-O, raised forty-three million dollars in Series A, taking its total funding above fifty-six million. Its founders previously led Gemini post-training and product work for Google AI Search. Cybersecurity company Zscaler participated as both a customer and a strategic investor.
All right, why is this interesting? Well, enterprise AI is increasingly using post-trained models on proprietary company knowledge, rather than one general public model. So proprietary knowledge bases, which is the core of GEOforge. So to take a step back, the entire premise of GEOforge is that you build content that is grounded in a proprietary knowledge base, and until you get that proprietary knowledge base built and robust enough, you don't even bother trying to generate any other type of content because it'll just be AI slop. So here, proprietary knowledge bases could become valuable, not just for public GEO, like in the case of what we're doing with GEOforge, but also as training and evaluation assets for clients' own AI systems that they will develop for themselves. I do think starting with enterprise companies, and more and more they'll be building internal AI systems, and in order to do that, it needs to start with a comprehensive knowledge base.
Next up, different generations are already choosing different AI discovery environments. YouGov data reported by Search Engine Journal shows ChatGPT remains the preferred AI tool across every age group. But Claude preference reaches ten point six percent among Gen Z versus one point four percent among Boomers plus. Copilot shows the reverse pattern, performing considerably better with older users.
So this is one of the first times we've looked at LLM preference through the lens of age. But again, you really can't separate AI visibility from audience composition, and it again brings up the importance of ICP development and persona development, and that includes age for some different brands and products. So engine prioritization, let's say for SaaS, it should reflect the personas who are buying the product. So developers may warrant having heavier Claude measurement, while corporate users, you might wanna lean more towards Copilot. But one thing is for sure that everybody is using ChatGPT.
All right, the last story of the day. Fresh Korean data reinforces that there is no universal Reddit strategy. This is a small observational study from RankIt AI, who tested thirty-three Korean market purchase intent responses containing four hundred and ninety-two citations and found zero citations from Reddit, Naver Blog, Tistory, or NamuWiki. Roughly eighty percent came from vendor and official sources.
Okay, what is this telling us? Why are we bringing up a Korean Reddit story today here? The broader point here is that citation source ecosystems can differ radically by market, by language, query class, and LLM engine. So if you are running an international GEO program, you really have to parse out each market. You need to map local sources for each country. You can't just simply translate a US Reddit/digital PR playbook because it's not gonna be localized enough.
All right. That's it for today. Thank you for tuning into episode 12, and we'll see you all in the next one.
