Direct traffic rose about 141% across 12 websites in Kevin Indig's research, while engaged direct sessions fell about 1%, pointing to bots rather than hidden AI traffic. Paris Childress covers the week's GEO news under one theme, tools commoditizing, attribution staying murky and agents getting infrastructure: Demandbase's AEO agent, Peec AI's agent and actions, Marie Haynes on ChatGPT sending 0.62% as much traffic as Google, Beacon's $20 monthly monitoring, Google's generative AI search controls, Anthropic's Claude Duet, Walmart's agent gateway, OpenAI models in Google Cloud Model Garden, and ChatGPT Ads creative enhancements.
Direct traffic up 141%: is it bots or AI? Paris Childress, founder of Hop AI and co-founder of GEOforge, covers this week's GEO news under one theme: tools are becoming a commodity, attribution stays murky, and AI agents are getting real infrastructure.
Demandbase introduced three AI marketing agents, including an AEO agent that tracks how AI assistants describe and recommend brands to members of B2B buying groups. It draws on a reported 342 million contacts across 13 million companies, flags visibility gaps, outdated information and influential sources, and connects findings to Demandbase campaign execution. Paris calls it the first time he has seen ABM move into GEO.
Peec AI formally presented its redesigned actions experience and Peec Agent, which turn visibility findings into prioritized recommendations across technical issues, content and earned media. It reports more than 25 integrations, and agent changes require user approval. Paris sees Peec AI moving from monitoring toward execution, with a smart focus on off-site earned media.
Search Engine Land reported Marie Haynes' analysis of more than 145 million sessions across 100 websites: ChatGPT sends just 0.62% as much traffic as Google organic, and all AI assistants combined reach about 0.73%. Google drove about 27 times more conversions. Paris notes AI referrals are only a slice of AI influence, since people also come direct or search the brand.
Kevin Indig examined 12 websites. Direct sessions rose about 141% while engaged direct sessions moved about minus 1%, and bot activity explained much of the growth. 632 tests found no reliable relationship between AI visibility and later direct or branded search. Paris still believes direct traffic to deep pages is predominantly AI.
Change Agents Corporation plans to launch Beacon on October 28 with a free AI visibility audit and a $20 monthly subscription covering about 100 queries across five AI assistants. Paris says tracking is now a commodity, and a price that low likely means shallow sampling and low confidence share of voice data.
Google updated its guidance to explicitly include robots meta tags, data-nosnippet, X-Robots-Tag and the Search generative AI control. These are existing controls, not new restrictions, and Paris recommends reading the documentation closely.
Anthropic opened a research preview of Claude Duet, a proactive agent that monitors connected apps and surfaces developments, via a waitlist for US Pro and Max subscribers. Paris sees it as Anthropic's answer to OpenClaw, Meta's Muse, Grok Bot and OpenAI's Dots.
Walmart announced agent profiles and an agent gateway for Walmart and Sam's Club, letting customers authorize agents, set spending limits and separate agent purchases, with structured access through MCP. Paris says this is the right infrastructure for agentic commerce.
OpenAI plans to offer frontier models and Codex workflows through Google Cloud's Model Garden, while Microsoft stays its primary cloud partner. Paris hopes this means AI avoids the walled gardens of past tech waves.
OpenAI told self-serve advertisers that creative enhancements will be enabled on October 30, adjusting headlines, descriptions, images, URLs and translations, possibly without individual approval. Advertisers can opt out. Paris sees ad optimization becoming autonomous, like Meta ads.
Paris: Hi, everybody. Welcome to episode 68 of The GEO Show, brought to you by GEOforge: full self-driving for AI visibility. Today's first story is coming to us from the ABM world, account-based marketing. Demandbase, one of the leaders in ABM, launches an AEO agent for B2B buying groups. Demandbase introduced three AI marketing agents, including an AI agent that tracks how AI assistants describe and recommend brands to members of B2B buying groups.
Paris: Its system draws on the company's reported intelligence footprint of 342 million contacts across 13 million companies. The AEO agent identifies visibility gaps, outdated information, and influential sources. It can connect those findings to Demandbase campaign execution. Really interesting.
Paris: This is the first time that I've heard of ABM getting involved in GEO, and Demandbase is one of the leaders in ABM, so it makes a lot of sense actually. I think that they have this massive database, and they have the ability to immediately trigger specific campaigns off of AI visibility intelligence information.
Paris: So it will be interesting to see how this develops with them. Next story comes from Peec AI. Peec AI releases its expanded AI agent and actions experience. Peec formally presented its redesigned actions experience and Peec Agent consolidating capabilities previously described in its October 13th changelog.
Paris: The system turns visibility findings into prioritized recommendations across technical issues, content improvements, and earned media opportunities. Peec reports more than 25 integrations enabling workflows involving CRM systems, content publishing, project management, and outreach. Agent-generated changes require a user approval, and the updated experience is available across Peec plans with plan-dependent credit allowances.
Paris: So Peec is continuing to strengthen its position as an execution environment rather than simply a monitoring tool. So they're doing pretty well here with the integrations, 25 integrations. Most of that seems to be with the CRM and more emphasis on off-site earned media opportunities, which is a really good idea in my opinion.
Paris: Next up, Search Engine Land reports new analysis finding ChatGPT sends just 0.62% as much traffic as Google. Marie Haynes analyzed more than 145 million sessions across 100 websites. ChatGPT referrals represented approximately 0.2% of Google organic sessions. Including other AI assistants raised the figure to 0.73% of Google organic traffic.
Paris: And just to note, this is not 0.73% of total traffic. It's just 0.73% of Google organic traffic, less than one percent. Among 55 sites tracking conversions, ChatGPT traffic had a higher reported conversion rate than Google organic traffic, but Google generated approximately 27 times more conversions overall than all AI assistants combined. So this does put things into the bigger perspective here.
Paris: Google organic still dominates. Google organic sessions still dominate, compared to the referrals, AI referral traffic. Now, AI referral traffic, I believe strongly is still only a small slice of what's really happening. The AI referral traffic that you see, let's say, in Google Analytics, I believe is just a fraction of all your traffic that is influenced by AI because people can be influenced by AI and then come to you directly, and they can also go and search for your brand, and neither of those channels or pathways would show up as AI referral traffic in Google Analytics.
Paris: So keep that in mind, but still this puts things into perspective that organic search is still really the number one dominant source of traffic to websites. Now, on that note, that leads us directly into our next story. New research from Kevin Indig challenges the theory that AI traffic is hiding in direct. Kevin Indig examined Google Analytics data from 12 websites.
Paris: Across the studied period, total direct sessions increased approximately 141%, while engaged direct sessions changed by approximately minus 1%. Further inspection suggested substantial bot activity behind the apparent traffic growth. Indig also conducted 632 tests of relationships between AI visibility and later direct or branded search activity.
Paris: He could not establish a reliable relationship after accounting for alternative time alignments and other methodological limitations. This does not prove that AI has no direct influence, though. Really interesting, and this pushes back against the popular belief of AI attribution.
Paris: Hidden AI attribution does remain plausible, but aggregate direct traffic growth is not sufficient evidence. I think that most of the direct traffic now that arrives to non-homepage pages of a website, I think that there's strong evidence to show that is predominantly AI traffic, 'cause just logically, it doesn't make sense that a lot of human users would be typing in direct, long, let's say, deep pages of a website.
Paris: Most of the direct from humans would come straight to the homepage. But direct traffic to other pages is most likely coming from AI, and it's not human-generated traffic. So the total direct sessions increasing 141% is not really coming from humans. Interesting, though, and this story continues to develop, and attribution remains quite murky for GEO. Next up, we have a new product entering the GEO space.
Paris: Again, it's a new day, and there's a new competitor. Beacon enters the category with planned $20 monthly GEO monitoring. Change Agents Corporation announced plans to commercially launch Beacon on October 28th. The product targets small businesses with a free AI visibility audit and a $20 monthly subscription.
Paris: Its announced capabilities include competitor comparisons, approximately 100 common customer queries, and monitoring across five AI assistants. Optional implementation services provide an additional potential revenue stream. All right. Well, another entry with a low-cost provider that's just proving more and more that AI visibility audits and AI visibility tracking tools are becoming pure commodities.
Paris: $20 per month most likely means that the depth of sampling of those prompts is very shallow and the confidence interval in reported share of voice data is most likely very low confidence. And they're probably not showing that either. But still, most buyers who are entering this space aren't really that educated about the measurement rigor that needs to take place and the depth of sampling.
Paris: So, more tools are just taking advantage of that with low-cost offerings. Next up, as reported in Search Engine Land, Google formally documents controls governing generative AI search use. Now, this is quite important. Google updated its guidance on controlling what website owners share with search.
Paris: The documentation now explicitly includes existing mechanisms such as robots meta tags, and also data-nosnippet, X-Robots-Tag, and the Search generative AI control. These are existing controls being documented more clearly. They are not newly introduced crawler restrictions. So publisher control over content access and generative AI presentation is now becoming much more explicit thanks to Google, and I think it's very important to read through this documentation and really understand how we can present the best possible data to AI search crawlers.
Paris: Okay, and the next story, another big one. Anthropic introduces Claude Duet, a proactive AI agent. That was announced yesterday. Anthropic opened a research preview of Claude Duet, a proactive agent designed to monitor connected applications, surface relevant developments, and assist with ongoing tasks.
Paris: A keyword here is proactive. The preview begins through a waitlist for eligible US Pro and Max subscribers. Claude Duet represents a move beyond isolated question and answer interactions toward persistent delegated work. Its commercial adoption and effects on brand discovery are not yet established, obviously.
Paris: So I think this is probably Claude's answer to the big wave started initially by OpenClaw, and then recently with the launches of Meta's Muse, GrokBot, and OpenAI's Dots. Users seem to want a more persistent, proactive AI assistant that is going to flag them when interesting things come up, and generally have this persistent kind of chat happening all the time with an AI agent that is constantly monitoring things and helping you to prioritize what it should be doing. The next story: Walmart introduces an agent gateway and controlled shopping profiles.
Paris: Walmart announced agent profiles and agent gateway and terms governing AI agents shopping through Walmart and Sam's Club. Agent profiles are intended to let customers authorize agents, establish spending limits, and distinguish agent purchases from their own. The gateway will provide structured access through a Model Context Protocol, MCP, rather than requiring agents to navigate conventional consumer webpages.
Paris: Developer access begins through a beta program while consumer profiles are planned for the coming month. Very interesting stuff here. Agent commerce is developing an infrastructure layer that involves authorization, identity, and machine-readable interfaces. This makes a lot of sense to me. It isn't efficient for an agent to arrive at, let's say, walmart.com or an e-commerce site and conduct a shopping experience by clicking through on a Chrome browser the way that a user would do.
Paris: It makes much more sense for that to be structured through an MCP. And Walmart is taking some really important steps here by making sure that these agents are authorized by the customer, that they have different spending limits, so they can't go crazy and go rogue and spend too much, and also just distinguishing the agent purchases from the human purchases.
Paris: So Walmart seems to be putting the right infrastructure in place for agentic commerce, which is coming at us fast and furious. It'll be interesting to see if Amazon follows suit in any way, and maybe also Shopify and the other big e-commerce players. Next story. OpenAI and Google Cloud announce access to OpenAI models through Model Garden.
Paris: OpenAI announced plans to make frontier models and Codex workflows available through Google Cloud's Model Garden. The arrangement will allow enterprise customers to access OpenAI models through Google Cloud infrastructure and procurement channels. General availability is planned in the coming weeks.
Paris: And just to note, Microsoft remains OpenAI's primary cloud partner. So enterprises are increasingly able to combine models from competing providers within shared infrastructure. This reminds me of what GrokBot did recently by basically allowing the users of GrokBot to use other models beyond simply Grok's models where there are limitations, but it can use Claude's models and OpenAI's ChatGPT models as well.
Paris: And so I think that this is a similar type of move where no matter what environment you're in, you could be in Google Cloud in this Model Garden, which I'm hearing for the first time, or you could be with your assistant, but you wanna have the ability to use whatever the best model is for the task.
Paris: And I think the whole ecosystem is really starting to adapt to that user preference, which is great. Meaning that we won't hopefully see the same type of walled gardens in AI that we have seen in previous waves of technology. And onto the last story. ChatGPT Ads plans automatic creative enhancements starting October 30th.
Paris: OpenAI notified self-serve advertisers that creative enhancements are scheduled to become enabled for new and existing campaigns on October 30th. The feature can adjust headlines, descriptions, images, destination URLs, and translations based on conversational context or user interests. Creative variants may run without individual advertiser approval.
Paris: Advertisers can opt out before the activation date, or they can adjust the setting later. Great. So paid AI advertising is becoming more dynamically personalized, potentially making individual ad exposures difficult to reproduce. So I think they're taking a page out of social media ads, particularly Meta ads, where basically AI-driven creative testing and iterative creative testing using AI-generated creative enhancements is gonna help improve the testing.
Paris: It's gonna help improve the conversions and the results and the return on ad spend. So a really positive move here by OpenAI, and I think that this means that ads, just like with Meta and even in Google Ads as well, ad optimization is becoming more and more autonomous, where the role of the human expert who's optimizing becomes more and more strategic and more focused on feeding the right type of data into the platforms so that they have better context, and then letting automation do things like audience identification and targeting and creative iteration and creative testing.
Paris: All right. That will do it for episode 68. Thanks for tuning in, and we'll see you all in the next one.
