Google, in their words, recently began “supercharging Search with generative AI.” Their new AI Overviews (AIO) product launched in May 2024 after testing with the name Search Generative Experiences (SGE) in Google Labs for over a year.
AIO pops up on the search engine results page and will show a snippet of AI-generated information about your search query with links included. In Google’s words:
"Normally, you might break this one question down into smaller ones, sort through the vast information available, and start to piece things together yourself. With generative AI, Search can do some of that heavy lifting for you."
Sounds great in concept. However, Google runs on advertising revenue, so don’t expect complete abandonment of site links in favor of a chatbot experience.
For insight on how advertising has changed, three sources provide valuable information. The first is a study of the SGE experience in Labs by SERanking pre-launch, the next are studies from SEO company BrightEdge post-launch of AIO, and the last is a case study: Bing already released a generative search product last year.
SERanking analyzed 100,000 keywords on SGE across 20 niches focused in the New York City area. Because of these pre-launch findings, advertisers were prepared for significant shockwaves when Google rolled SGE out.
In 27% of cases, no ads accompanied SGE snippets. In 73% of cases, ads were present. Ad testing was a heavy component of SGE snippets with product recommendations being a key feature. In SGE demo videos and pictures, you’d often see AI recommending a handful of products in its SERP feature.
(Source: Google blog)
In this testing phase in Google Labs, ads at the top of the SERP accompanied SGE snippets more often than shopping ads, but shopping ads tended to appear above the SGE snippet.
In addition to these findings that would potentially impact advertising, SERanking made some other key observations:
Long-tail keywords were more likely to trigger AI-powered responses.
Keywords from the Food and Beverage niche trigger SGE snippets most often, while the News and Politics niche gets SGE responses less often.
64% of the analyzed keywords had an SGE answer or a Generate button; 18.5% of results supplied the snippet outright.
After expanding its snippet, the SGE answer most often had 8 links.
85.5% of SGE snippets linked to one or more domains from the top 10 organic results.
SGE would not take the place of ads, but Google seemed to want to give primacy to shopping links and shopping ads. Over the last few years, they added shopping links to Google Images, created a reverse image search that primarily recommends ecommerce products, and created the Performance Max campaign structure for ecommerce in Google Ads that runs responsive ads across all of Google’s ad placements with a black-box algorithmically driven optimization scheme.
It makes sense that Google would want to preempt visits to Amazon.com by creating an alternate shopping platform, collect referral fees, and take market share back from Amazon’s growing shopping ads platform. In fact, part of the DOJ’s lawsuit against Google alleges that
"Google actively tried to weaken companies like Expedia, OpenTable and Amazon. ... Google both limited their visibility in searches and demanded access to their data — all while making its own widgets to do the same things."
The research findings that long-tail and news keywords receive more SGE snippets suggests Google wants to maintain search ad relevance, so they preferentially serve ads for queries with clear intent or immediacy while returning explanatory content for information seekers or queries that would feel awkward to monetize.
When Google officially launched AI Overviews, the public-facing product was much different from the tested version. BrightEdge tracked the differences between SGE and AIO with their Generative Parser, and the results reveal a rollback on Google’s AI.
After launch, the portion of search results that featured an AI snippet dropped from 84% to 15%. This is mainly because Google removed the ability for users to opt-in to the feature to generate an AI snippet from the search results. This opt-in option is no longer available in AIO:
(Source: SE Ranking blog)
Even still, AIO prevalence continued to decrease, dropping to 7% of queries by the end of July.
The release of AIO also brought shrunken snippets, taking up 30% less space than SGE results:
(Source: BrightEdge’s blog)
Google rolled out product carousels in AIO for ecommerce search queries in May, but apparently decided on greater caution with ecommerce. By July, the prevalence of carousels in AIO fell in favor of educational Overviews. Keywords featuring product carousels went from their peak of 7% in May down to only 2%. Currently, the vast majority of search queries with purchase intent return results pages with only ads at the top, free of AIO:
(BrightEdge’s query for aloe vera juice in June 2024 vs. our query in September 2024.)
Google’s rollout of AIO seems cautious in the face of risks to revenue with search advertisers and the reputational risks of generating misinformation and inflammatory content. Bing’s foray into generative AI search helps explain why Google pulled back on their bullish push toward AI.
Bing already released generative AI search, so it isn’t exactly a new frontier. Bing received a burst of interest at the time but also a lot of scrutiny for misinformation and biased content.
(Source: Bing blog)
Large Language Models do not really possess knowledge or even understand language; they are more like word prediction algorithms that operate so well they create the illusion of understanding. Bing and ChatGPT have suffered some reputational damage from the nonsense or harmful answers they provide, as evidenced by the Verge article above.
Google acknowledged the reputational risk the product poses, especially if rushed to market. This, along with the fast pace of OpenAI, helps explain why they are second to market (in a market of two).
When Google introduced Bard, an ad promoting the chatbot showed the bot sharing inaccurate information. Alphabet, Google’s parent company, temporarily lost $100 billion in market value. Google had the opportunity to learn from Bing’s example and no longer needs to worry about rushing to earn first-to-market.
Google search is bad enough that a lot of users append the word “Reddit” to their searches to find more relevant, helpful information. Google acknowledged this and even released a short-lived product called “perspectives” as an attempt to keep people on Google’s properties rather than Reddit. If you Google “search has gotten worse” you will find a number of articles written about research from a German study which states in its abstract:
"We further observe an inverse relationship between affiliate marketing use and content complexity, and that all search engines fall victim to large-scale affiliate link spam campaigns. However, we also notice that the line between benign content and spam in the form of content and link farms becomes increasingly blurry—a situation that will surely worsen in the wake of generative AI. We conclude that dynamic adversarial spam in the form of low-quality, mass-produced commercial content deserves more attention."
It feels inevitable that affiliate spam campaigns and SEO strategists will quickly find ways to game generative search as well. For now, the rollback in prevalence of AI Overviews and the smaller snippet sizes mean advertisers will not feel the immediate impact of AIO too dearly. Publishers report the impact of AIO on search traffic as negligible. The reprieve may prove temporary though, so stay on the lookout because Google still plans on scaling AIO up.
Last updated on October 4th, 2024.
