Latest / The SEO Podcast with Fexingo: Search Engine Optimization, Google Rankings, and Organic Traffic / How Google Rewrites Search Results With AI Overviews
Transcript
- Lucas: Look at the SERP right now. You type in a complex query about enterprise software procurement, and half the screen is taken up by a box generated by Google's AI models. It’s not just a featured snippet anymore. It’s a synthesized summary that pulls from three different sites and cites them at the bottom. For a lot of publishers, that feels like an eviction notice. Luna: It definitely changes the click-through dynamic. If the user gets the full answer in that box, why do they need to visit your site? I’ve seen some metrics where the top-ranking URL still gets clicks, but the volume is just... thinner. Lucas: Thinner is one way to put it. I was talking to a friend who runs a niche B2B SaaS blog last week. They specialize in compliance workflows for mid-sized fintech firms. Their traffic dropped by forty percent in the last two quarters since Google rolled out these AI Overviews more aggressively. And it wasn’t because their content got worse. It was because Google started summarizing their guides directly in the overview box. Luna: Forty percent is huge. Did they try to block the AI scraping or change their formatting? Lucas: They tried everything. Noindex tags, schema markup, even rewriting their headers to be less directive. But here’s the thing about the new system: Google isn’t just looking for the best single page anymore. It’s building a knowledge graph on the fly. So even if you don’t get the top spot, your content might be used as a source citation inside the AI response. You’re losing the click but keeping the attribution. Or so we hope. Luna: That attribution piece is tricky though. If I’m a reader and I see four sources cited, I’m probably going to scan those links rather than clicking the first one. It fragments the traffic. Lucas: Exactly. And it creates this weird incentive structure where you want to be cited, but you don’t necessarily want to be summarized poorly. There’s a case study from a major financial news outlet where they found that their articles were being pulled into AI Overviews for stock market queries, but the summary missed key nuances about risk factors. That’s a brand safety issue now, not just a traffic issue. Luna: Right, because accuracy in finance is everything. If the AI glosses over a regulatory nuance, that could lead to real liability. How are other industries handling this? Lucas: Healthcare is the most cautious. Many medical associations have explicitly told their members not to optimize for AI snippets because the stakes are too high. But for B2B and e-commerce, it’s a wild west. We’re seeing retailers trying to structure their data so specifically that the AI has no choice but to link back to their product pages. It’s almost like they’re feeding the machine exactly what they want it to chew. Luna: So it’s become this game of structured data tennis. You serve the perfect json ld, hoping the AI returns the ball directly to your cart. Lucas: That’s a good way to describe it. And the server side is getting faster. Earlier this year, we talked about how Google uses entity clusters to understand topics. Now, those clusters are being used to generate the actual text you read. It’s not just indexing; it’s creating. Which means the barrier to entry for simple informational queries has basically vanished. Anyone can write a FAQ page, but the AI can synthesize ten FAQ pages in seconds. Luna: Which makes unique human insight even more valuable. If the AI handles the generic stuff, the only reason to click through is if you know something the AI doesn't. Like internal data or recent experiments. Lucas: Precisely. The winners right now are publishers who are doing original research. Not just aggregating news, but actually running surveys, analyzing proprietary datasets, or interviewing industry leaders. Because the AI can’t fabricate that without hallucinating, it tends to cite the original source more heavily. It’s rewarding depth over breadth. Luna: That sounds like a shift back to quality, which is nice. But is it sustainable for smaller teams? Original research is expensive. Lucas: It is. And that’s where the gap is widening. Large enterprises can afford to commission reports that become the default citation for AI Overviews. Small businesses are left competing for scraps of attention in the footnotes. It’s a consolidation of power in search, similar to what happened with ad auctions in the early two thousands. Luna: I remember that era well. The big agencies bought all the prime spots. Now the big data sets buy the prime answers. It’s the same cycle, different technology. Lucas: Spot on. And for listeners who run small marketing teams, the advice is surprisingly simple. Stop trying to game the snippet. Start trying to own the data point. If you can publish a statistic that no one else has, the AI will likely use it and link to you. It’s the new backlink strategy, but based on truth rather than hyperlinks. Luna: Truth as a ranking factor. I like that. It forces us to be more honest about our expertise. Lucas: It does. And it also means that transparency is crucial. If you’re citing sources in your own content, make sure they are authoritative. The AI is learning to trust certain domains more than others. Building a reputation as a reliable node in the knowledge graph is becoming as important as having fast load times. Luna: So technical SEO isn’t dead, but it’s secondary to credibility. You can have the fastest site in the world, but if the AI thinks you’re fluff, you won’t get cited. Lucas: Correct. Speed gets you indexed. Credibility gets you included in the answer. Those are two very different hurdles. And honestly, I think we’re only at the beginning of this shift. As the models get better at understanding context, they’ll stop relying on simple keyword matches and start looking for nuanced argument structures. Luna: Which means long-form content might come back into style. Not just for length, but for the complexity of thought it demonstrates. Lucas: I suspect so. The AI can summarize a listicle instantly. It struggles to summarize a well-reasoned essay without losing the thread. So publishers who invest in narrative depth might find themselves in a safer position than those churning out list-based content. Luna: That’s a hopeful note for writers. Less spam, more substance. Though I worry about the short-term pain for sites that relied on volume. Lucas: The pain is real. My friend’s B2B SaaS blog is going through a hard transition. They’re cutting low-performing posts and doubling down on three flagship research pieces. It’s a pivot, but it’s a necessary one. The old playbook of publishing five hundred thin articles a month just doesn’t work in an ai first world. Luna: Five hundred thin articles versus three deep dives. The math changes completely. Okay, let’s take a brief pause here. If these marketing conversations have sparked something you've actually used, consider supporting the show directly. A handful of listeners chip in monthly through buy me a coffee dot com slash fexingo, and that's literally what funds making this many of these shows possible without ads cluttering your feed. Lucas: Yeah, it keeps us independent. We don’t have to answer to advertisers or chase viral clicks. Just straightforward analysis for people who care about how these systems actually work. Luna: Thanks to everyone who supports us. It makes a real difference in our ability to dig into these topics deeply. Lucas: Back to the topic. One final thought on the future of this. We’re going to see more tools emerging that help marketers monitor their presence in AI Overviews. Tracking which queries trigger your citation, and how accurate the summary is. It’s going to become a new KPI alongside impressions and clicks. Luna: Citation accuracy as a metric. That’s going to be interesting to track. I imagine there will be plenty of A-B tests on how different phrasing affects the AI’s summary. Lucas: Absolutely. And eventually, we might see interactive elements within the search results themselves. Not just text, but calculators, configurators, live data feeds. The search bar is becoming a command line for the web. Luna: A command line. That implies a higher level of intent. Users aren’t just browsing; they’re executing tasks. Marketers need to build for execution, not just consumption. Lucas: Exactly. Build tools that solve problems, not just pages that describe them. If you can provide the utility that the AI summarizes, you win. If you can provide the utility that the AI replaces, you survive. It’s a clear distinction to make. Luna: Utility over description. Simple enough in theory, hard enough in practice. But it gives us a clear direction for the next quarter at least. Lucas: It does. And for those listening, ask yourself: does my content teach, or does it tell? In an age of synthetic answers, teaching is the only thing that drives a human to click through. Luna: Teaching wins. Good reminder. Thanks for walking us through this shift, Lucas. Lucas: Anytime, Luna. Let’s see how this plays out over the next few months. The landscape is moving fast.