Are Google’s AI-generated summaries in their Search Generative Experience (SGE) interface unfairly stealing clicks from the website publishers who originally created those answers? This is the question everyone in SEO and digital marketing has been asking for a couple of months now. With Google’s latest update displaying links to SGE’s sources more prominently, now seems the best time to revisit that question. Let’s stop talking about AI for a moment But before we start, please do me a favor. for the next couple of sections, let’s set aside how the snippets in SGE are generated. Why? Because I’ve noticed that too many discussions about SGE devolve into arguments about generative AI and large language models. It’s a worthwhile discussion to have, and we should be talking about it. But we’re missing the point by reducing our thinking about SGE to the particular algorithm that’s currently powering it. Fixating on this particular iteration of an algorithm that Google uses to generate previews of search results distracts us from asking much more interesting questions. As SEOs, we should think about features in terms of the search experience itself and ask: What is the search page displaying? In what ways is the search page presenting that information? How do the structure and content of the search page impact those who create the content being displayed (website publishers) and those seeking answers to their search queries (users)? How do those impacts compare against alternative implementations of a search results page? Which of those implementations makes for the best search experience (whatever that might mean)? We can’t ask any of those questions when we are getting stuck in the complexities of how Google’s generative AI model works. So, for a few minutes, let’s pretend it doesn’t matter how Google generates the answers displayed in SGE. Our discussion won’t depend on whether those answers are crafted by generative AI, a different algorithm, submitted by website owners, hand-written by a Googler stuck in some basement office, or magically cast into existence by a team of little green fairies. Can you do that for me? Great! Now, let’s talk about SGE. SGE’s new link attribution On Tuesday, Google introduced a new design to SGE that made links more visually prominent. Previously, the default SGE interface didn’t show sources in an obvious way. Instead, users had to click on a button to see links explicitly within the text of that snippet: Old SGE interface as documented by yours truly on May 26. As of Aug. 1, there is only one mode, which shows a clickable chevron at the end of each paragraph: The SGE snippet from one of my searches earlier today. When you click on that chevron, you’ll see a dropdown with clickable links for webpages where SGE sourced its summary: Same search result, but now I’ve clicked to see expanded link information. As you can probably tell, this new UI looks very similar to the expanded mode on SGE’s old interface. However, I suspect this will be a major improvement for most users. I’m probably in the minority for having regularly clicked on that tiny expansion button. Why do we care? Even small design tweaks to how Google displays attribution are important to anybody creating and publishing content on the web. After all, Google’s previous iteration of SGE received a lot of public criticism. Most notably, Avram Piltch from Tom’s Hardware wrote in early June: “Instead of highlighting links to content from expert humans, the ‘Search Generative Experience’ (SGE) uses an AI plagiarism engine that grabs facts and snippets of text from a variety of sites, cobbles them together (often word-for-word) and passes off the work as its creation.” Many have already praised Google’s new UI as a significant improvement. SGE “is learning to give credit where it’s due,” says Android Police. 9to5Google hopes this design will end up “directing more traffic through to the sources.” I even heard good things from one of the more vocal critics of SGE’s initial lack of attribution, Lily Ray. Ray, who serves as the Senior Director, SEO and Head of Organic Research at Amsive Digital, told me: “I’ve been very vocal since day one about the importance of Google incorporating links into SGE answers, just like Bing Chat has done from the start. I was excited (and relieved) today to see that Google took our concerns seriously and seems to be testing multiple layouts for incorporating links in SGE answers.” Slow down: Is SGE actually a ‘plagiarism engine’? But wait. What did Tom’s Hardware mean by “plagiarism” in the quote above? Piltch’s description is genuinely fascinating. Read this again, but apply every statement to Google’s pre-existing featured snippets: “…engine that grabs facts and snippets of text from a variety of sites, cobbles them together (often word-for-word) and passes off the work as its creation.” The only real difference between SGE’s summaries and featured snippets is the number of links they include (and a lack of pretty colors): Featured snippet result for “how to learn Italian”, featuring only one website. You could say that the example above, like SGE, also “grabs facts and snippets of text” explicitly “word-for-word” from Untold Italy’s website. So why is one interface treated as searching while the other as stealing? I believe SGE includes quotes from websites functionally the same as the text displayed in featured snippets or most other rich results within Google’s SERP. Danielle Stout Rohe, Sr. Data Program Manager, Content and SEO at Cox Automotive Inc., agrees. As she told me: “When SGE first came out, it felt like an expanded featured snippet (FS) to me. The only major difference was that the default view in SGE didn’t clearly indicate what text was tied to what site, but the other SGE view does.” SGE feels like a natural development from previous SERP features. In 2007, Google wrote about meta descriptions that they “want snippets to accurately represent the web result.” In 2012, they introduced rich snippets guidelines to “provide even better summaries” to users. In 2014, they added “structured snippets” for additional data. In 2018, Danny Sullivan announced a relaunch of featured snippets, saying: “We display featured snippets in search when we believe this format will help people more easily discover what they’re seeking, both from the description and when they click on the link to read the page itself. It’s especially helpful for those on mobile or searching by voice.” Most of this language tracks pretty directly across all those developments, from 2007 to the 2023 test with SGE. But I wasn’t there in those early days of SEO. Thankfully, I know someone who was. So, I asked fellow marketer and my occasional collaborator Rand Fishkin (CEO of SparkToro; formerly Moz), who told me: “Even in the pre-2010 days, most SEOs I talked to had the sense that the “10-blue-links” era was fading (especially after the integration of things like Google Maps, weather, sports scores, etc. between 2006-2009). I never believed much in predicting the future, but I don’t think the rise of zero-click searches and instant answers would have been a huge surprise to anyone who watched the SERPs in those years.” We can speculate how much SGE was influenced by the recent popularity of OpenAI’s ChatGPT. But I don’t think these features are simply Google’s attempt to capture industry hype. SGE’s summaries are simply the newest form of “instant answers” in a long line of previous iterations. Get the daily newsletter search marketers rely on. Why were we afraid of featured snippets? What do we fear when we accuse a search engine like Google of “stealing”? One way to approach this problem is to look backward. How did we feel when featured snippets and their lengthy quotes first launched? Featured snippets also caused a fair amount of anxiety. As Paul Shapiro,…
