Google has made Gemini 4 Hydrogen the default model in AI Mode worldwide, and it also handles the hardest questions in AI Overviews. Host Paris Childress says citation baselines from before November 12 may no longer hold. The same briefing covers an EU draft law on paid AI shopping agents, a Peec AI survey on AI search spend and citations, a Search Console logging error, Bing's AI opt-out, a Hostinger survey on double-checking AI recommendations, Sitecore's Marketing IQ, and a Google analytics help page that leaves out AI features.
Google has made Gemini 4 Hydrogen the default model in AI Mode worldwide, and it also handles the hardest questions in AI Overviews. Host Paris Childress says that when models change, the sources pulled into the answers can change too, so AI Mode and AI Overviews citation baselines from before November 12 may no longer hold.
The same briefing covers the EU's Digital Fairness Act proposal on paid AI shopping agents, a Peec AI survey on AI search spend and a sharp drop in Reddit citations, a Search Console logging error, Bing's AI opt-out, a Hostinger survey on people double-checking AI recommendations, Sitecore's Marketing IQ, and a Google analytics help page that leaves out AI features.
Google says Gemini 4 Hydrogen is now live in Google Search for all users globally. It is the new default model in AI Mode, and it handles the hardest questions in AI Overviews. On November 12, Nick Fox and Robbie Stein from Google said on X that Hydrogen is available globally in Search for all users. Stein called it the new default in AI Mode, also used for the hardest questions in AI Overviews. Google says it improves query fan-out and linking, connecting users to a broader range of helpful links, and that it gives more visual answers, including more images, comparison tables, and custom tools, plus better language understanding in every AI Mode language. There are no benchmark figures yet. Paris notes that Google's blog post introducing Hydrogen could not be opened at the time of recording.
The change comes two days after Gemini 4 Carbon reached AI Ultra subscribers in AI Mode, and during the November Google core update. Paris says we should start to notice changes across rankings, AI Mode, and AI Overviews.
When models change, it also changes which sources get pulled into the answers.
His take is that AI Mode and AI Overviews citation baselines from before November 12 may no longer hold, and that share of voice reports for those surfaces may get volatile.
The European Commission adopted its Digital Fairness Act proposal on November 11. The proposal is COM/2026/1100, and Article 26 covers AI shopping agents. As reported, AI shopping agents would have to disclose any payment they get from traders, and shoppers would have to confirm those purchases. On first use, the consumer chooses how much confirmation they want. Article 26(4)(b) requires disclosure of any payment from traders, and a paid or conflicted purchase must be confirmed whatever the setting. Fines in large EU-wide cases could reach 6% of worldwide turnover.
Recital 59 extends personalized advertising rules to product rankings and recommendations, including in natural language systems. That this covers sponsored AI chat answers is PPC Land's reading. The draft also includes a general dark patterns ban and influencer marketing labeling, with brands and agencies liable. PPC Land links it to United Airlines' ban on AI agents booking flights. Paris's point is that this is the first major draft law in the EU that names paid placement inside AI agent recommendations, which is directly relevant to ChatGPT ads and sponsored agents.
A Peec AI survey of 700 marketing decision-makers found 89% raised AI search spend, but only 24% can prove a return. Sapio Research ran the survey for Peec: 700 managers and above at companies with 100-plus staff, 300 companies from the US, 200 from the UK, and 200 from Germany. 81% of respondents expect to raise their spend again. 64% of those who raised it significantly cannot prove a return. Paris calls the ROI gap a real budget risk for 2027.
Peec's own platform data covers more than 100 million AI chats from July to September. ChatGPT conversations with a single search fell from 98% in July to 39% in September, which Paris reads as people going deeper and having longer, multi-turn conversations. Reddit citations fell from 59 in 100 to 9 in 100. Company websites rose from 52% of citations to 58%. Product pages rose from 13.8% to 20.1%. "Best X" listicles fell from 16.2% to 7.3%. If the citation shift holds, the weights are moving from third-party listicles and Reddit toward a brand's own pages.
Google confirmed a logging error that lowered impressions in Search Console's Generative AI performance report from November 9 to 11. Google says only data logging was affected, not actual performance. The window overlaps the start of the Google core update, so week-over-week AI impression reports and any reading of core update impact will show a false dip during those days.
Microsoft is promoting two ways to turn off AI answers in Bing: a Bing AI search choice browser extension, still in preview, and adding minus AI to a query. Both first appeared in June. PPC Land reports that the extension also makes Bing the default search engine. Paris sees building the opt-out with a default engine switch as mostly a distribution play, with little effect on how many people see AI answers. As far as he knows, Google is not offering a turn-AI-off option.
In a Hostinger survey of 1,014 US adults, 93% say they have received a business recommendation from AI, but 94% also check another source before acting, and only 6% would act on AI alone. 65% say they have left Google results for Reddit or TikTok. Paris's takeaway is that AI often puts a brand on the shortlist, but the website and the reviews still help close the decision.
Paris closes with a lightning round: ChatGPT Ads adds region targeting within a country and combined product set filters; Profound says ChatGPT commercial conversations rose 119% from June 2025 to June 2026, with no stated method; Imprint raises $130 million at a $1.9 billion valuation and offers a server that lets AI agents make purchases; and a Google paper on 2.1 million image generation chats finds people appear in nearly 80% of them.
Paris: Hi, everybody, and welcome back to The GEO Show, brought to you by GEOforge: Full Self-Driving for AI Visibility. I'm your host, Paris Childress, recording on Friday the 13th of November, and we have some great stories to run through today. So let's get right into it. This first story is about Gemini 4.
Paris: Gemini 4 Hydrogen becomes the default model in Google AI Mode globally. Google says Gemini 4 Hydrogen is now live in Google Search for all users globally. It's the new default model in AI Mode, and it handles the hardest questions in AI Overviews. Google says it improves query fan-out and linking and gives more visual answers. On November 12th, Nick Fox and Robbie Stein from Google said on X that Hydrogen is available globally in Search for all users. Stein called it the new default in AI Mode, also used for your hardest questions in AI Overviews. So here's a list of some of the changes. Google says there is better query fan-out and linking, connecting users to a broader range of helpful links. So maybe that means more citations. Let's see. Also, more images, more comparison tables and custom tools, and better language understanding in every AI Mode language. Google's blog has stated that they're introducing Gemini 4 Hydrogen, but we could not open that post itself, and there are no benchmark figures yet.
Paris: And this comes just two days after Gemini 4 Carbon reached AI Ultra subscribers in AI Mode, and also during the November Google core update. So a lot of things moving here in Google. We've got a core update, we've got Gemini 4 Carbon, which is really their new frontier model, and now Gemini 4 Hydrogen, which is powering AI Mode and the most challenging, hardest questions in AI Overviews. So we should start to notice now a lot of changes happening in Google across rankings, across AI Mode, and across AI Overviews. It's important to keep in mind here that when models change, it also changes which sources get pulled into the answers. So AI Mode and AI Overviews citation baselines that you maybe have been tracking from before November 12th may no longer hold up. So expect some volatility in your share of voice reports for Google AI Mode and AI Overviews.
Paris: Next up, the EU's Digital Fairness Act proposal would make AI shopping agents disclose payments. The European Commission adopted its Digital Fairness Act proposal on November 11th. As reported, AI shopping agents would have to disclose any payment they get from traders, and shoppers would have to confirm those purchases. Fines in large EU-wide cases could reach up to 6% of worldwide turnover. Ouch. So here are some details. The proposal is COM/2026/1100. Its Article 26 covers AI shopping agents. On first use, the consumer chooses how much confirmation they want. And in Article 26(4)(b), it requires disclosure of any payment from traders. A paid or conflicted purchase must be confirmed whatever the setting. Recital 59 extends the personalized advertising rules to product rankings and recommendations, including in natural language systems. This covers sponsored AI chat answers, that's PPC Land's reading. It also includes a general dark patterns ban and influencer marketing labeling, with brands and agencies liable. PPC Land links it to United Airlines' ban on AI agents booking flights. So this is the first major draft law in the EU that names paid placement inside AI agent recommendations. So that is directly relevant to ChatGPT ads and to sponsored agents.
Paris: Next up, we've got some fresh new research from Peec AI, a survey that they ran which showed 89% of companies raised their AI search spend, but only 24% can prove a return. A Peec AI survey of 700 marketing decision-makers found 89% raised AI search spend, but only 24% can prove a return. Peec's own platform data also shows Reddit citations falling sharply and company websites gaining share. Let's dig deeper. Sapio Research ran the survey for Peec, so it was 700 managers and above at companies with 100-plus staff, 300 companies from the US, 200 from the UK, and 200 from Germany. And 81% of respondents expect to raise their spend again. 64% of those who raised it significantly can't prove a return. That's a little bit troubling. Platform data covers 100 million-plus AI chats from July to September, and there's some very interesting results from that.
Paris: First, that ChatGPT conversations with a single search fell from 98% in July to 39% in September. So let that sink in. ChatGPT conversations with just a single search, meaning they did not have follow-up questions, or it was not a multi-turn conversation, went from 98% single-search conversations to 39% just between July and September. So that's telling me that people are going much deeper and spending longer and having longer conversations in ChatGPT. Reddit citations fell from 59 out of 100 to just 9 out of 100. So a massive sharp decline in Reddit citations. And what picked up the slack in part was company websites. Company websites rose from 52% of citations to 58%. Product pages rose from 13.8% to 20.1%. And best X listicles, which we all know and love, fell from 16.2% to 7.3%. So taken in whole, this is very, very interesting. Reddit is declining sharply. Listicles are declining sharply as well, and company websites are on the rise when it comes to citations. So if the citation shift holds, the weights are moving from third-party listicles and Reddit towards a brand's own pages. Another big takeaway here is that the ROI gap for GEO is still a real thing, and it's a real budget risk for 2027.
Paris: Next up, Google confirms Search Console's AI report undercounted impressions between November 9th and 11th. Google confirmed a logging error that lowered impressions in Search Console's Generative AI performance report from the 9th to the 11th of November. Google says only data logging was affected, not actual performance. The window overlaps the start of the core update from Google, which is kind of ironic. So week over week, AI impression reports and any reading of core update impact will show a false dip during these days. Keep that in mind.
Paris: The next story is about Bing. Bing re-promotes two ways to switch off AI answers. Microsoft is promoting two ways to turn off AI answers in Bing. A Bing AI search choice browser extension, which is still in preview, and adding a minus AI to a query. Very interesting. Both first appeared in June. PPC Land reports that the extension also makes Bing your default search engine. Okay. So building the opt-out with a default engine switch makes it mostly a distribution play, and I think it will have little effect on how many people see AI answers. And as far as I know, Google is not offering a turn AI off option.
Paris: The next story has to do with the search everywhere thesis. A Hostinger survey says that these are actually US adults: 93% have had an AI business recommendation and 94% double-check it. So in a Hostinger survey of 1,014 US adults, 93% say they've received a business recommendation from AI, but 94% also check another source before acting, and only 6% would act on AI alone. 65% say they've left Google results for Reddit or TikTok. So very, very interesting here, especially that the large majority of people are not just taking the AI answer at face value, but that they are going to another source to double-check and confirm those answers. And you would naturally think that they would go to Google next, but actually they're going to Reddit and TikTok as well, and using those more and more as search engines to validate their AI responses. So the important takeaway here is that AI often puts a brand on the shortlist, but the website and the reviews will still help to close that decision.
Paris: Next story. Sitecore launches Marketing IQ built on Scrunch to draft fixes when AI gets a brand wrong. Sitecore launched Marketing IQ inside Sitecore AI at its symposium on the 11th of November. It uses data on how brands appear in ChatGPT and Gemini to draft content changes, which marketers approve before anything is published. So this data comes from Scrunch, which Sitecore bought in June. The price was reported at about $225 million, which Sitecore declined to confirm, but I think that is well accepted by now. It's included with Sitecore AI CMS and DAM licenses. There is no price, general availability date, or customer data yet. What's really important here, first with this acquisition and now with the integration of Scrunch into Sitecore's Marketing IQ, is that AI visibility is moving into the CMS. So that means that standalone tracking tools now compete with platform bundles. And it raises the question, will the CMS become the natural home of GEO? We'll see about that. I think that that is the vision for Sitecore here.
Paris: Next up, Google publishes a help page on spotting search traffic in analytics, and it leaves out AI features. Google published a help page on identifying Google Search traffic in analytics and server logs. It confirms the referrer shows only the origin with no query. And what's interesting, an interesting omission is that AI Overviews and AI Mode and Gemini traffic are not addressed. So there is still no referrer-level way to separate AI Mode or AI Overview clicks, unfortunately. That being said, the UK competition regulator's December 3rd deadline for Google to provide AI click data is looming.
Paris: All right, let's wrap up with some lightning round. First, ChatGPT Ads adds region targeting within a country and combined product set filters. Well, that's fantastic. Regional targeting in ChatGPT Ads. Great. Next, Profound says that ChatGPT commercial conversations rose 119% from June of last year to June of 2026 with no stated method. That kind of supports the earlier research that we just spoke about. People are more and more using ChatGPT for commercial conversations. Next, Imprint raises $130 million at a $1.9 billion valuation. It offers a server that lets AI agents make purchases. And last, a Google paper on 2.1 million image generation chats shows people appear in nearly 80% of image generation chats. Very, very interesting. So let's end it there. And Friday the 13th, it's been a very, very dynamic week. Thank you all for listening and see you on the next one.