Citation persistence, not one-time coverage of Google AI answers, is the visibility metric Organic KPI broke open by rerunning 160 commercial queries six weeks apart. Coverage rose from 72.5% to 80.6% while total citation slots fell 12.5% and median source retention landed at 40%. A one-time citation win is not a ranking, and this weekend edition of The GEO Show treats that volatility as the week's throughline.
Citation persistence, not one-time coverage of Google AI answers, is the visibility metric that broke this week. Organic KPI reran the same 160 commercial queries six weeks apart. Coverage of Google AI answers rose from 72.5% to 80.6%, including a 20-query B2B SaaS sample that hit 100%. Total citation slots still fell 12.5%, and median source retention landed at 40%.
Paris Childress, recording the first weekend edition of The GEO Show on Sunday 30 August 2026, treats that rerun as the opposite of a ranking win. A brand can appear in an AI answer on Monday and vanish by the next measurement. Celebrating the first screenshot is the wrong ritual. The work is whether the citation is still there.
Median source retention is the share of previously cited domains that still appear when the same query is run again. After six weeks, Organic KPI found that figure at 40%. Most of the sources that occupied a citation slot in the first pass were gone in the second.
That is the collapse hiding under a coverage increase. Eight out of ten commercial queries in the set now trigger a Google AI answer, and every query in the B2B SaaS sample did. The inventory of slots those answers point to still contracted 12.5%. More answers, fewer seats, higher churn.
The lesson here is never to celebrate winning a citation once, because tomorrow you could be gone, and this is nothing like securing a number one ranking in Google in the SEO days.
Expanding coverage does not create more room for brands. It creates more answers that still have a finite list of sources. Organic KPI's 160-query panel shows the two series moving in opposite directions: presence of Google AI answers up from 72.5% to 80.6%, total citation slots down 12.5%.
Search Engine Land separately reported that Google is automatically expanding some AI answers, skipping the Show More step when its systems predict the expanded answer will be useful. Organic results move further down the page. Traditional rank, already a weak visibility metric inside an AI-answer interface, gets weaker as those answers auto-expand.
An arXiv pre-registered field experiment with 1,100 participants found that removing Google AI answers and Google AI Mode increased publisher click-through rates. An AI Mode-only experience cut publisher traffic and lowered reported user experience and trust. Clicks and visibility can move in opposite directions. Coverage going up is not a traffic forecast.
A classic number-one ranking was a relatively stable asset. An AI citation is a slot in a generated answer that can turn over without a ranking report ever flickering. Paris's line on the show is that this is nothing like securing a number-one ranking in the SEO days, and that teams need to measure citation persistence over time.
That is also why a one-day GEO screenshot is a vanity number. The 40% retention figure is the one that should sit on the report next to coverage. If coverage is 80% and four out of ten sources do not come back, the program is leaking the thing it claims to have won.
Citation persistence is a longitudinal measurement on a fixed query set, not a one-off audit. Organic KPI's method is the template: the same 160 commercial queries, six weeks apart, with coverage, slot count, and source retention reported together.
Paris's measurement stack on the show is still a "How did you hear about us?" field, sales-call confirmation, and brand-search lift. GEO looks commercially insignificant in GA4 or HubSpot source fields if those questions are missing.
00:00. Search Engine Land: Google is automatically expanding some AI answers, dropping the Show More step and pushing organic results further down. Rank as a visibility metric keeps weakening.
03:00. Omniscient Digital: first-touch analysis missed 85% of AI-influenced leads. Of 189 leads who said AI influenced them, 28 (15%) were classified as AI referrals in the CRM. 87% of AI citations lived on third-party pages.
07:00. Semrush launched an official Claude connector on 26 August, exposing 28.8 billion keywords, 43 trillion backlinks, and 317 million LLM prompts. Ahrefs already had a connector. SEO dashboards are becoming data layers behind agents.
10:00. An arXiv field experiment with 1,100 participants found that removing Google AI answers and Google AI Mode increased publisher CTR, while an AI Mode-only experience reduced traffic and reported trust.
11:00. BCG (26 August): nearly one-third of surveyed customers use AI in the purchase journey, and AI introduced users to brands they would not have considered in roughly 63% of AI-assisted journeys.
The rest of the rundown covers Salesforce and Anthropic's CloudForce (37 pre-built sales skills, open beta expected in September), Press Ranger and Otterly matching 129.3 million citations against 91 publisher licenses, and Cairrot 2.0 adding more than 20 GEO and AI-search capabilities on 28 August. The throughline stays the Organic KPI rerun: coverage expanding, slots shrinking, persistence at 40%.
Hi, everybody, and welcome to the first weekend edition of The GEO Show. We are recording on Sunday, August 30th, and we are now on episode 11. This show is brought to you by GEOforge, full self-driving for AI search visibility. And I am your host, Paris Childress, the co-founder of GEOforge. We've got some great stories to get into this weekend, so let's dive right into it.
First up, this is from Search Engine Land. Google is automatically expanding some AI answers. Google says that some AI answers now expand automatically into full answers, removing the Show More step and pushing organic results further down. Google says this happens when its systems predict that the expanded answer will be useful.
Okay. So Google's now effectively moving more of the standard search queries into more of an AI Mode style interface. And of course, this is pushing those ten blue links even further down the page and the march towards AI Mode continues. So the traditional rank alone is becoming an increasingly weak visibility metric.
What we really need to be doing is tracking whether our clients or brands are cited, mentioned, recommended, and clicked inside of the AI layer itself. Next story, also about Google Search AI answers, comes from Organic KPI. Organic KPI reran the same one hundred and sixty commercial queries six weeks apart.
And AI-answer coverage rose from seventy-two point five percent to eighty point six percent during those six weeks, while total citation slots fell twelve point five percent. In its twenty-query B2B SaaS sample, AI-answer coverage reached one hundred percent.
Median source retention after six weeks was just forty percent. So there's a lot going on here. But what we are seeing is that AI-answer coverage for B2B SaaS queries continues to rise. Now it's, according to this study, at eighty percent, eight out of ten queries, and in some cases, coverage rose to one hundred percent.
But also what we're seeing is a lot of variability and a high degree of volatility among which sources are retained, which citation sources are retained. So AI-answer presence is expanding while citation competition is becoming even tighter and more volatile. The lesson here is never to celebrate winning a citation once, because tomorrow you could be gone, and this is nothing like securing a number one ranking in Google in the SEO days.
You really need to measure citation persistence over time. Next up, first touch analysis may miss eighty-five percent of AI-influenced leads. This is coming from Omniscient Digital, who compared CRM attribution with self-reported acquisition data. Of one hundred and eighty-nine leads who explicitly said that AI influenced them, only 28 of those, which is 15%, were properly classified as AI referrals in the CRM.
Most appeared as organic or direct. It also found that 87% of its AI citations mentioning the company lived on third-party pages. There's two really interesting points that jump out at me. First is that there is major misclassification of attribution source referrals in CRMs when it comes to AI-influenced lead gen, and also that of course, the vast majority of the citations are coming from third-party sites that mention the brand as opposed to the brand's owned channels, websites, and channels themselves.
So GEO may look commercially insignificant if you're relying right now on just GA4 or HubSpot or your typical CRM source fields unless you really make the effort to do it the right way, which is adding a "How did you hear about us?" question on your lead capture forms, also confirming that in sales call analysis and really measuring the brand search lift for GEO measurement.
I think this is still the only and the best way to capture the true ROI of an investment in AI search and GEO. All right. The next story also has to do with CRMs and the biggest CRM, Salesforce. This is a major story in the SaaS world because Salesforce is now making Claude an interface to its CRM. Salesforce and Anthropic launched CloudForce, beginning with Salesforce and Claude.
Thirty-seven pre-built sales skills that can interrogate CRM data, evaluate pipeline, and take governed actions directly from Claude. Open beta is expected in September. I think that this is a major, major move from one of the absolute top brands in SaaS, Salesforce, and I do think that it will force the hand of many other CRMs and other SaaS brands to follow suit, which means that they are accepting now that the AI layer has been wedged between themselves and their customers.
And so if you can't beat them, join them, as they say. SaaS interfaces are starting to become back-end infrastructure for AI agents rather than destinations that humans must visit. So the next evolution after GEO is agent discoverability and operability. So these are things like APIs, Model Context Protocol or MCP, documentation, and permission-aware actions.
So this was a very interesting move by Salesforce. I think what they're admitting now is that most people are going to want to manage their CRM through an AI interface, and Claude, of course, is leading the way here via its MCP technology. Next story is similar because Semrush is doing the same thing.
Semrush put its SEO intelligence directly inside Claude. They've launched an official Claude connector on August twenty-sixth. Customers can query its keyword, backlink, competitive, and AI visibility datasets conversationally and automate workflows such as competitor analysis and reporting. Semrush says its connected dataset includes twenty-eight point eight billion keywords, forty-three trillion, that's trillion with a T, backlinks, and three hundred and seventeen million large language model prompts. All right.
That's a lot of data. It's mind-boggling to try to even imagine forty-three trillion backlinks. But basically, Semrush is effectively handing over its entire database to Claude via its Claude connector. And its biggest competitor, Ahrefs, has had a fantastic connector for a long time now that our teams use every day.
Probably the competitive pressure was on for Semrush to follow suit here. But just like in the world of CRM and SaaS, SEO dashboards are now becoming data layers behind agents. So routine keyword analysis and reporting is commoditizing very rapidly, and the human value here moves more towards strategy, experimentation, and business interpretation.
All right, let's move to the next story. Publisher licensing may influence ChatGPT citations. Okay. A company called Press Ranger and Otterly AI matched 129.3 million citations against 91 confirmed AI publisher licensing agreements. Publishers with OpenAI deals averaged 48% more ChatGPT citations per cited page, and OpenAI-only licenses showed 112% difference.
Comparable home platform effects did not appear for Google or Perplexity. So not a big surprise here. If a publisher has a licensing partnership with ChatGPT, then those are going to be sources that tend to get cited more in ChatGPT citations.
So the citation ecosystems may be partly shaped by commercial content access relationships, not just purely page-level optimization. So keep this in mind when you prioritize the publications that you go after for citations, especially for ChatGPT. And it's better to do that rather than just to treat all earned media as interchangeable.
This is a major factor. In fact, it actually makes me think that in CiteForge, in our product, GEOforge, CiteForge is the citation module. We have a priority score, and I think we might want to bake that in to the priority score, which is, does this publisher have any licensing deal with OpenAI? Okay, let's move on to the next story coming from arXiv.
A controlled experiment finds that Google AI answers reduce publisher clicks. This sounds like a big duh, but here's the data. A pre-registered field experiment with eleven hundred participants found that removing Google AI answers and AI Mode increased publisher click-through rates, unsurprisingly. An AI Mode-only experience reduced publisher traffic and also lowered reported user experience and trust.
So this is just providing stronger causal evidence than most SEO studies that AI search can change click behavior, not simply correlation. So, of course, when forecasting clicks with AI, we should assume that just because organic clicks are declining, it doesn't mean that visibility is also declining.
In fact, visibility can even rise while organic sessions decline. All right, next story comes to us from none other than the Boston Consulting Group. BCG reported on August twenty-sixth that nearly one-third of surveyed customers now use AI somewhere in the purchase journey. More importantly, AI introduced users to brands that they otherwise would not have considered in roughly sixty-three percent of AI-assisted journeys.
So AI is actively changing which brands even enter the consideration set. It can alter the competitive shortlist itself, and particularly for SaaS brands, and I would include cybersecurity here, I think it's very important to measure where the brand appears in these real bottom-of-funnel comparative prompts, like best X for Y or alternatives to X and which vendor should I shortlist?
Because people now are turning to LLMs really as a shortcut to a short list. So something that in the past might have been many, many different Google searches and dozens of clicks into websites to build a comparative table. Now they're just taking a shortcut and having LLMs do that right away, and then moving right from the shortlist into those branded search-driven website visits.
All right. We're now on the last story of the day, which is another GEO platform is moving from monitoring towards recommended actions. This is a company called Cairrot, and this is spelled C-A-I-R-R-O-T. And I have no idea if I'm pronouncing that correctly, but I'll just call it Cairrot. Cairrot 2.0 on August 28th has added more than 20 GEO and AI-search capabilities and an insights engine, which is intended to turn visibility data into prioritized SEO and prescribed GEO actions.
So as we have seen several times and talked about a lot on this show, the GEO software category is quickly moving from dashboard to diagnosis to prescribed execution. So pure monitoring will be very difficult to defend as a premium agency service. It's becoming more and more of a commodity. And the moat now here is increasingly about proprietary knowledge, offsite authority, implementation, testing, and revenue attribution.
This is where agencies should focus their GEO service efforts, not on monitoring and delivering only dashboards. All right. That wraps up episode 11. Hope you all have enjoyed this weekend edition, and we'll see you on the next one.
