Search moved fast in the first half of 2026. This is your halftime briefing.
Two core updates. AI visibility reporting finally arriving in Bing and Google. A Google I/O that rewrote the rules. ChatGPT referral traffic spiking overnight. And an AI commerce landscape that looked completely different by June than it did in January.
For the Search 2026 Halftime Show, Cara Corbett is joined by headline guest Michael King (Founder & CEO, iPullRank) for a live update on the biggest shifts in search in H1 2026 and what it means for your agency and your clients going into H2.
We cover:
→ The biggest H1 2026 shifts in SEO and AI Search
→ Key predictions for H2 and how agencies need to prepare
See the recording below.
Click here to see the presentation slides.
Webinar Transcript
00:00:00 → 00:00:05 Cara Corbett: Welcome, everyone. Really glad you could join us today. I’m Cara Corbett, VP of Partner Growth at SUSO Digital.
00:00:06 → 00:00:21 Cara Corbett: We’ve got a really special one for you today. This is the Search 2026 Halftime Show: What Happened and What’s Next. The reason we did this one is because one of the biggest things I hear from our agency partners is: things are moving so quickly in AI search. How do we stay up to date with all the changes, and who are some good people to follow to keep up with all this news?
00:00:22 → 00:00:33 Cara Corbett: So today I’m super excited, because not only are we going to dig into all of the major changes that have happened in search so far this year, but we’ve also got a very special guest joining today to help me do that. I’d love to give a warm welcome to Mike King.
00:00:34 → 00:01:01 Cara Corbett: He’s a man who probably needs no introduction to most people in the SEO world. Mike is the CEO and founder of iPullRank, which is a New York City-based SEO agency. He’s also the two-time winner of Search Marketer of the Year, and he’s the founder of SEO Week, which is probably one of the industry’s biggest and best SEO conferences out there. I got to attend it this past April. It was incredible. I even saw Mike freestyle with his eyes closed.
00:01:02 → 00:01:35 Cara Corbett: Something I want to highlight about Mike is that he comes from a computer science background and really looks at everything with an engineering-type lens. He’s the guy that’s actually taken the time to read a lot of the patents from Google and understand how search algorithms work at a very technical level. And so what Mike does at his agency is not called SEO or GEO; he calls it relevance engineering. Mike, do you want to explain your take on this?
00:01:36 → 00:02:00 Mike King: Sure. The way I look at it is a lot of SEO is somewhat outmoded, and there’s a lot that we can understand about how things actually work by paying attention to what these platforms are doing. There’s open source equivalence to everything that they’re doing, and you can take that stuff apart and figure out how things actually work and then apply that to the marketing work that you’re doing. Relevance engineering is effectively the confluence of artificial intelligence, information retrieval (which is the science behind search), user experience, content strategy, and digital PR.
00:02:12 → 00:02:28 Cara Corbett: I love it. I also want to pump your tires a bit. Mike’s also the guy who publicised query fanout. So if you’re in the business of doing GEO for your clients, you should probably already know what that is. But if you don’t, it’s one of the methods that LLMs use to go and retrieve information for their responses, where they take your query and break it down into multiple sub-queries in the background. You can read more about that in Mike’s AI Search Manual, which I believe is one of the most relevant and comprehensive resources for learning and doing GEO.
00:02:44 → 00:03:03 Cara Corbett: The final thing I’ll say about Mike is that he’s got a lot of strong opinions and they often spark some debate in the search community. So I’m really excited, Mike, to get your unfiltered takes on some of the changes that we’re going to walk through today. Do not hold back; it’s what’s going to make this webinar so great. Did I miss anything about you that you want to share with everyone?
00:03:04 → 00:03:05 Mike King: No, I think that was a good summation.
00:03:07 → 00:03:58 Cara Corbett: Great. So fairly simple webinar today. First, Mike and I are going to walk through the key events that happened in the first half of 2026. We’re going to group these into five different topics: one, volatility; two, reporting; three, ChatGPT; four, Google I/O, which is their annual developer conference, and regulatory pressures; and five, the agentic web. That should take up the majority of the session, but there will also be a quick regroup on what to focus on, and we’ll get our thoughts and predictions for the second half of the year. And as always, if there are any questions about anything we go over today, go ahead and use the chat box on the right side. We’ll have time to answer questions at the end of the session. Everybody will also receive a copy of the presentation within the next couple of days. Okay, so let’s jump into it.
THEME 1: VOLATILITY
00:03:59 → 00:04:12 Cara Corbett: We’re kicking it off with volatility. The first half of the year was no walk in the park. We saw tons of changes on both the traditional SERPs with core updates, but also with the way that AI tools were operating.
00:04:12 → 00:04:41 Cara Corbett: Starting in February: Profound came out with some cool research that found LinkedIn was the number one cited domain for professional queries across every major LLM. A separate Semrush study also corroborated this evidence; they found that LinkedIn was the second most cited domain across their data set. This is a bit of a switch-up from the narrative around earned media’s dominance of citations, and it really just highlighted another surface that we need to think about when we’re optimising for AI search. This makes a lot of sense because LinkedIn profiles and company pages carry high trust with LLMs. Mike, what’s your take? Are you seeing LinkedIn as a highly cited domain across your client base?
00:04:56 → 00:05:38 Mike King: Yeah, absolutely. The way I think about this is it’s really about your content ecosystem. I talk about it in terms of your shared, owned, and earned properties. LinkedIn being one of the shared properties, what we see is that LinkedIn posts (when you’re writing articles on LinkedIn) work like gangbusters. I’ve seen clients publish things and they become the first part of the response in ChatGPT. So it’s still valuable; it’s still something that should be a part of your bucket of tactics. And it tends to be one of the quicker ways to get in there. I would also say YouTube also works really well too.
00:05:41 → 00:06:10 Cara Corbett: Yeah, YouTube especially with Google Search, obviously because they own the platform. All right, in the following month, Google began rolling out its first core update of the year. This rollout took two weeks to complete and officially wrapped up on the eighth of April. As always, Google didn’t really get into the details about what they’re changing, but they did say this was another update designed to better surface relevant, satisfying content for searchers.
00:06:10 → 00:06:32 Cara Corbett: This one in particular presented a lot of turbulence within the SERPs. According to SE Ranking analysis, nearly eighty percent of the top three results shifted, and almost one in four top-ten pages fell out of the top one hundred. Many other analyses saw that niche sites with original research and insights held up, whereas bigger domains that were running AI rewrites of old content got absolutely crushed. None of that surprises me. Mike, I know this one was a couple of months ago, but what’s your take, specifically around the importance of unique, or as we call it, non-commodity content?
00:06:43 → 00:08:58 Mike King: Sure. For us, only one of our clients got hit by this, and it was the client we’ve been working with the longest. We actually helped them recover from various algorithm updates over the years. But this one seemed to be a fundamental difference in what Google wanted to show. They’re in the automotive space; they do car reviews. And Google really shifted to showing more of the local dealerships and manufacturers, but showing a lot less of the review sites. In a situation like that, there’s not really anything you can do. Your business model is around reviews and Google just decided they don’t care about that anymore.
Beyond that, we didn’t see much volatility for our clients. But what I will say is that there’s a big gap in the SEO space around this sort of thing. You’ve got tools that are tracking volatility broadly, and then they make these assessments after the fact. They say, OK, these sorts of sites got hit, so this is what happened. That’s not necessarily true; that’s correlation, not necessarily causation.
So what we do for this sort of stuff is we track rankings, but on top of that, we’re also storing copies of pages for our clients and competitors and extracting features from them. There’s an array of data points we look at, including link metrics and other external factors, so we can do regression models when these algorithm updates roll out and see precisely what Google wanted more of.
00:09:00 → 00:09:08 Cara Corbett: Great. Giving us some insight into the secret sauce, building all of these fine-tuning points for algorithms. I love that.
00:09:00 → 00:11:28 Cara Corbett: Let’s move over to the Google Search Console impressions error. This was a big one too. On the third of April, Google confirmed they had a logging bug in Search Console that had been over-reporting impressions since the thirteenth of May 2025, which is almost a full year. How do you go that long without noticing? Clicks were not affected, but because impressions were inflated, click-through rates appeared lower than they actually were throughout that period. And Google said historical data would not be corrected. So any click-through rate analysis or benchmarking from that window was effectively unreliable. Mike, do you think this was a genuine accidental bug, or could there have been something more cheeky going on with Google to make themselves look better?
00:09:43 → 00:11:28 Mike King: I think it was an actual bug. I think that they have been a bit of a mess for the last two years because they’ve been so heavily focused on AI for everything. The resourcing for these teams; GSC has never been a largely resourced tool. A lot of the features and functionality are so far behind what you get out of GA and things like that, and they’ve had indexing problems. There’s been so many problems with Google Search over the last two years, so I think this is legitimately a problem that happened.
But it also brings up the point that accuracy in analytics is not what people believe it is. You look at things as a source of truth and they’re not always right. Even if you look at regular analytics, no matter what you use (whether it’s GA, Adobe Analytics, or whatever), they’re not accurate, they’re precise. There are a lot of situations where a user may not have JavaScript on, or they may be blocking cookies, so they get double-counted or not counted at all. You can’t really trust any of these data sources to be completely accurate. You can only count on them to be precise. But in this case, we found that GSC wasn’t even precise in that time frame.
It’s good to keep a pulse on this sort of stuff so you do have an understanding, because a lot of people have been reporting that impressions jumped and were assuming it was a function of all the generative AI stuff. But no, it was just a bug. So it’s good to know that, so you can update your assumptions.
00:11:28 → 00:11:37 Cara Corbett: Fair. And I wonder how big of a difference it would have made to that great decoupling as well. Like, is that whole thing a farce? Crazy times.
00:11:40 → 00:12:22 Cara Corbett: Moving on. The ChatGPT referral jump happened in May. On May seventh, OpenAI made a quiet algorithmic change that now automatically links brand mentions to their website’s homepage directly within the ChatGPT response. So now when a user searches for specific brand recommendations, they can click through directly to the website, which was not something they did before this. Similarweb tracked the result: ChatGPT referral traffic jumped approximately one hundred and fifty seven percent week-on-week. I don’t know how much merit the zero-click argument has anymore. This is the clearest example of why monitoring AI platforms actively really matters. The impact was real and measurable, but there was no official announcement from OpenAI.
00:14:08 → 00:14:42 Cara Corbett: On May twenty-first, we saw another core update, less than two months after the March update, which was completed on June second. This one ran simultaneously with Google I/O, which was deliberate: it specifically targeted the scaling problem of programmatic AI-generated web spam. The clearest signal in this update was which kinds of sources Google now chooses to show more of. Reddit’s share of the top three positions rose to 10.24% after May. Two core updates in under eight weeks made it a particularly difficult reporting period for agencies. Anything jump out to you around this time, Mike?
00:14:43 → 00:17:10 Mike King: So I believe this one and the next one are actually tied together, and I think there’s a bit of a misunderstanding of how algorithm updates roll out. It’s not just one single thing that they’re adjusting; it can be a variety of teams all pushing something around the same time. And just like any other software company, they may roll components of it back and further adjust certain components.
A lot of people in the SEO space love to talk about this idea of what they call Mount AI, where you have this huge hockey stick of growth because you’re using generative AI content across millions of pages, and then it peaks and drops off. That’s not just a function of it being generative AI content, because Google cannot reliably detect generative AI content with reasonable confidence. They have a lot of false negatives and false positives, so they can’t use that as a single signal to demote content. Instead, they’re using user signals to indicate whether or not the content is good.
The reason you see that spike is because when Google sees new content, they compare it to existing content and they have a site quality score. They say, OK, this looks like content we rate a six; let’s see how it performs. The thing about generative AI content is that users can tell. It typically has a high bounce rate and low time on site. These are the equivalents of signals used in information retrieval to determine whether something is good. This is a UX problem. It’s not a problem of having used generative AI; it’s a problem of having slapped it on a website in a way that nobody wants to actually read. They’re not getting any value out of it. So I think these two updates were them going after that, and then refining it a bit more as well.
00:17:11 → 00:17:46 Cara Corbett: OK, cool. Speaking of spam, Google had a June 2026 spam update on the twenty-fourth, which ran for roughly two days. While no new spam policies were announced, their existing spam policies already explicitly cover attempts to manipulate generative AI responses. These were the content types that took a hit: websites churning out massive amounts of automated, unoriginal AI content. But I think that ties in nicely to what you were just saying, Mike, about it being an engagement signal versus them actually being able to tell it was AI-generated.
THEME 2: REPORTING
00:17:47 → 00:18:08 Cara Corbett: All right, let’s move over to reporting. We saw some of the biggest changes ever this year, especially when it comes to measuring AI search visibility, even if it’s still not where we want it to be.
00:17:56 → 00:19:24 Cara Corbett: We’ll start with Bing, because in February, Microsoft launched its AI Performance report in Bing Webmaster Tools, which was the first first-party AI search reporting tool from any major platform. They came out with something before Google did. Obviously ChatGPT doesn’t give us any information. So this tool provides the data to help validate which pages and content are being used as citations in real Copilot AI answers. It also gives information around the key grounding terms, in other words those query fanouts that AI uses to retrieve content that was cited. This was officially the first tool of its kind to help us spot opportunities to improve clarity, structure, or completeness on pages that are indexed but less frequently cited. And they even expanded their features: fast-forwarding to June, they added four new capabilities announced at SEO Week: intents, topics, citation share, and compare. So if you’re working with clients and want to get real insights, definitely recommend checking it out in Bing Webmaster Tools.
00:19:02 → 00:19:48 Cara Corbett: Moving forward into May, Google Analytics added a dedicated AI Assistant channel that automatically identifies and labels traffic arriving from AI tools like ChatGPT, Gemini, and Claude. So if you or your clients have GA4, it doesn’t require any additional setup. It already applies to all existing GA4 properties, so you should be able to see exactly which AI tools are sending referral traffic to your site now. Previously, you would have had to do it with a regex code that somebody would have had to send you. Something worth noting: Perplexity is actually absent from this reporting. That referral traffic still lands in the referral segment, so you’re still going to need to do a custom channel group if you want to catch that one.
00:19:49 → 00:20:26 Cara Corbett: Moving on to May, on the e-commerce side of things, Google announced native AI reporting capabilities for Merchant Center, which will give us a taste of how products are surfacing in Google’s AI-powered shopping results. More specifically, it will report on things like share of voice (so how are you doing compared to your competitors), shopping funnel performance, and product term and attribute insights. Unfortunately, this isn’t fully live yet. It is being rolled out over the coming months, but it definitely gives us signals as to where Google is heading. Have you seen any of these updates yet, Mike, in Merchant Center?
00:20:28 → 00:20:35 Mike King: No, I actually don’t live in Merchant Center anymore. It’s been a long time.
00:20:33 → 00:20:38 Cara Corbett: Fair enough. Leave that to the team, I guess.
00:20:36 → 00:21:03 Cara Corbett: Moving over to Google Search Console: on the third of June, they launched a dedicated generative AI performance report. So for the first time, we’re able to see our impressions inside AI Overviews, AI Mode, and Discover’s AI features, broken out from regular search data. What’s your take on this one, Mike?
00:20:53 → 00:20:54 Mike King: That report is a joke.
00:20:57 → 00:21:03 Cara Corbett: I was going to ask you about this. So they don’t have click data. There’s no query data yet. They’re not really useful at the moment.
00:21:04 → 00:22:23 Mike King: I want to give a shout out to Krishna Madhavan from Bing. He is fighting the good fight. The reports that they’re giving in Webmaster Tools are fantastic, and it really just reinforces a lot of the things I’ve been saying about how things work and what you need to be focusing on. I love that what he’s doing is pushing everyone else forward. I don’t think we would have ever gotten this GSC report had Bing not created one first. And it’s my hope that they then add these similar features.
But what Google has been doing is saying, hey, you don’t have to do anything special for AI search; just do what you always do for regular SEO. But everyone else that actually creates these platforms is saying the opposite. They’re saying query fanout is important, all these different nuances of how the index is used are different. And you’re getting that directly from Bing, not from some small company that no one’s heard of. It’s not like Neva (rest in peace to that search engine); you’re getting it from the long-held competitor, and it just reinforces everything that we need to do in order to get the visibility that we want.
00:22:25 → 00:22:29 Cara Corbett: Nice. That covered all my questions for you on that one.
THEME 3: CHATGPT
00:22:30 → 00:22:33 Cara Corbett: Moving on to the next segment, which is all about ChatGPT.
00:22:34 → 00:23:12 Cara Corbett: So we kind of alluded to this already, but of course ChatGPT did something we were all anticipating: they officially launched advertising on the ninth of February. Currently, ads show only to Free and Go tier users; paying subscribers don’t see them. The platform launched at a sixty dollar CPM with a two hundred thousand to two hundred and fifty thousand dollar minimum spend, so this was pretty much invite-only for large brands and holding company agencies. That floor dropped fast: by April it was fifty thousand dollars, and on May fifth, OpenAI eliminated the minimum entirely and opened a self-serve ads manager to all US businesses, slowly rolling out to other countries as well. Is that something you guys do at iPullRank, Mike?
00:23:15 → 00:23:24 Mike King: No, we don’t. We have partners that do that sort of stuff, but we do report on it. Profound has great reporting; they have a whole ads section now, which is cool.
00:23:25 → 00:24:51 Mike King: But what I do want to say is that I don’t necessarily believe these ads are the right modality for chat. I think the whole pay-per-transaction model probably makes a lot more sense when the user wants the agent to go out and buy something on their behalf. I think that’s the model that’s going to catch on, because these ads are just not good. And then there’s the big brand safety problem that’s not answered here either, because if someone asks something bad about your brand and then your ad pops up, you don’t want that next to ChatGPT saying it read a Reddit thread that says your brand sucks. There’s no indication that they’ve solved that problem. So I just don’t believe in these ads until I’m proved otherwise.
00:24:55 → 00:25:40 Cara Corbett: Right. And I haven’t dug into this, but is it still a cost-per-click model? Like, if your ad shows up next to something negative about your brand, are you potentially paying for that? Well, it’s not even cost-per-click; it’s cost-per-impression. So you can be paying for impressions that are saying your brand sucks. Yikes. And there’s no reporting comparable to what you’re used to in Google Ads, is there?
00:25:40 → 00:26:10 Mike King: Yeah, there’s no real visibility yet. Like, there’s no equivalent of what you’re used to in Google Ads. There are just so many variations in the ways things show up that they haven’t built it out yet.
00:26:11 → 00:26:38 Cara Corbett: For the first time since their launch, ChatGPT’s global market share among all LLMs dropped below fifty percent in March 2026, according to analytics firm Sensor Tower’s State of AI report for 2026. By May, it had settled at 46.4%. Claude, Gemini, and Perplexity all gained ground in this period. I would say a big reason for this was OpenAI’s affiliation with the Department of Defense, and a lot of people had some problems with that. But also at this time, Claude really started to rise in popularity. Mike, do you prefer Claude?
00:26:20 → 00:26:52 Mike King: Yeah, Claude’s just better. It’s so good. I mean, everyone got better. Even Gemini had gotten better around that time. I wouldn’t say they’re a leading provider right now, but Claude is just dramatically better, and Claude Cowork is so valuable. And then with the announcement of Claude’s Skills feature, and then you’ve got OpenClaw. So there’s so many reasons to use Claude as your daily driver, as your operating system for this sort of stuff.
00:26:53 → 00:27:14 Cara Corbett: Yeah, totally. It’s been super helpful. And I saw your presentation in Miami where you gave a great talk on how you’re using that and other things to create efficiencies in your agency, which was really cool.
00:27:19 → 00:27:47 Cara Corbett: On the fourth of March, ChatGPT pulled back from its Instant Checkout feature just five months after it launched, due to way too many infrastructure hurdles. E-commerce is messy. The system struggled with basic retail needs like promo codes, shipping choices, and real-time inventory tracking. Plus, only a handful of merchants actually signed up. Most importantly, user behaviour doesn’t really match that vision. Most of us love using ChatGPT to research and compare products. When it comes time to buy, a lot of us still prefer checking out directly on the brand’s website because you know you’re going to get a secure, seamless user experience. So instead of trying to be a digital marketplace, ChatGPT is really leaning into what it does best: acting as your ultimate product discovery and comparison assistant. Maybe one day they’ll get there, though.
00:28:08 → 00:28:50 Cara Corbett: All right, and speaking of AI shopping: a large-scale study done by Peec AI, analysing over forty thousand ChatGPT carousel products, found that eighty-three percent matched organic Google Shopping results. I’m sure we’ve all heard that ChatGPT runs on Bing. It’s not the case for shopping, it seems. Semrush also ran a separate experiment and came to the same conclusion. So if you want your product to show up in ChatGPT’s shopping carousel, you need strong structured data, a clean product feed, and of course domain authority. We actually did a whole webinar on this a few weeks ago. There’ll be a QR code at the end of this presentation if you want to scan and rewatch that. Anything you want to add to this section, Mike?
00:28:49 → 00:28:51 Mike King: No.
THEME 4: GOOGLE I/O AND REGULATORY PRESSURES
00:28:52 → 00:29:05 Cara Corbett: OK, moving over to Google I/O and then regulatory pressures. I know Mike’s going to have a lot to say about this.
00:28:56 → 00:30:00 Cara Corbett: We’ll start with March, when the headlines were rewritten. Google confirmed they’re experimenting with AI-generated headline rewrites in search results, basically deleting your carefully crafted title tags and replacing them with text written by their own AI. Google claims this is just a small, narrow test, but I would take that with a grain of salt. This is exactly the same kind of language they used right before they made AI-rewritten headlines permanent in Google Discover. Why does this matter? It completely strips away your editorial control. We’re not talking about Google just pulling words from your page anymore; this AI can actually fabricate entirely new phrasing, sometimes completely flipping the sentiment of an article or stripping away critical context. And Google isn’t displaying these AI headlines with a label. So if AI writes a terrible, misleading headline, your audience is going to blame the brand, not Google’s algorithm. This is particularly a problem for news publishers and outlets that rely on headlines to hook readers into their story. Mike, I’ve got a question I’ve gotten before from other agencies in our Partner Club: is it still worth writing your title tags and meta descriptions in this era if they’re just going to be rewritten by Google?
00:30:12 → 00:31:30 Mike King: So they have rewritten both of those for a long time. From my perspective, the meta description is not really worth writing because they’re going to pull what’s most relevant from the page, and that’s going to yield a higher click-through rate. But as far as the page title goes, what you should be doing is ignoring the whole sixty-character thing that’s been the quote-unquote best practice for a number of years. We’ve done tests a number of times where we just take the top three keywords that are performing in Google Search Console and append them to the end of the page title. It always yields like a ten to twenty percent increase in click-through rate, because Google just has more to choose from when it rewrites it. You don’t have to necessarily take the approach that we do, but you should provide more information about the page in the page title than you’re used to, so it has more to choose from.
Now, what we don’t know is whether they’re just summarising the page and ignoring everything in your page title. My experience has been that they reduce the page title, not that they summarise the page. But if they’re moving to summarising the page, then you’re just going to want to be more focused on what you’re talking about on the page itself.
00:31:32 → 00:31:53 Mike King: And I think it was on one of your webinars, if I’m remembering correctly, that we were talking about how the page title is actually like a first entry point for AI, right? When they’re trying to deduce which page to reference when doing a query fanout and finding the chunk of content they want to use. So if they’re rewriting your page title, are they accessing the Google page title or your original version?
00:31:55 → 00:32:56 Mike King: So it depends on the source. If we’re talking about Google specifically, then Google is going to pull from their own search results, so whatever they have in their search results is what’s being used. Whereas if you’re talking about ChatGPT or Perplexity, there are different data sources. In some cases it could be Google, in some cases it could be Bing, in some cases it could be Exa Search. And they all provide a different level of content. Exa Search specifically will extract like the first hundred words from the page. So it’s not just the page title that they’re getting. The problem is you don’t necessarily know which source they’re getting it from. So you want to provide as much as you can within the environments and the tags that you can.
00:32:56 → 00:34:28 Mike King: So there’s no easy answer to that. The best way to do it would be a more technical approach, meaning that you detect what user agent is accessing the page and then provide whatever is most valuable to them. So for Google, you would give a longer page title so they can choose from it. If instead it’s Exa Search, you’d give a reduced page title that is more aligned with the query you’re targeting. It needs to be a nuanced approach rather than a one-size-fits-all. And in order to serve a different experience to different agents: in WordPress, for example, there are user agent detector plugins where you can specify a different version based on that user agent. You can say: for Bing, here’s my page title. For Google, here’s my page title. For ChatGPT, here’s my page title.
00:34:30 → 00:34:45 Cara Corbett: Would that not be considered cloaking in Google’s eyes?
00:34:33 → 00:34:54 Mike King: So cloaking has a very narrow definition: when you show something different to Google than what you show the user. That’s typically about what is on the page itself. Just having a different title tag, I don’t think would be enough to trigger some sort of filter.
00:34:54 → 00:35:25 Cara Corbett: OK, cool. Lots of knowledge, love it. Now, in May, Google published official documentation titled “Optimising Your Website for Generative AI Features” on the Google Search Central blog, under a brand-new Generative AI Fundamentals navigation section. This documentation largely stated that good GEO is just good SEO, and focused on the importance of non-commodity content and clear technical structure. They also included a section around myth-busting where they denounce tactics such as LLMs.txt as well as chunking. Now Mike, I know you’re a large believer in chunking. There’s a whole section on it in your Search Manual. And I know that you don’t always think Google is truthful in their advice. What’s your unfiltered take here?
00:35:40 → 00:40:08 Mike King: So I wrote a blog post about this and I think it was titled something to the effect of “Google’s guidance is naive and self-serving,” because that’s exactly what it is. Sure, this is guidance for the Google environment, which I don’t think is accurate either. But it’s really them trying to regain control over the fact that we are moving from a Google-shaped web to an agent-shaped web. The last twenty-five years, people have built websites in alignment with best practices for what works for Google. But now that there are a multitude of players treating the web in a different way, we are all moving in the direction of creating websites that work well within these agents. And so when you say something like LLMs.txt doesn’t matter, that’s not true. It’s not true because everyone is using Claude now, and Claude uses it. And in fact, if you look in the agentic guide in another section of Google, they say you should have an LLMs.txt. You’re spoiling what’s coming up.
So on chunking: there are really two definitions. One is what the actual systems do when they decompose your page to understand it. They break it into a series of what are called chunks. That terminology comes from information retrieval, specifically retrieval augmented generation. They break down your page into chunks, measure those chunks against the synthetic queries in the query fanout, and then say, OK, this is relevant; feed it to the language model. The second definition, from a marketing perspective, is about how you structure your content so it works in that environment: using headings, having paragraphs that are about one thing rather than two or three things, because it’s easier for it to perform in those environments. But it’s also easier for a user to consume when you do it that way.
Google says you don’t have to do that. But if you read their research about what they actually do, everything points to these atomic units of content that speak about one thing. Google will tell you to do a thing when it works for them. Like they told everyone to switch their site to the secure protocol and said they’d give you a rankings boost, so then everyone did it. And you didn’t get a rankings boost because everyone did it. But it made things better for Google. Same thing with Core Web Vitals: hey, make your site faster and we’ll give you a rankings boost. No one got a rankings boost, but everybody made their site faster, and that made Google’s crawling costs go down. Same thing is true of structured data: hey, give us all the structured data. We can understand your pages better, we can rank you better. Everyone did it. Nobody really got value out of it because everyone did it. But Google got a stronger signal to make their search engine better. So they are guiding us in the way that works for them, not necessarily what works for us.
00:40:10 → 00:40:46 Cara Corbett: OK, and then also in May, we had Google I/O, which is their annual developer conference. During this event, Google made its clearest statement yet that search is an AI product now. In fact, Google used the phrase “AI-first” repeatedly, and they announced the biggest redesign to their core homepage interface in over twenty-five years, called the Intelligent Search Box. This transforms the traditional search bar into a flexible, conversational, agentic AI field. So the SERP that agencies have been optimising for the past decade looks fundamentally different today. Personally, I think it’s a better experience. What about you, Mike?
00:40:51 → 00:42:00 Mike King: I want to sidebar real quick. I wasn’t at I/O this year, and I’ll tell you why: because the things that I’ve been writing have made some key people at Google upset. So I want everyone to know that I’m fighting for you and I’m missing out on opportunities because of it. Now, I think what they rolled out at I/O is always obviously important. I mean, you can see all that at the Google Blog, so I don’t want to spend too much time on it. But I do want to talk more about what this means. So Google’s overall strategy here is to get you as dependent on them as possible for answering your questions, right? Because when you have AI Mode, essentially, Google is synthesising the answer and then maybe sending you off to the website afterwards. What that does is it keeps you within Google’s experience rather than going to individual websites. And that’s obviously very bad for websites.
00:42:01 → 00:42:38 Cara Corbett: Yeah, I totally agree. And moving on, still in May, Google expanded the “Highly Cited” label to more web article links, which helps surface original reporting and primary sources. This is really relevant for PR and comms agencies: earned media in credible publications gets a visual signal in the SERP that other content doesn’t. It’s one of the clearest ways original reporting and third-party coverage translates into search visibility.
00:42:39 → 00:43:10 Cara Corbett: Moving on to the third of June: the UK’s Competition and Markets Authority, or the CMA, introduced a new conduct requirement on Google that publishers must be able to opt out of AI Overviews or any of Google’s AI services, and also opt out of having their content used for AI model fine-tuning, without losing their standard organic search visibility. Essentially, Google Search Console added a toggle that helps users decide whether their site can appear in and help ground Google’s generative AI search features. Any sites that opt out are just not going to receive traffic or impressions from these generative AI features. It’s being rolled out in the UK first, and then a global rollout is expected to follow. Mike, what’s your take on people opting out of Google’s AI features?
00:43:26 → 00:43:25 Mike King: I think it’s a pyrrhic victory. This is where everything is going. If you opt out, I think that’s a silly thing. But I think they should be able to. And I think it should be divorced from classic organic search visibility. Google just wasn’t doing that, which was smart strategically on their side. But yeah, it’s just not fair.
00:43:30 → 00:44:46 Cara Corbett: On the fourth of June, Google launched Search Profiles for creators and publishers, which are essentially dedicated, customisable pages that pull together articles, videos, and social posts in one place, with a Follow on Google button. Eligibility requires one hundred thousand followers on YouTube, Instagram, or X, or three hundred thousand-plus on TikTok. Creating a profile doesn’t directly affect rankings, but it can trigger or enhance a Knowledge Panel, which obviously gives us a lot more real estate on the Google search SERPs. Right now it’s US-only at launch, but there is international expansion planned. This is really relevant for media clients and thought leaders: just another way for them to take up more real estate on Google SERPs, but also have all of your accounts connected in one authoritative place. I wonder if this will also help with E-E-A-T and help Google connect those entities a bit more. Mike?
00:44:29 → 00:44:41 Mike King: Yeah, I think it’s great. I don’t like that it requires you to have one hundred thousand followers just because I don’t have that. But I think that’s cool.
00:44:47 → 00:45:39 Cara Corbett: On the seventeenth of June, the CMA introduced a second requirement called Fair Ranking. Google must rank organic search results, including AI Overviews, using objective and non-discriminatory criteria, and they must give advance notice of significant ranking changes. So maybe this will mean no surprise core updates or algorithm shifts right around Christmas when people are going on holidays. Give us warning, apparently. As for rollout, they were given a six-month implementation window. Two CMA rulings in two weeks is a meaningful shift in regulatory pressure on Google. Mike, would you say things are going to change with the way Google operates with all this pressure?
00:45:27 → 00:45:32 Mike King: Go ahead.
00:45:27 → 00:45:35 Cara Corbett: People got to understand that Google has review cycles. There’s a review cycle around that time of the year, so of course there are going to be things that hit.
00:45:35 → 00:45:38 Mike King: I didn’t know that.
00:45:38 → 00:46:14 Cara Corbett: So give me that insight. As for rollout, they were given a six-month implementation window. So two CMA rulings in two weeks is a meaningful shift in regulatory pressure on Google. Mike, would you say things are going to change?
00:45:52 → 00:46:14 Mike King: It sounds unrealistic what they’re asking for. Objective ranking criteria: how do you define that? I think that ranking criteria is inherently subjective because it’s reflective of what’s happening on the internet. So I think it’s an unrealistic thing that they’re looking for.
THEME 5: THE AGENTIC WEB
00:46:15 → 00:46:21 Cara Corbett: Interesting. All right, moving on to the final section, which is the agentic web. There’s been a lot of movement on this this year. If you’re dealing with brands that sell online, this is the section for you.
00:46:21 → 00:47:09 Cara Corbett: On the eleventh of January at NRF, which is the National Retail Federation’s annual conference in New York and one of the biggest events in the retail and commerce calendar, Google announced the UCP, the Universal Commerce Protocol. This is essentially an open standard for agentic commerce, co-developed with Shopify, Walmart, Target, Stripe, and twenty-plus other partners. It’s live for eligible US retailers in Google AI Mode and Gemini from day one. The idea is a standardised way for agents to discover, evaluate, and complete purchases across any retailer. So instead of a user clicking through websites, comparing prices, and filling out shipping forms, their personal agent now uses the UCP framework to negotiate and execute the purchase securely.
00:47:10 → 00:47:37 Cara Corbett: Also in January, at the same time, Shopify announced its Agentic Plan. We talked about this one in our last webinar, but this is specifically designed for businesses that don’t actually use Shopify for their online storefronts. It allows them to seamlessly sync their product data into the Shopify catalogue so that autonomous AI assistants can find and buy their products. Shopify is really positioning itself as that commerce layer underneath the agentic web, which is pretty cool.
00:47:38 → 00:48:11 Cara Corbett: UCP hit Shopping in March. They expanded to include cart and product discovery capabilities. So essentially agents can now add one or more products to the cart, and agents can pull real-time details from the store like variants, inventory, and pricing. And critically, UCP and AI product features began appearing in standard Shopping results, not just AI Mode. This is really the convergence point: the same product data signals that drive traditional Shopping visibility are now powering AI-driven commerce. Clients who maintain clean product feeds and strong SEO are going to be ahead.
00:48:14 → 00:49:32 Cara Corbett: And then we already kind of alluded to this one. In May, Google’s Chrome team added an LLMs.txt check to Chrome’s Lighthouse Agentic Browsing audit, essentially treating the file as a relevant thing for AI agents navigating websites. But then simultaneously, Google Search explicitly stated that it doesn’t use LLMs.txt for training purposes. And in their official guide, they said you don’t need to do it. So there are kind of two opposing official statements from Google giving contradictory guidance. Mike, I know you kind of spoke to it a bit, but what’s your take? I’ve seen some SEOs start to come around on LLMs.txt, but I’ve also seen a lot of pushback and some research against it. What’s your take?
00:48:53 → 00:49:32 Mike King: Yeah, I think LLMs.txt is worth using at this point. I wouldn’t spend a bunch of effort on it, but I would definitely use it. And then the UCP stuff; that’s where all this stuff is going. Shopping is going to become very agentic. It’s going to be a pay-per-transaction thing that’s really going to drive monetisation here. And you’ve also got the Agentic Commerce Protocol, which is what Stripe and OpenAI developed. I keep talking about this idea of machine media, where it’s going to be mostly these agents consuming the web on behalf of users, and these are the sorts of on-ramps to make that happen.
00:49:34 → 00:49:52 Cara Corbett: OK, Claude Cowork: we already kind of talked about this one too, but Anthropic launched Claude Cowork in January. It’s an AI agent that can read local files, execute code, and complete cross-app browser tasks. This was the first major step toward AI that operates your computer on your behalf. Pretty cool.
00:49:53 → 00:50:46 Cara Corbett: And then Apple and Google announced a multi-year partnership on the twelfth of January: Gemini is going to power Apple Intelligence and also a redesigned Siri. The rollout is happening in two phases. Phase one already arrived in the spring, delivering on-screen awareness and contextual understanding. Phase two will be a full conversational Siri, coming with iOS 27 in September 2026. So as of this webinar, it’s live but only partially. For agencies, the implication is that Gemini’s citation behaviour is becoming even more important; a brand’s visibility in Gemini is increasingly also its visibility through Siri, across two billion Apple devices. You have a take on this, Mike?
00:50:37 → 00:51:04 Mike King: I think Siri has been bad for so long, so I think it’s good that they finally got something decent in there.
00:51:05 → 00:51:13 Cara Corbett: I remember when I first started in SEO, seven or eight years ago, I was seeing headlines like “oh no, SEO is dead because of the rise of conversational search.” And here we are so long later. I’m looking forward to seeing what they come up with.
00:51:15 → 00:52:38 Cara Corbett: This next one is exciting, and I know that Mike’s team has done some cool research on it. In January, Google announced Personal Intelligence. Gemini and AI Mode can now cross-reference your Gmail, your photos, your YouTube history, and search history to give more personalised responses. So search is no longer a shared experience with a single set of results. Personalisation at this level means brand presence across multiple Google services, not just on your website, becomes more important for showing up in responses. Mike, Gareth did some cool research on this that he unveiled at SEO Week. Do you want to tell everyone about it?
00:51:43 → 00:52:38 Mike King: Yeah. So one of the things that Gareth led was an experiment where he had two different Gmail accounts. One was a control with no manipulation. The other one he sent emails from both fake brands and also signed up to mailing lists of real brands, and we looked to see how that impacted the search results. It does, pretty significantly. From the real brands it was something like a sixty percent increase in citations. And for the fake brands, they would get mentioned but no citations. The whole point is that because they were in emails, they were showing up in AI Mode.
00:52:29 → 00:52:38 Cara Corbett: Yeah, pretty crazy. I remember seeing the screenshot where he was asking about what streaming platform he should use, and it said something like “because you have a three-year-old daughter named X, you should be using Disney Plus.” I just thought that was so creepy. But yeah, it really shows how personalised these are becoming.
00:52:50 → 00:53:30 Cara Corbett: OK, and then Cloudflare launched a feature that automatically converts HTML to Markdown when an AI crawler requests it, cutting token usage by up to eighty percent. That was in February. Worth noting that Google pushed back on the broader trend shortly after: John Mueller called creating parallel Markdown versions of web pages a stupid idea, and argued that HTML conversion is trivial for any crawler and that Markdown strips away structural signals that both search engines and AI use to understand a site. We’ve seen some debate in the SEO community on this. Mike, would you say Markdown is worth it, or is this a genuine place where we should listen to Google?
00:53:30 → 00:53:51 Mike King: So my position has evolved on this. I originally thought it wasn’t valuable because I hadn’t seen any impact from it. But what it also does is it speeds up the loading of the page. And we have seen that when your pages are slow, ChatGPT will give up on waiting for it to load. So there can be some value from it.
00:53:52 → 00:54:10 Cara Corbett: OK, we’ve only got four minutes left, so I’m actually going to breeze through the next point. I’ll keep it on the screen; everyone’s going to get a copy of this. It’s essentially some tips on what we think agencies should do right now. But Mike, I want to ask you before we move on to questions: do you have any predictions for the remainder of the year?
00:54:11 → 00:54:36 Mike King: Just more personalisation across everything. And I’m not just talking about generative AI or LLMs or whatever. I think that Google Search is also going to become hyper-personalised, even in classic search results as well.
00:54:37 → 00:54:59 Cara Corbett: Interesting. And just so everyone’s aware, Mike’s been pretty good with his predictions. He called the AI Overviews interface; he built his own version of it before it even came out. So listen to this guy, because he knows what he’s talking about.
RESOURCES AND Q&A
00:55:17 → 00:55:31 Cara Corbett: Just a quick thing before we wrap up. If you’re not already part of the SUSO Partner Club, it’s a free resource for agencies. You get free access to our SEO team if you don’t have one in-house, a helpdesk to answer any questions that you or your clients might have, free one-on-one training on the foundations of SEO and GEO (so you can learn all about query fanout, RAG, and all this technical stuff we talked about on the call), and we also do free analyses and reports that can be white-labelled in your branding. So why not join? It’s completely free.
00:55:36 → 00:56:28 Cara Corbett: And finally, we’ve got some resources to share on the screen while we get to some questions. I mentioned our last webinar that went over how to make e-commerce stores more discoverable for AI search through structured product data; you can scan that QR code there and rewatch that if you like. We also have to give a shout out to iPullRank’s AI Search Manual. Like I said, it’s the most comprehensive GEO resource out there. Mike, do you want to talk a little bit about it?
00:55:45 → 00:56:09 Mike King: It is twenty-four chapters and we keep it up to date on everything you need to know about AI search. There are no vagaries; we walk you through how it works, what you do, and what tools to use. It’s everything you need to know. And it’s a big reason why we show up before everybody in your ChatGPTs and Geminis. And they have a TLDR version, so you don’t have to dig through the whole thing.
00:56:11 → 00:56:28 Cara Corbett: Yeah, it’s really comprehensive and good. And then of course, a link to our SUSO Partner Club. If you haven’t joined already, like I said, it’s completely free. I’d love to work with you.
