Google organic search still sends far more traffic than AI: a study of 95 properties by The SEO Works found organic sent about 39x more sessions than AI, while Marie Haynes found ChatGPT visitors converted at close to 14% vs 3% for organic. Paris Childress argues GEO is predominantly a brand channel that should be funded from brand budgets, then covers ChatGPT Ads in 89 countries, Semrush finding AI Overviews on 88% of commercial keywords, Sistrix research on the page features that go with AI citations, altered results served to rank trackers, Google's Universal Cart, Walmart's agent gateway, Seer Interactive on ChatGPT fan-out variance and Meta's delayed crawling terms.
Google organic search still sends far more traffic than AI. In the largest of three recent studies, The SEO Works found organic sent about 39x more sessions than AI across 95 properties. Paris Childress's point is that traffic is the wrong way to judge GEO, because AI's influence shows up in places analytics cannot see, and because GEO is predominantly a brand channel.
Far less, according to all three studies. Marie Haynes looked at 100 sites and 12 months of GA4 data and found ChatGPT sessions were just 0.62% of Google organic sessions. Across the 55 sites that track conversions, Google organic drove 964 conversions, while all LLMs combined drove 36. The SEO Works looked at 95 properties between September 2025 and August 2026 and found organic sent about 39x more sessions than AI, a gap that narrowed from 44x to 33x after May 2026. Kevin Indig, writing in Growth Memo and reported by Search Engine Roundtable, looked at four sites and found no reliable link between AI visibility and branded or direct traffic. Rand Fishkin summarized all three, arguing AI search is over-hyped and over-invested.
Sometimes. In the Haynes data, ChatGPT visitors converted at close to 14% vs just 3% for organic. The SEO Works found AI traffic converted at 1.18x the organic rate overall, but worse than organic on about a third of sites: about 23 of the 63 sites.
GA4 counts clicks from AI Overviews and AI Mode as organic. Those clicks are AI-influenced, but GA4 cannot separate them from clicks on the blue links. The samples are also small: only 55 of Haynes' 100 sites track conversions.
Mostly a brand channel. Mike King of iPullRank argues GEO is a brand channel, not a performance channel, and should not be judged on traffic alone. Paris agrees and goes further:
"I believe that GEO should be funded from brand budgets, not funded from performance or SEO budgets."
In his view GEO contains SEO but also involves brand marketing, product marketing, PR and some performance marketing, so it deserves a seat at a larger table than the SEO team.
Very. Semrush tracked 400 US commercial keywords for a year and found AI Overviews on 88% of them in August 2026, up from about 46% a year earlier, and on 97% when weighted by search volume. By query type:
Paris reads this as a methodical move by Google toward AI Overviews on practically all queries, and eventually a default AI Mode experience, which raises the question of whether that traffic counts as organic or AI.
Sistrix compared about 48,500 pages cited by ChatGPT, AI Mode and AI Overviews with uncited pages from the same sites, using 96,418 URLs from 17,380 domains in the DACH region (Germany, Austria and Switzerland). Cited pages were more likely to have:
llms.txt, breadcrumbs, HowTo and QAPage schema and readability scores showed no lift, and product schema scored 0.8x. ChatGPT favored question-style titles (1.9x), while Google AI Overviews favored article schema (2.0x) and disfavored product pages (0.5x). Paris's advice: don't worry about llms.txt, include article schema, put the current year in the title, and use question-style headers.
The traffic debate will keep going, but the direction is clear: as AI Overviews and AI Mode absorb more of Google, the line between organic and AI traffic gets harder to draw, and measuring GEO by referrals alone gets less useful.
Paris: Hi, everybody, and welcome to episode 71 of The GEO Show, brought to you by GEOforge: full self-driving for AI visibility. I'm your host, Paris Childress. This is a weekend edition, and what is really dominating the headlines is a recent and still-churning debate between the SEO community and the new GEO, let's say, the GEO defenders.
Paris: And it has to do with a few studies that have been recently put out about the measurable attributable impact of AI traffic. So let's get right into it. Here's what happened. There were three studies of interest here. The first study came from Marie Haynes, who looked at 100 sites and 12 months of GA4 data across those 100 sites.
Paris: She found that ChatGPT sessions were just 0.62% of Google organic sessions. ChatGPT visitors, though, converted at close to 14% versus just 3% for organic. And across the 55 sites that track conversions, Google organic drove 964 conversions versus all LLMs combined driving just 36 conversions. That's study number one.
Paris: Study number two was from The SEO Works, which looked at 95 properties between the period of September 2025 and August 2026. And what they found was organic sent about 39 times more sessions than AI, and that gap narrowed from 44x to 33x after May 2026. But what differed here in this study is about the conversion rate.
Paris: AI traffic across this pool of 95 properties, AI traffic converted at 1.18 times the organic rate, so just a slightly higher. But it converted worse than organic on about a third of those sites. About 23 of the 63 sites, conversion rates were actually worse from AI-referred traffic. And the last study came from Kevin Indig, who puts out the fabulous Growth Memo newsletter, and it was reported in Search Engine Roundtable.
Paris: He looked at four different sites, and he found no reliable link between AI visibility and branded or direct traffic. And so he looked deeper at the theory that AI-referred traffic is only a fraction of AI's influence, but you really need to look at the branded traffic and the direct traffic because a lot of people who are not clicking or rather going and searching for the brand on Google, and also they may be just copying the links from AI answers and then dropping those into their browser, which shows up as direct traffic.
Paris: But he did find no reliable link or strong correlation between AI visibility and that branded and direct traffic. And last, Rand Fishkin came in and summarized all three of these reports, arguing for the most part that AI search as a new channel is a bit too hyped up and over-invested. Now, there are a few caveats, and this is a very contested story.
Paris: The samples are relatively small. Only 55 out of Haynes' 100 sites track conversions. And one very important thing to note here is that GA4 counts clicks from AI Overviews and AI Mode as organic, so they don't show up as AI traffic. But this is very much AI-influenced because it is coming from AI Overviews and AI Mode.
Paris: But GA4 cannot distinguish between clicks coming from AI Overviews and AI Mode versus clicks from the blue links. And the GEO crowd has responded, and the best response I believe came recently from Michael King, Mike King from iPullRank, which is that GEO isn't in fact a performance channel, it's a brand channel, and we shouldn't be measuring the impact of GEO on the basis of its traffic alone 'cause it has greater influence.
Paris: And that does beg the question that if AI is a brand channel, then what is the KPI and who owns it? Should SEOs be the ones in charge of managing GEO? And I think that is right now where most of the debate comes from. My opinion is that GEO is something much, much larger than SEO. It contains SEO and in a way it can be seen partially as a performance channel, but it is predominantly a brand channel, and that it also involves brand marketing, product marketing, and definitely PR, as well as some aspects of performance marketing. And in that sense, I think that GEO deserves a seat at a larger table, and I believe that GEO should be funded from brand budgets, not funded from performance or SEO budgets.
Paris: Okay, let's move on to the next story. ChatGPT ads are now in 89 countries. OpenAI added ChatGPT ads in 26 more countries, bringing the total to 89. It says that more than 100,000 advertisers have run campaigns. From 30th of October, AI creative enhancements switch on by default for self-serve campaigns.
Paris: So unfortunately, this rollout has not yet come to Bulgaria, but it has come to a lot of other places, 89 countries in total now. Reuters notes that ads revenue run rate for OpenAI is now at one billion dollars a year. And the creative enhancements are really interesting because they can rewrite headlines and descriptions, they can resize images, and they can optimize destination URLs and even translate ads.
Paris: So right away, OpenAI is moving in with AI enhancements that we've seen recently being pushed by Meta Ads and also Google Ads.
Paris: So this is a big deal. Ads now in ChatGPT are really a global channel covering pretty much every major country in the world. So if you're advertising globally, ChatGPT Ads has you covered. And interestingly, that AI rewritten ad copy for these multilingual and multinational campaigns will run without the approval of each version unless the advertisers opt out of that feature.
Paris: So a bit of advice for anyone who is running global ad campaigns on ChatGPT, beware of this October 30th deadline and make a decision on these creative enhancements before that deadline, especially if you're in regulated categories. Next up, we've got really important new information coming out of Semrush about AI Overviews presence.
Paris: AI Overviews is now on 88% of tracked commercial keywords. Semrush tracked 400 US commercial keywords for a year. AI Overviews appeared on 88% of them in August 2026, which is up from about 46% year on year. And the best X for Y queries, which is a really popular query format, reached 96% for AI Overviews.
Paris: So across 400 keywords, and these are commercial high-intent bottom-of-funnel keywords, coverage rose from 45% a year ago to 88%, or actually 97% when weighted by search volume. This is how this breaks down. Best X for Y query types went from 42% to 96% year over year. Reviews queries went from 30% to 89%.
Paris: Brand A versus brand B went from 63% to 86%. Alternatives, queries with the word alternatives went from 48% to 82%, and pricing queries went from 43% to 82%. So this is bottom of funnel. The first big wave of AI Overviews really took away all the clicks from the top of the funnel from more of the early research or educational informational intent type of queries.
Paris: And now AI Overviews has moved down to dominate the bottom of the funnel as well, which is really clear from this research. I think this is a methodical move by Google to eventually have AI Overviews cover practically 100% of all Google queries. That's including the commercial queries, bottom of funnel, branded queries, local intent, basically everything, because this is their way to slowly transition the entire Google experience over to AI Mode. And when you put this in context, I think the inevitability of Google AI Mode in the context of the first debate, what really happens then if all of the AI referred traffic, let's say, is gonna be coming from Google AI Mode, which becomes the default?
Paris: Well, is that then organic traffic or is that AI traffic, and how do you define that? And is ranking number one still a meaningful goal at all for SEO when it's a default Google AI Mode experience? Next up, some research from Sistrix around which page features show up alongside AI citations. Sistrix compared about 48,500 pages cited by ChatGPT, AI Mode, and AI Overviews with uncited pages from those same sites.
Paris: A table of contents, schema markup, and clear heading structure went with more citations. llms.txt showed no measurable advantage. Okay, so their method was they took 96,418 URLs from 17,380 domains. The cited and uncited pages were paired by site, and it looked at data from the DACH region.
Paris: And what they found was how much more often each feature appeared on cited pages. When the current year in the title was on the page, it was cited 2.7 times more. When there was a table of contents, 1.7.
Paris: When there was article schema, 1.6 times more. When there was five or more H2 headings, 1.6 times more. When there was FAQPage schema, 1.5 times more. Tables with a header row, 1.5 times more, and an author box with a photo, 1.4 times more. Specifically, there was no lift from an llms.txt file, breadcrumbs, HowTo or a QAPage schema.
Paris: These were all basically 1.0x, or readability scores. In fact, product schema scored a 0.8. And when we look at this by platform, ChatGPT did favor question-style titles. So pages that had question-style titles appeared 1.9 times more often than pages without. And AI Overviews from Google favored article schema with a 2.0 times more citation rate than pages without article schema, and they disfavored product pages with 0.5 times the multiple there. So a few big takeaways here for me.
Paris: The biggest one is putting the current year in the title. That gave you a 2.7 times greater likelihood of getting that page cited versus pages that don't have the current year in the title. So whenever you can, I think putting the year in the title and then updating that is really important. The second big takeaway is that llms.txt files really don't matter.
Paris: They don't have an impact. And even all these different schemas, article schema seem to matter maybe a little bit for Google AI Overviews, but not so much for ChatGPT. And maybe the last takeaway for ChatGPT is ChatGPT's favoring of question-style titles.
Paris: So if I were to roll all this up, I would probably give three or four pieces of advice. First of all, don't worry about llms.txt. Not really important. Second of all, do include article schema because it does seem to matter. You have a two times greater likelihood of citation on average when you have article schema for Google AI Overviews.
Paris: The next piece of advice here would be, when possible, do include the current year in the title. That has a big impact. And finally, I would use question-style headers because ChatGPT does favor question-style headers quite a lot. The next story is also quite controversial. Google is serving altered results to rank trackers.
Paris: Two SEO data companies say that Google is showing automated rank trackers results that look normal but have been swapped. Nozzle measured 8.4% of its tracked results being affected since mid-September. Ahrefs puts it at 5% to 20%, depending on the data provider.
Paris: Here are the details. Nozzle, which has about 27 million results pages across 52 countries, reported that 8.4% of its results pages were altered since September 13th, up from just 0.17% before that date. The peak was 28% on the 29th of September, and altered pages kept only 22% of the usual sites.
Paris: Ahrefs reported between 5% to 20% of results affected depending on the provider. It didn't disclose its method, and other trackers reported different rates. DemandSphere reported 11%. AccuRanker reported under 1%. And there is no Google statement on these findings.
Paris: If this is true, then this is really a big blow to the rank tracking tool space. And in general, I think we should all be a little bit skeptical of any sort of data that gets scraped from search results. Bear in mind that Google Search Console data here is not affected at all, and I think that's still probably the highest quality data set when we wanna see SEO results from Google as well as the generative AI results that are now reported there.
Paris: And getting back to what I said earlier, the inevitability of Google AI Mode, where there will just no longer be pages ranked at all, I think that will make rank tracking, you know, pretty much just a useless exercise anyway. Okay, next story. Google's Universal Cart is live in US Search and in Gemini.
Paris: Google's Universal Cart, which was first shown off at I/O in May, is now appearing in US Search and Gemini. Shoppers can add items from product listings to a cart that pays with Google Pay. The cart icon sits in the top right of the results page, and product listings show an Add to Cart button. It uses Google Pay and has a My Orders tab.
Paris: Now, Google did not make any official announcement here. There was no Google rollout post, and which merchants are eligible is also not described. But what matters here is that Google is shortening the path from discovery in its own results. And that includes search results, shopping results, and that is gonna get brought into AI Mode, I'm sure.
Paris: So they're shortening the path from discovery within Google to purchase inside of its own pages, and that means fewer visits to merchant sites. So in many cases, if users have enough information from the snippets that are provided by Google and they're convinced enough to make a purchase at that point, they can just add those to a checkout cart inside of Google and check out without a visit to the merchant site at all.
Paris: And this could have major implications on the visits to those merchant sites, but it also could have big implications on the sales. By smoothening or shortening the path to purchase, some of these vendors might even see an increase in sales, but at the same time, they might see a big, big drop in the visits to their pages. And more on shopping in an AI-first world, Walmart opens a gateway for AI shopping agents.
Paris: Walmart announced Agent Profiles, a developer agent gateway, and seven rules for AI shopping agents. Agents that break the rules will be blocked. Here are the details. Agent Profiles will launch in the next month. Customers approve every purchase or set a spending limit with $50, $100, and $200 given as examples.
Paris: The agent gateway is a beta built on an MCP. Agents never receive card details, and the integrations that are named so far are Sparky, Meta's Muse, ChatGPT, and Gemini. Walmart gave no adoption figures and no firm dates, but this is the largest US retailer now setting very clear and specific terms for agent traffic, which other retailers are likely to follow on with.
Paris: And of course, we're all watching to see what Amazon will do next here, because right now their current position is that they're blocking all of these personal agents from coming to Amazon and shopping on the user's behalf, which I think is a big mistake, and I think that they're gonna reverse that decision eventually. All right, next story.
Paris: Google tests AI-written snippets in AI Overview citation cards. Google appeared to be testing AI-written summaries in the citation cards next to AI Overviews, replacing the publisher's own snippet text. Brodie Clark spotted it first. Barry Schwartz saw it in some of his queries, but not all, and there are two screenshots so far.
Paris: The summary appears to be based on what Google judges most relevant to the query. Now, this is a small sample, and Google has not yet commented. If it does roll out, publishers will lose control of the text shown beside their link. And this has been happening for many years now with classic SERP snippets, which is that Google does not, by default, always show your meta description, but it might pull from other parts of the page if it thinks that it is more relevant to the user.
Paris: So that begs the question, who's accountable if the card misrepresents your page? Google is already writing, of course, and rewriting most titles and snippets anyway, and they're generally doing a pretty good job of this. Let's also consider the fact that a lot of real-time search retrieval now is reading only the metadata from the snippets.
Paris: They're only reading the meta title and the meta description. So if Google is rewriting those in the citation cards as well, this is gonna affect the type of data and the information and the entities that is getting retrieved by AI search. Bottom line, though, there's probably not a whole lot that we as marketers can do about it if Google is deciding to rewrite those snippets.
Paris: Okay, this next story comes from research by Seer Interactive. And it's about how AI visibility tools measure ChatGPT. Seer Interactive ran one ChatGPT prompt repeatedly. Two identical runs about 8 minutes apart shared none of the same search queries. So they're looking at the fan-out queries that result from a prompt, basically.
Paris: Two identical runs that are 8 minutes apart shared zero of the same fan-out search queries. Very interesting. URLs retrieved for the prompt went from 14 to 40 between runs. Of the 40, 34 were never cited in the answer. So when they're doing these repeated runs, the URLs sometimes were 14, the URLs that were retrieved.
Paris: Other times they were 40. That was the range. And when they were 40, 34 were never cited in the answer. Okay. The format of the underlying search data also changed between runs. Seer recommends running each prompt repeatedly and reporting visibility as a range, not as a single number. And I am really happy to see this because this is exactly our philosophy with SignalForge, is we do 30 runs in a row for every prompt, every time we're updating the visibility of that prompt, so that we can account for this very, very high variability.
Paris: And we are reporting the results as a range by showing the average as a line and showing a margin of error above and below that line.
Paris: So again, a single run of a prompt is not a reliable measure of visibility. This is not like rank tracking. Onto the last story of the day. Meta delays its website crawling terms after publisher backlash. Meta pushed back new terms that would let it crawl websites linked to its ad and business tools from October 30th to November 20th.
Paris: Publishers say they can't block the crawler without hurting their ads. Here are the details. Reuters reported the delay and Meta reportedly kept the disputed wording. The News/Media Alliance, which represents more than 2,000 US publishers, demanded that Meta withdraw the terms. The UK competition regulator, CMA, has asked Meta to explain the terms. So where this is headed is that advertising with Meta may come to imply consent to crawling.
Paris: And this does open a whole new front in the fight over AI access to websites that goes beyond just robots.txt, 'cause if you're advertising with Meta, Meta is now taking the position that it has the right to fully crawl your website, not just the landing page that may be used in the ad, but your entire site. But Meta says that robots.txt does still work, actually. All right.
Paris: That is the last story of the day, but we now have an additional lightning round as part of a new format that I'm gonna test out, and I wanna run through some quick stories of this lightning round. First, it is confirmed Claude Duet is live. Anthropic's new agent runs its own browser, which can buy through Stripe's Link and Shop Pay, as reported by Anthropic.
Paris: So this might be the first real shopping agent from Claude Duet. Next, Microsoft Clarity's AI visibility features were tested by Brainlabs. AI bots fetched pages about 90 times for every visit they sent on average. Next, Google's Gary Illyes on robots.txt. Robots.txt controls whether a crawler can access a page, not how data already crawled is used.
Paris: Okay, that's a clear, interesting clarification. Next, AI Mode loses Export to Docs. Google removed that shortcut. I wonder why that's the case. And last, Profound adds Moz-powered Google data to its page reports. Okay, that is a wrap. That'll do it for episode 71.
Paris: Thank you all for tuning in, and we'll see you all in the next one.
