A First Page Sage study of 1,089 brands finds that clear offerings and consistent brand information beat schema markup for AI recommendations, while FAQ schema and llms.txt showed no measurable lift. In this solo news episode, Paris Childress also covers DemandSphere data showing AI Overviews on 82% of tracked branded keywords, Cloudflare Pay Per Use and Google paying publishers for AI answers, the SPUR content telemetry standard, the dismissal of the Chegg and Penske Media suits, Safeway shopping in ChatGPT, ChatGPT ads attribution gaps, publishers selling GEO, and Profound's research on personal shopping agents.
A 1,089-brand study finds that clarity beats schema markup for AI recommendations. In this solo episode, Paris Childress, founder of Hop AI and co-founder of GEOforge, walks through this week's top GEO stories.
Less than you might think. First Page Sage ran 4,213 commercial prompts and 657 agent shortlisting tasks across 1,089 brands in 14 industries. After controlling for authority, clear offerings and suitability correlated with an 11.2 percentage point recommendation lift and consistent brand information with 9.4 points. Product schema in ecommerce added 4.8 points, generic schema 0.4, and FAQ schema and llms.txt showed no measurable lift.
DemandSphere reports that the share of its tracked branded keywords returning an AI Overview rose from 26% on September 1 to 82% on September 29, peaking at 90.48% on September 27. Google has not announced a change, and an AI Overview does not mean the brand was cited.
Cloudflare's Pay Per Use beta lets AI companies pay publishers per cited answer, agent report, or shopping recommendation. Google is reportedly paying about 100 publishers based on how much their content contributes to AI Overviews, AI Mode, and Gemini. The SPUR Coalition released a content telemetry standard for tracking how AI uses content, and a judge dismissed the Chegg and Penske Media antitrust suits over AI Overviews.
Safeway shopping in ChatGPT turns a recipe or list into a cart and checkout. ChatGPT ads still face attribution gaps that keep budgets in test mode. Publishers now sell GEO services, and Profound is researching personal shopping agents like Meta Muse. Paris argues that proactive, always-on agents will make share of voice a short-lived KPI, and that share of conversions is what to measure next.
Subscribe to The GEO Show wherever you get your podcasts, and watch the full episode on YouTube: https://www.youtube.com/watch?v=hCl1-SqgR-c
Paris Childress [00:00]: Hi, everybody. Welcome back to The GEO Show, brought to you by GEOforge: Full Self-Driving for AI Visibility. We're on episode 55. It is October 3rd, a beautiful Saturday in Sofia, Bulgaria, and this weekend I'm gearing up to speak at BrightonSEO next Thursday, and gearing up for that talk. I'll be speaking about information gain and the role of information gain in the content production and distribution process, and as a critical element of a comprehensive GEO strategy.
Paris Childress [00:43]: So if anyone is planning to attend BrightonSEO in Brighton, catch my talk. It's at 9:30 AM on day one of that conference, and, uh, really looking forward to it. All right. Let's get into today's top stories, starting with the first story coming from Digiday reports.
Paris Childress [01:05]: AI content telemetry gets a common event model. The Standards for Publisher Usage Rights, or SPUR for short, Coalition released version one of its content telemetry standard on October 2nd. The framework tracks five distinct elements: content being retrieved, grounded, cited, presented, and engaged with.
Paris Childress [01:32]: Members include The Guardian, Financial Times, BBC, Sky, The Times, Mediahuis, and AP. SPUR has invited OpenAI, Anthropic, Google, Meta, and Microsoft to an advisory board and plans to implement pilots. So the industry, the publisher industry is starting to formalize the exact event chain that GEO platforms have been optimizing towards, and that is retrieval, grounding, citations, presented, or I guess that's mentioned, and engaged with.
Paris Childress [02:10]: So there seems to be some alignment forming between the major publishers and the GEO industry. Next up, some pretty big news from Cloudflare. Cloudflare creates a market for paying per AI use. Cloudflare's new pay-per-use system lets an AI company specify what downstream use it will pay for and at what price.
Paris Childress [02:40]: Publishers choose whether to participate. Cloudflare then records usage, bills the AI buyer, and pays the publisher. Paid events can include a cited AI search answer, a source used in an agen- an agent report, or a review that shapes a shopping recommendation. Cloudflare explicitly connects the product to answer engine optimization tooling.
Paris Childress [03:04]: This is very interesting. So the economics here are such that publishers will actually get paid to have its data-- to have their data become cited in AI responses, and the, the one paying will be the AI buyer. So the AI buyer here will pay for this exclusive publisher content so that they can crawl it effectively, and then when that content gets surfaced in an AI answer, I suppose through either a citation or used, used in an agent report, like an agent recommendation or shopping recommendation, at that point in time, they're paying, they're paying per use.
Paris Childress [03:51]: So that is a really a fascinating business model, and most of us will not-- m-most websites don't have the, the leverage to do that, but the largest publishers now have more options. They can allow AI to crawl them freely and cite them, or they can disallow crawling, they can block the crawlers, which pretty much takes away all chance of AI visibility for those publishers.
Paris Childress [04:19]: And now there's a new option, which is pay per use. They can let them crawl, but then they would get paid whenever their content surfaces in answers or recommendations. Really interesting stuff here. Next up, more on publishers winning in the GEO era. Google reportedly begins paying about 100 publishers for AI contribution. The Verge, citing The Information and earlier Digiday reporting, says roughly 100 publishers are participating in a Google pilot where compensation is tied to how much a publisher-- how much publisher content contributes to AI Overviews, AI Mode, and Gemini.
Paris Childress [05:02]: Sounds like the previous story, kind of like a pay per usage or pay per contribution. One early participant reportedly earned more than one million dollars over a year. Another newer participant reportedly earned roughly fifty to sixty thousand dollars. So Google's apparently experimenting with moving publisher relationships from the tra-- the classic traffic exchange, which was really the unwritten agreement for the better part of two decades, now towards content contribution economics.
Paris Childress [05:36]: Also, another fascinating move, Google is paying for publishers' contributions to AI answers. So they're not just paying straight up licensing fees to access the content, but it's more like paying for performance.
Paris Childress [05:53]: Next story. In light of all this new, uh, emergence of new business models between AI engines and content publishers, a publisher antitrust lawsuit against Google AI Overviews has been dismissed. US District Judge Amit Mehta, uh, we've heard of that name before. That was the judge on the massive Google antitrust lawsuit from a couple of years ago.
Paris Childress [06:22]: That judge dismissed separate antitrust suits from Chegg and Penske Media Corporation, alleging that Google's AI-powered search features improperly diverted publisher traffic. The court held that the claims did not establish an antitrust violation. So it looks like in the near term here, publishers are less likely to force a return to the old click-based search bargain through antitrust legislation.
Paris Childress [06:51]: Um, now licensing, regulation, technical controls, and commercial deals like the last two stories that we just reported on are more likely to shape the market. This next story is a fun one. ChatGPT turns a grocery idea into a basket and checkout. Albertsons and OpenAI launched a Safeway shopping experience in ChatGPT.
Paris Childress [07:16]: Users can begin with a recipe, photo, a digital list, or a request. ChatGPT surfaces relevant products and discounts, builds a cart, and directs the shopper to a Safeway checkout. Albertsons operates more than twenty-two hundred stores serving over thirty-six million customers weekly, and the experience is planned to expand to other Albertsons banners.
Paris Childress [07:40]: Really cool idea here. AI discovery is compressing need to recommendation to product selection to basket collection and to transaction in one conversational workflow. The future is really rushing towards us here. We see agents now acting and shopping on buyer's behalf.
Paris Childress [08:05]: So what-- for the GEO industry, um, I think commerce GEO, the subset of GEO, commerce GEO, will eventually need SKU and product-level measurement, not just brand visibility. And that's going to include things like product eligibility, recommendation, offer selection, cart inclusion, and of course, cart abandonment, and purchase.
Paris Childress [08:30]: And these should become distinct events in a, uh, f-fully measured GEO strategy for physical e-commerce. Now, this next story is pretty shocking. AI Overviews on branded Google queries jumped from twenty-six percent to eighty-two percent, as reported by DemandSphere. DemandSphere reports that the share of its tracked branded keywords returning a Google AI Overview increased from twenty-six percent on September first to eighty-two percent on September twenty-ninth, briefly reaching ninety point four eight percent on September twenty-seventh.
Paris Childress [09:14]: Wow. From twenty-six percent to eighty-two percent in one month, last month, September first to the twenty-ninth. Keywords were checked once daily across markets and devices. Google has not announced this change, and DemandSphere notes that the AI Overview-- AI Overviews appearing does not mean that the brand itself was cited.
Paris Childress [09:37]: So this is-- has long been considered the safest part of traditional search. Safest meaning the most insulated for AI Overviews, crowding out traditional results and taking away clicks. But now it appears that this is going away too. Uh, the-- You have a user explicitly typing your brand name, presumably in order to navigate to your website.
Paris Childress [10:00]: And now, as opposed to that website link being at the very top of the page, it's getting pushed down by AI Overviews. So maybe Google is experimenting here. Um, maybe Google is trying to test the real search intent behind these, what were presumed to be strongly navigational queries. And maybe they think now we're just gonna provide an AI Overview and a summary of this brand to give people a sense of what this brand is all about.
Paris Childress [10:30]: And if they wanna navigate, of course, they'll have to scroll down past the AI Overview. So that is gonna have a major impact on what's usually the strongest and most reliable organic source of traffic, which is people searching for your brand name with navigational intent. So now that appears to be under threat if this trend is gonna continue.
Paris Childress [10:50]: We're not really sure if this is just an isolated test yet. This next story is about schema markup. 1,089 brand study, clarity beat schema markup by a wide margin. First Page Sage ran a four thousand two hundred and thirteen commercial prompts plus six hundred and fifty-seven agentic shortlisting purchasing tasks across one thousand and eighty-nine brands in fourteen industries.
Paris Childress [11:21]: So quite a large control set here. After controlling for authority signals, clear offerings and suitability correlated with a plus eleven point two percentage point recommendation difference. Consistent brand information across first and third-party sources with a nine point four percentage points recommendation difference, and product schema in e-commerce with just a plus four point eight percentage point recommendation difference, and generic schema markup with just zero point four percentage point recommendation difference.
Paris Childress [12:03]: FAQ schema and llms.txt showed no measurable positive difference in this data set. So this flies in the face of one of the most popular and talked about technical on-page tactics of GEO, which is you should have an llms.txt file that AI bots can crawl, and all of your pages need to have schema, especially the FAQ pages.
Paris Childress [12:31]: They should have schema markup. Well, this is actually really refuting that, that, uh, commonly held belief here, because what it's showing is that clarity is what really gets you the recommendations. Um, so clear offerings and suitability and also consistent brand information across first and third-party sources.
Paris Childress [13:02]: Consistency and clear offerings. So that is a very interesting finding. So don't over-index and think that because you've got schema on your pages, you're checking the box for GEO. It, it really means that you've gotta pay a lot more attention to the actual content of that page.
Paris Childress [13:23]: Next story, some attribution woes for ChatGPT ads. ChatGPT ad attribution problems are keeping budgets in test mode. Digiday reports that Accuracast saw lead form submissions from a ChatGPT-only campaign, while ChatGPT still showed zero conversions two weeks later. It also reported roughly one hundred platform clicks versus about twenty in the agency's tracking.
Paris Childress [13:55]: Other agencies did not universally report the same issue, so this is not evidence of a platform-wide effect. One executive said clients spending ten million dollars monthly on Google were putting less than a hundred thousand dollars per month into ChatGPT. While Accuracast said measurement uncertainty was preventing ten to twenty thousand pound per month tests from scaling towards a hundred thousand per month.
Paris Childress [14:24]: So clearly here, a paid AI discovery is developing faster than its attribution infrastructure, and this is really artificially holding back a lot of potential ad spend for ChatGPT ads because there are, what appears to be in this test, massive differences between the conversions that are reported between ChatGPT and the, uh, the internal platforms, and the same is true with clicks.
Paris Childress [14:54]: So here in this example, ChatGPT showed zero conversions while the agency had clear lead form submissions from a ChatGPT-only campaign. And on the other side, ChatGPT appears to be over-reporting the paid ad clicks.
Paris Childress [15:16]: Next up, as if there weren't already enough, uh, SEO adjacent players and agencies moving into GEO, it's now happening at the publisher level. Publishers are turning GEO into a sellable media product. Reported also by Digiday, Future's GEO product has thirty plus clients and renewals. Ziff Davis has multiple clients, and at least half a dozen large publishers in the US and Europe now offer some form of GEO service.
Paris Childress [15:52]: Publishers offering this service combine branded content, existing editorial authority, AI visibility measurement, and in some cases, agent-readable content formats. Publishers also acknowledge that attribution remains unsolved, as we just saw earlier. So now the publishers themselves are becoming both competitors and potential execution partners to the GEO SaaS category.
Paris Childress [16:25]: And of course, they own one thing that most GEO software platforms do not: source authority and distribution. And today's last story is about none other than Profound. Profound shifts its research focus toward personal shopping agents, and this is not a surprise given that these personal agents are now red hot.
Paris Childress [16:53]: Profound published research on October second examining Meta Muse and Instinct-style personal agents. And of course, this comes on the heels of OpenAI's major announcement and launch of Dots. And they are tracing Muse's shopping process across fifty test prompts and early website activity signals.
Paris Childress [17:15]: Profound's framing is that agent visibility depends increasingly on accessible product information, catalogs, and clear signals about who a product is appropriate for. And this term of clear signals is appearing yet again in the same episode. Clear signals beating classic structured data. So what's the interpretation here?
Paris Childress [17:41]: Um, the major players in the space, they are already seeing the change that is happening. Conversations are moving beyond AI search results coming from LLMs, and they're moving towards agents that compare products, and they make decisions on behalf of users.
Paris Childress [18:02]: And this is already gonna be a whole, a whole new world. Practically before GEO even had a chance to get established, it's already changing in a major way because classic LLMs were, for the most part, reactive. The user has, has, has to go to an LLM with a problem, with a question, and initiate that conversation with a prompt, and then get some kind of response back.
Paris Childress [18:27]: And then brands, of course, they wanna be part of that response. But now with these agents, with these a- these proactive agents are anticipating the needs of their users, and they're proactively pinging, and there's a persistent always-on conversation going on. So you don't really have this prompt anymore, which is what people think about as a, a kind of an evolution or a longer form of a keyword or a search term.
Paris Childress [18:52]: What you have is an ongoing conversation that never really stops. And so then, you know, trying to think about how do we measure this? How, how are we going to measure our vi- visibility for a set of prompts? That really doesn't take into account this type of user and agent always-on conversational behavior.
Paris Childress [19:12]: So I, I don't really know what's gonna come next here, but I do think that this is more evidence and strong evidence that this share of voice KPI is really short-lived because what we really need to see now is, is just what share of conversions, not, not even necessarily coming through your website because the agents are gonna be the ones going to your website to convert and transact on behalf of their users.
Paris Childress [19:37]: But what share of actual monet- monetized revenue-generating events are happening by agents and being referred by AI? Not necessarily just how often is your brand becoming visible in an LLM's response to a user proactively prompting an LLM. So that'll be very interesting to follow. All right. That will wrap it up for today's episode, 55.
Paris Childress [20:07]: Hope you all enjoyed it, and see you all on the next one.
