AI Overviews Hit 43%, Search Console Adds Social, and the AI Correction Gap

📬 You like this podcast? You’ll love the newsletter.
Join the weekly The Visibility Report: subscribe

AI Overviews now appear in roughly 43% of Google searches, up from 15% a year ago. But buried in the same dataset is a number almost nobody is quoting: as of May 2026, only about 6.8% of U.S. ChatGPT desktop queries returned an answer that cited anything at all. Citations are growing fast. They're also still the exception.

This week's news update covers four developments that all point the same direction — your brand's visibility is increasingly being determined on surfaces your content team doesn't own and can't see.

• The Similarweb data behind the 43% figure, why the 6.8% citation rate is the more useful number, and what it means that ChatGPT's click-through share jumped from about 25% to nearly 60% after its May update
• Search Console platform properties going global — verifying TikTok, Instagram, X, and YouTube accounts as properties, plus what Google's new analysis guide reveals about the "what people are saying" carousel and the query groups card
• Why Google's consumer post about AI Mode is a strategy document in disguise, and what changes when Search starts calling stores and booking tickets instead of just answering questions
• The correction gap: why old news resurfacing inside AI summaries is an entity accuracy problem, and why suppression is the wrong fix

I'm an AI search visibility consultant and a fractional content strategist for startups and enterprise brands. If that sounds like the kind of help you're looking for, email me at cassie@cassieclarkmarketing.com. 

Or request your 7-Day AI Search Visibility Audit: https://cassieclarkmarketing.com/ai-search-visibility-audit/

Let’s connect:

LinkedIn → Cassie Clark | AI Search Visibility Consultant
Website → https://cassieclarkmarketing.com

Transcript

SPEAKER_00 0:31

Hey, welcome back to Found in AI. I'm Cassie Clark, an AI Search Visibility Consultant, and this is the Thursday news update where we go through what actually happened in AI Search this week and what, if anything, you should do about it. Today is Thursday, July 30th. There were quite a few things actually that happened this week, and we kind of have a theme, which doesn't always happen. The theme is AI Search stopped being a thing that you optimize for and started being a thing that happens to you. Which is a little bit different, just stick with me. We got a number that kind of reframes the entire conversation. We got a measurement tool for content that you don't own. We have Google showing its hand about what AI mode is actually becoming, and we got the first real sign of the entire industry forming around cleaning up what AI systems say about you. There are four stories, it's probably gonna be a little bit longer than normal. Let's get into it. So let's just start with that data because everything else kind of sits on top of that one. Similar Web Publisher 2026 Generative AI Landscape Analysis and TechCrunch picked it up on Monday. I think it was Monday. The headline figure though is that Google AI overviews now appear in about 43% of searches. A year ago that number was 15%. And AI mode grew alongside of it. So roughly 126 million visits in June of last year compared to 279 million by May of this year. And Similar Web also found that the average length of a Google search is going up, which is people really just replacing that three-word keyword strings with full conversational questions. We've talked about this on the show before, you might already be seeing this in your Google Console data. Now there's really just data behind it. Now, quick little note before we get going. Citations inside AI responses are up more than fivefold over the past year. Great, that's that's good news. But yet, as of May 2026, only about 6.8% of US chat GPT desktop queries returned a response that included any citation at all. That's 6.8%. Again, third-party measurements, but that's a number that really sticks out. AI answers are becoming that default surface, and the overwhelming majority of those answers are not showing anyone where that information came from. Which means the model is just drawing on what it knows about your category, your competitors, and you, and then not just giving a link or attribution, it's just saying the thing with out of click anywhere. Now I've been constantly saying this, and being mentioned and being cited are not the same event, and most of what's happening to your brand right now is the first one. There's just now a number attached to it, it's 6.8%. If your entire AI search visibility strategy is get cited, you are optimizing for that 6.8%. I'm gonna keep saying it because it's such a tiny number. The other 93%-ish is where your positioning is being decided. And it's being decided by what the model has absorbed about you from everywhere else, not just your own brand blog. Now, there is an encouraging data point to balance all this out. After OpenAI's May 7th search update, the share of US Chat GPT desktop visits that landed on an actual web page went from about 25% in March to nearly 60% by the end of May. So that means when their links are there and they're prominent, people do actually click on them. Though the link economy isn't really dead, it might just be on live support. But it's conditional now. Okay, moving on. We've got Google to talk about. Yesterday, July 29th, Google rolled platform properties out globally in Search Console. They announced it back on July 7th and it's been trickling out since. And as of yesterday, it is now available to everyone. If you missed the original announcement, platform properties let you verify an Instagram, TikTok, X, or YouTube account as a Search Console property, and then you can see how that content performs in Google Search, Discover, and Google News. You get information about clicks, impressions, and queries, which is the same metrics that you've been staring at your website for a decade now. It is the first Search Console property type that you can verify without actually owning a domain. That's not really a small thing, that's kind of a big deal. And then they also announced that they were pushing out a guide that goes along with it. So here's about that part. The guide is actually how to analyze social and video content performance. I read most of all of it this morning. I was skimming pre-coffee, so you know how that goes. But it's really more of a substantial kind of update that they give than there's their typical launch day help page that they roll out. And there are four parts of this that actually really matter. So the first one, this tells you where this content is actually showing up. So the guide opens by naming three search surfaces where your social and video content can appear. That's in the short video carousel, it's in the latest post carousel, and it's under the what people are saying carousel. Now I want to call out that what people are saying part because that's a third-party opinion service. It's basically Google just assembling what other people are saying about a topic and then putting it on a results page. If your content is landing there, you're not really being cited as a source, but you're just being used as evidence. And you can measure that now. The second thing is those query groups. Inside the insights report, there's a card that takes the search terms that are driving traffic to your social content and then it clusters them into themes, and then it sorts some of the themes from top, trending up, and trending down. Now, if you're a longtime listener of the show, you might say, hey, that sounds familiar. Back in February, I covered Bing's grounding queries, and what I said then was that the useful part about all of this wasn't really the keyword list that they were giving. It was just really getting to see how the system categorizes you. This is kind of that, but it's for content that you don't host. Google is showing you which themes it associates with your off-site presence, and then whether that association is getting stronger or weaker over time. That's really entity data with a trend line on it. So it's really nice to have. The third thing is that Google really poses the gap question itself. In the performance report section, the documentation lists sample questions you can answer with the data. One of them is roughly, are there search terms you'd expect to be getting traffic from that you aren't? Google is telling you in a docs page to audit for the absence, not just the presence. Now the fourth thing is the test mechanics, which are unusually specific. There's a 24-hour filter, so the day that you maybe try that again. So the day after you post, you can see which search term is suddenly bringing people to that particular piece of content, and then you can act on it while it's hot. There's comparison mode with URL pattern matching, and the guide gives you actual strings. So on YouTube, compare URLs containing watch against URLs containing shorts. On Instagram, you can compare P against Reels. There's a playlist walkthrough where you filter for the word playlist and then pull CTR in average position. And then the piece that I'd actually build the workflow around is annotations. The guide specifically tells you to drop an annotation, which is just essentially a bookmark in your data, where you rewrite a YouTube title or a TikTok caption and then compare performance on either side of that. If you really think about it really hard, Google is basically telling you to run controller before and after tests on the platforms that you don't own, and then watch what it does to your Google search performance. That's a feedback loop on content that you don't own that did not exist eight weeks ago. Here is my take on all of this. For as long as I've been doing all of this work, the honest answer to well, how is our YouTube channel affecting our search visibility? Has been it helps. We can see some of those data on YouTube, but that's about as far as we can get. We're not really sure how it's showing up in Google other than we know it's showing up. That measurement was kind of a black hole, and everybody just kind of agreed to live with it, which is kind of like what we're having to do with wonky. I say wonky because some of the AI search visibility measurements that we need are wonky, and we're trying to piece them together. This also has created a very specific organization problem, which is the one that I'm running into with almost every engagement that I have with brands. Social owns the account, SEO owns the reporting, nobody really owns that overlap in between. So the overlap doesn't get optimized, it doesn't get measured because it can't be measured, and because nobody can pull that number, that just kind of just gets forgotten. But now, with this, somebody can pull that number, which means that somebody now has to own it. And that brings me to the FSA framework. This is your first time hearing that. That's freshness, structure, and authority. And with this new thing, freshness, you can watch older content regain traction and then decide what to update or expand. Structure, you can test whether a clearer title or better organized description makes content more retrievable. And authority, oh, you're building evidence that your expertise shows up consistently across surfaces that you don't actually host. That is all entity building. That is what I talk about a lot. Okay, now third thing is that Google published a post for users that actually kind of reads as a strategy document. On Tuesday, Google's blog ran a piece called Five Ways AI Mode and Search helps you enjoy the real world. It is a light consumer post about, you know, touching grass, which I think more of us need to do more often. And it also mentions that searches for like adult tennis lessons and trail running and run clubs are way up. Searches for how to do a digital detox are up 110% since the start of the year. That is not lost on me. I feel like everyone should do a digital detox at least once. That's another topic. Anyway, if you skimmed this blog post, I get it. Or if you didn't read it, here's what it says. One, AI mode connecting to your Google Calendar through personal intelligence is helpful so that it can recommend classes that fit your existing schedule. Two, there's shopping inside AI mode, which they've been moving towards for the last two-ish months. They've been building out that infrastructure. It can filter by nearby for local stock. And Google, you can ask Google to call stores on your behalf to confirm availability. Three, you have Canvas building you a custom strategy guide and then building a simulation to practice against it. Four, you can use it to find and book event tickets. So when you give it your budget and then quantity, and it just does it for you. And five, those connected apps. You can tell AI mode to make you a party fire and then have Canva generate the design inside your search results. So those are five examples, and four of them are search doing something rather than search telling you something. Now, there are two takes takeaways here. First one is kind of tactical. Google publishes example prompts in these consumer posts. And if you read it, you'll find that they are actually real prompts in natural language at real length. For example, can you help me find affordable hiking boots with good ankle support? That's a demand sign query data, and it's straight from the company that has it all, published for free. And most marketers are probably gonna read this post, go right past it, and think, ah, that's lifestyle content. I am begging you, start reading these posts as keyword research and apply it to your strategy. That's strategy information in a lifestyle post, but it's all right there. Second one though is kind of bigger. When search is placing the call, like actually placing the call for you, booking the ticket, checking whether the boots are in stock at the store four miles away. The question is no longer does my page rank for this keyword? The question now becomes is my inventory, my pricing, my hours, my availability legible to a system that's acting rather than answering. That's not a content problem, that's a structured data problem, a feed problem, a merchandising problem, and an ops problem. And that all of that sits nowhere near your blog. So, which is the argument that I keep making on the show roughly every other week. So I'm not gonna relitigate it. But this is one more surface where AI visibility turned out to be a whole organization question and not just a content team question. And if we're honest, Google has been telling us for a hot minute that this is where search is headed. They've been talking about this for the better part of a year at this point. Okay, so let's move on to the fourth story, which is not really Google related. I not really, but earlier in the week, an online reputation management firm called removenews.ai, built by a company called Reputation Resolutions, they put out a press release saying inquiries are climbing sharply because of a specific new problem. Old damaging news coverage resurfacing through AI search. Now, I talked with a brand earlier in the year who mentioned that some of their old press releases were pulling in some of their past positioning and they were wondering how to get rid of that. This is sort of similar, but with just damaging news coverage. Standard disclosure, this is really a vendor announcement about a vendor's own demand, so treat that accordingly. I'm not gonna pass off their groove claims as findings, that's just their numbers, but this is really interesting because the mechanic they're describing is real and it's worth a few minutes of your attention. Their argument kind of runs like this In classic search, a negative article was a link on a page. You could see it, you can measure its position, you can work on pushing it down. If you get it down far enough, no one's gonna see it. In AI search, though, the article doesn't just stay a link, it gets synthesized into a narrative about the person or the company, then it's presented a summary, delivered with confidence, often without the user ever visiting the source. And we might not actually see the source. Think about those numbers I gave you at the beginning of this episode. The part that actually really matters here though is that AI systems gain access to information much faster than they build mechanisms for correcting or moving it. Google's policy has always been that content doesn't come down just for being negative. AI engines have fewer formal channels for that, and then I'm not really sure that they're gonna make those channels to begin with. I mean, now that I think about it and I'm saying it out loud. So there really is a gap here. That damage propagates at retrieval speed and remedy moves at well, whatever speed the correction process exists. And mostly one really does it. You have to it takes a while to do that. Here's what I'm thinking about all this, though. I don't really think the answer is suppression. Suppression is reactive, it's awfully expensive, and in a synthesis environment, you're playing a whack-a-mole with an ecosystem that you don't really control. The answer though is corroboration. Enough current, consistent, well-structured sources saying that accurate thing, and that's what the model synthesizes is, and then it defaults to accuracy. That's kind of the weight of evidence support that just kind of cancels everything out. It's not really a reputation tactic, it's more entity work. And it's the reason that I have clients build a strategic source of truth and then have them refresh that every few months just to make sure that it is as truthful as possible. And it's why entity accuracy testing belongs inside of an AI visibility audit. Those audits should not just be citation counting. I have seen some of those audits, and that's exactly what they are. They're just auditing for presents. But fewer are auditing for accuracy. Accuracy is the one that bites you if it's not correct. If you have never opened four engines and then asked them to describe your company, your leadership, and your last three years, please go do that today. It's gonna take you about 20 minutes. I have yet to run that test for a client and not find something worth fixing, so it's helpful to know what's out there so you can be aware of it and then do your best to fix it. Again, I know this one's kind of long, so let's put it all together and give you the TLDR summary. AI answers are now the default service, and most of them are not citing any brands at all. Google just handed you a way to measure content on platforms that you don't own. Search is becoming something that acts, not just something that answers, which is what, again, Google's been saying for the better part of a year. And there's now a market forming around fixing what AI systems say about you because the systems can say it much faster than anyone can correct it. All four of those kind of point in the same direction. Your visibility is being determined across surfaces that your content team doesn't own and maybe can't see. This week, though, one of those surfaces got a dashboard and the rest still didn't, but we are getting closer. Okay, that's it for this week. If you want this in writing every Friday with some deeper insight, the visibility report is the newsletter you want it to be on. Subscribe, the link is in the show notes. And if you're looking at your own numbers, wondering how much of this actually applies to you, that's what the AI Search Visibility Audit is for. Details for that are in the show notes, or you can head over to CassieClark Marketing.com to get started. I'm Cassie Clark, an AI Search Visibility Consultant. I will see you on Tuesday. Until then, stay visible.