OpenAI is building ChatGPT answers off licensed partner datasets such as Daloopa, PitchBook, and LSEG News, expanding the source graph beyond the crawlable public web. On episode 27 of The GEO Show, Paris Childress covers AirOps CMS and PM publish integrations, the ChatGPT Work data agent, Claude share gains, ecommerce AI referrals, and crowded GEO agencies.
OpenAI is building ChatGPT answers off licensed partner data, not only the crawlable public web. On episode 27 of The GEO Show, Paris Childress argues that commercially important evidence may live in hosted datasets GEO practitioners cannot influence through conventional content optimization, then walks ten stories spanning CMS abstraction, warehouse agents, ads co-opetition, ecommerce referrals, LLM share shifts, crawl-to-refer ratios, persona simulation, shopping citation pillars, and a crowded GEO agency category.
The throughline is source control. When indexes include licensed financial feeds and work moves into the LLM via MCP, visibility strategy has to cover rooms beyond ranking pages for human traffic.
ChatGPT for financial services now includes premium datasets from Daloopa, PitchBook, and LSEG News. OpenAI says it indexes and hosts partner data itself and uses granular citations so users can trace figures and claims. Public-web optimization still matters, but it cannot rewrite evidence that lives only inside licensed partner indexes.
The source graph behind AI answers is expanding beyond crawlable web pages, and some commercially important evidence may live inside licensed or connected datasets that GEO practitioners cannot influence directly through conventional content optimization.
AirOps launched native integrations with seven CMSs and four project-management tools so insight can reach publish without manual handoffs. Separately, OpenAI's ChatGPT Work data agent connects major warehouses and investigates metric moves in natural language. Paris ties both to the headless SaaS pattern: marketers live in the LLM while warehouses and CMSs become connected backends.
Similarweb's ecommerce 2026 report shows AI referrals up more than 200% year over year on still-small volume, up to a 2:1 purchase advantage from recommendation, and 89% overlap with conventional search. August web share puts Claude at 9.3% while ChatGPT falls to 55.5% and Gemini rises to 25.6%. Add Claude to tracking and keep watching branded bottom-funnel search.
Search Engine Journal documented Anthropic crawl-to-refer ratios via Cloudflare from about 71,000:1 to 2,200:1. Bots crawl more while human referrals drop. Chattopic argues visibility should start with personas and journeys, not static prompts, matching Gumshoe-style persona-first prompt discovery for more runs and better statistical relevance.
Hexagon's 50,000-recommendation study attributes 73% of citations to authoritative content, third-party validation, and structured data accessibility, with strong DA60+ roundup correlation. Paris agrees on authority and third-party validation while noting structured-data correlation remains debated versus earlier episodes. Good Firms lists 1,178 GEO agencies across 80 countries at a $37 median hourly rate, evidence that SEO-to-GEO rebranding has erased most early-mover advantage.
00:31. AirOps: seven CMS and four PM native integrations.
02:42. OpenAI licensed financial data: Daloopa, PitchBook, LSEG News.
04:35. ChatGPT Work data agent across warehouses.
11:14. Claude at 9.3% Similarweb August web share.
12:40. Crawl-to-refer ratios ~31x apart.
16:00. Hexagon 50K shopping recommendations and three pillars.
The throughline is licensed and connected evidence: when answers leave the public crawl, GEO has to track rooms content optimization cannot enter alone.
Paris: Hi, everybody. Welcome back to another episode of The GEO Show, brought to you by GEOforge: full self-driving for AI visibility. We have got some great stories to run through today on Friday, September the 11th. The GEO world is active, and I am trying to bring you everything that you should be paying attention to: news stories, trends, experiments, hype versus reality. So our first story today is more on the reality front. It's coming from one of the leaders in this space, AirOps.
Paris: AirOps closes the gap from insight to published content. AirOps has launched native integrations now with seven content management systems that include Webflow, WordPress, Contentful, Sanity, Contentstack, Ghost, and Strapi, and also four project management platforms that include Asana, ClickUp, Monday.com, and Airtable. Its explicit pitch is removing manual handoffs between insight, content production, review, and publishing.
Paris: So we are seeing a move here by AirOps similar to what recent moves have come from Profound, which is that they want to position as the marketing system of record. just like Claude and other AI tools that are wedging themselves between the human marketers and their tech stacks, AirOps and Profound are doing the same thing. They wanna make it so that you can publish to your CMS without actually having to open your CMS.
Paris: They want you to be able to manage tasks and update tasks between their platform and your project management platform, be it Asana, ClickUp, or Monday, without having to go to those platforms. So it's really abstracting away both the CMS and the project management tool, which are both major tools in a marketer's toolbox So I think it's an interesting development, and I do think that this trend is gonna continue.
Paris: we also have integrations with Webflow and WordPress, native integrations that allow publishing straight to those CMSs. And I can tell you that it's not that simple. You actually have to create and map several custom fields, and we have actually created a WordPress plugin specifically to do that as part of our WordPress integration. So expect more like this from the major tool providers. Next story, AI answers are moving from the public web to licensed data. This is coming directly from OpenAI.
Paris: ChatGPT for financial services now includes premium datasets from Daloopa, PitchBook, and LSEG News. OpenAI says that its indexes and hosts this partner data itself and uses granular citations so that users can trace figures and claims back to their evidence. So what are we seeing? We have reported recently that OpenAI is building its own index, which bypasses the public web, and part of that effort is these licensed data deals that they're doing with these large data providers.
Paris: So the The source graph behind AI answers is expanding beyond crawlable web pages, and some commercially important evidence may live inside licensed or connected datasets that GEO practitioners cannot influence directly through conventional content optimization. I think this trend is gonna keep continuing. OpenAI does not wanna rely on real-time search through Bing and through Google to provide its answers, and these are the majority of the answers that it provides to users. It performs this real-time search. They want to be more self-sufficient.
Paris: They wanna build their own index, and as part of that effort, they wanna tap into proprietary large datasets through content providers. So optimizing and being retrievable is gonna be more than just being indexed on the public web, but it's going to be inclusion in these public datasets where it makes sense All right, let's move on to the next story here. The analytics dashboard is becoming a conversation. OpenAI launched the data agent in ChatGPT Work.
Paris: It connects to platforms including BigQuery, Databricks, Snowflake, Redshift, and ClickHouse. It investigates metric changes, creates interactive dashboards, and supports follow-on action through natural language conversation. So OpenAI is doing what we've seen from Claude. It is abstracting away a lot of the dashboards from, uh, financial and analytics point solutions, and here it's also happening with the large data warehouses and data lakes, so the likes of BigQuery, Databricks, Snowflake, Redshift. So standalone analytics interfaces are becoming less defensible.
Paris: The users are increasingly expected to ask, "Why did this number move?" And then have a system that investigates, rather than them having to manually navigate charts So SaaS tools that are specifically focused on analytics and data visualization, I think are really at threat here.
Paris: But even more so, as we've seen in the CRM space and in other major SaaS categories, because this was reported by Salesforce basically going headless, and that was followed by HubSpot, and I believe Zendesk was doing it as well. These big SaaS companies are basically saying, "We acknowledge the fact that users don't necessarily wanna log into multiple platforms. They don't necessarily wanna come in here and manipulate their data and build charts and graphs here in our tool.
Paris: They would rather do it in Claude or in ChatGPT Work." I think that again, this layer is getting more and more solid, where users are just gonna be living inside of their LLM of choice for doing all sorts of work and really their tool stack just becomes large data warehouses that are feeding ChatGPT and Claude via MCP connectors. All right, next story. AI discovery platforms can also become competitors to advertisers.
Paris: PYMNTS, citing The Information reported on September tenth that OpenAI told some partners it would no longer accept ChatGPT advertising for image and audio generation products that compete with OpenAI's own products. Adobe Firefly was reportedly affected by this. OpenAI's public ad policy gives it broad discretion over advertisers, but does not explicitly enumerate the specific competitor restriction. So this really reminds me of Google in some ways, both allowing advertisers in certain verticals to advertise through Google Ads, but also competing directly with them.
Paris: I'm thinking about, uh, Google Travel and flight search, Google Finance, and there's probably lots of other examples where Google has a friendly co-opetition where they are certainly accepting advertising and ad dollars from large players in verticals that they have direct solution. Here, it looks like OpenAI is maybe taking a different route and even blacklisting some of the most direct competitors to their own solutions.
Paris: So I think what it means for GEO practitioners is that we have to really also watch very closely what's happening in the ChatGPT ad space, and we always wanna separate AI visibility within ads as compared to organic AI visibility. All right, our next story comes from Similarweb. Similarweb's new State of Ecommerce 2026 report says that direct referrals from AI platforms increased more than 200% year over year, yet remain relatively small in traffic volume.
Paris: More importantly, an AI recommendation gave the recommended brand as much as a 2:1 purchase advantage in some cases. Eighty-9% of consumers using AI for shopping research also use conventional search. All right, so we're seeing two things happening. We are seeing that referral traffic overall is still growing very, very fast. Referral traffic from AI is still growing very, very fast year over year.
Paris: But overall, the actual number remains very small in terms of traffic volume and probably a site's percentage of overall traffic from referral sources, it's probably still a small percentage, but it's a very valuable percentage. And eighty-9% of consumers using AI for shopping research also using conventional search.
Paris: I'm gonna guess that follows a typical pattern where they start with ChatGPT, or they might even start with Google, but because they're starting at an earlier stage, they will be getting AI overviews crowding out the typical search results. They're gonna do their research, they're gonna shortlist brands, and then they're gonna go to Google to do branded search. And there's gonna be the first high-intent search.
Paris: Real search that happens at the bottom of the funnel on Google is gonna be for those brand names specifically with the navigational intent, not informational or educational intent. And I think that's the most common path, and that's why we always need to really watch closely what's happening with branded search volume inside of Google Search Console. And this data refers to e-commerce, but I think it's also very relevant for B2B lead gen as well Let's move on to the next story.
Paris: Claude is no longer a rounding error in AI usage. Similarweb observational data from August web traffic has said it puts ChatGPT at 55.5%, Gemini at 25.6%, and Claude now at 9.3%. Rounding it out, we have DeepSeek at 3.4%, Grok at 2.4%, and Copilot at 1.6%. Perplexity at 0.9%. Ouch. So twelve months earlier, ChatGPT was at 73% versus 55% today. Gemini was at 13% versus 25% today. Claude was at only 1.9% versus now 9% today.
Paris: So a lot of moving and shifting and of course, this is just web traffic. This does not capture all of the usage on mobile apps and desktop applications, but it's still pretty indicative of market share, relative market share among the big LLMs. So ChatGPT is dropping pretty fast. They're really losing share to Gemini primarily, and Claude has made major leaps. Claude is even approaching a ten percent market share, and that's probably with a lot of knowledge workers, I'm guessing.
Paris: So those of you who are tracking AI visibility in major engines like ChatGPT and Gemini, uh, you may wanna think about adding Claude to that list. Okay, next story. One widely cited AI metric produced answers more than thirty times apart. Search Engine Journal documented Anthropic's crawl-to-refer ratios attributed to Cloudflare ranging from 71,000 to 1 down to 2,200 to 1, which is roughly a 31x spread. So what we're talking about here is crawl-to-referral ratio.
Paris: That means how many times a page is crawled by AI bots relative to how many times AI refers actual human traffic to that page. And these numbers are getting worse and worse for when it comes to referral traffic from humans. So it's AI bot is exhaustively crawling, but not really referring human traffic. And these discrepancies, these huge discrepancies of 30x apart can arise from a measurement window, from bot grouping, website sample, and missing referrals from environments without a referrer header.
Paris: But the bigger picture here is that AI is increasingly crawling the web much more often, which is good. But on the downside, referrals, human AI referrals through those pages continues to drop, and that critical ratio of crawl to referrals keeps looking worse and worse. Okay, on to the next story. Prompt tracking is expanding into persona and buying journey simulation. Chattopic's September 9th product positioning says AI visibility should begin with the user rather than a static prompt, and I fully agree.
Paris: Its platform now emphasizes buyer simulation, persona-level visibility, AI persona chat, and AI focus groups, arguing that the same question can produce materially different answers for different customer contexts. I think this is very important, and very few visibility tools that I have seen are doing this the right way.
Paris: They're actually putting different prompts into the contexts of different buyer personas, meaning that you could have the exact same prompt searched three times by three different buyer personas, and the LLMs will give different answers and recommend different brands to each of those different personas. And the best tools that are, that are doing this, I think Gumshoe comes to mind here.
Paris: they are going through the process of creating unique buyer personas first, and then doing prompt discovery after you have those buyer personas in place. So then you align the prompts to those already established buyer personas, and I think this is the right way to do it.
Paris: And Gumshoe also uses this as an opportunity to build more statistical relevance because that's what gives them the ability to do more runs because they have multiple prompts per persona, and that's how they get to up to 30 to 40 runs per prompt All right, next story. A 50K recommendation study points to three AI discovery pillars.
Paris: Hexagon says it analyzed 50,000 shopping recommendations across ChatGPT, Perplexity, and Claude over twelve months and claims three factors: authoritative content, third-party validation, and structured data accessibility explain 73% of brand citations. It further reports that 92% of recommendations contained at least one brand appearing in a domain authority sixty-plus roundup page that-- and that robust structured data appeared among eighty-9% of consistently recommended brands versus 31% of rarely recommended brands.
Paris: All right, let's unpack this dense set of statistics here because it does contradict what we have stated earlier in this show in other episodes. One thing that I think we can all agree with are the three factors that drive GEO today, and that is authoritative content, and I will add to that authoritative content, high information gain content, which carries authority by virtue of its information gain. The second is third-party validation. Absolutely. That's probably the biggest factor.
Paris: And the third is structured data accessibility. Here, there's a bit of debate because we've seen earlier studies that show there isn't a strong correlation of structured data on a page and that page having high citation rate. But here, the report says that Well, that's not the case. Also, what's interesting, we've reported that domain authority, which is a real SEO... It's a SEO dominant metric. Um, it's even really the currency metric for link building.
Paris: Domain authority, DA, that 92% of recommendations were coming from brands that were appearing in DA sixty-plus roundup pages. So that's interesting. That's arguing that getting inclusion and reaching out for inclusion in roundup pages with high domain authority is still really worthwhile, and I generally agree with that. All right, onto our last story of the day. GEO is already a crowded global agency category.
Paris: Good Firms' September tenth update says its directory now contains 1,178 GEO agencies across eighty countries, with a reported median hourly rate of $37 and 9% of those agencies employing at least fifty people. So this is directory classification data. It's not a census of pure-play GEO providers. But this is no big surprise. I think that we have seen hundreds and hundreds and maybe more than a thousand SEO agencies now rebranding themselves, repositioning and rebranding themselves as GEO agencies.
Paris: So maybe six to eight or twelve months ago, repositioning in the GEO by an agency was probably-- you were probably one of the first movers, and there were some early mover advantages, and I already think that that's gone. I think now that probably the majority of SEO agencies are now repositioning and rebranding as GEO agencies, or maybe they're still straddling that fence and saying, "We do now SEO and GEO," and and they're adding GEO to their list of service offerings.
Paris: But the trend line here is very, very clear. As we see the zero-click trend, as we see AI overviews taking over, as we see Google's march to AI mode, it's clear that SEO, traditional SEO, which is designed to rank pages so that they get human traffic, that is really fading away pretty quickly. And now GEO first is the best strategy to win a GEO and still get as much, squeeze as much juice out of what remains of SEO as possible.
Paris: All right. That wraps up today's episode 27. Thank you all for listening, and we will see you all in the next episode.
