TL;DR: Google’s AI Overviews
- What They Are
-
- AI Overviews are Google’s AI-generated summaries at the top of search results, powered by Gemini LLM, giving concise answers with linked sources.
- How They Work
- Pull from multiple sources: web content, Google’s Knowledge Graph, entities (people/places/things).
- Combine indexing + LLM reasoning to generate cohesive responses.
- Offer transparency via “More Options” showing how results are determined.
- Rollout & Purpose
- Launched May 2024 in the U.S., expanded to 120+ countries and 11 languages.
- Designed to simplify complex queries, boost satisfaction, and increase diverse site visits.
- Sites below AI Overviews often see sharp traffic declines.
- Impact on Publishers
- Studies show 7–25% traffic drops (sometimes up to 89% for certain publishers).
- CTR falls by ~35% when AI summaries appear.
- Top-ranking pages displaced by AI Overviews can lose up to 79% of traffic.
- Some publishers are adapting by optimizing for Generative Engine Optimization (GEO).
- GEO (Generative Engine Optimization)
- Extension of SEO for AI-driven engines.
- Focuses on producing unique, structured, authoritative content that LLMs prefer.
- Early adopters see massive gains (e.g., 1,300% increase in AI referral traffic).
- Benefits & Challenges
- Pros: Quick answers, planning tools, more diverse sources surfaced.
- Cons: Accuracy issues (hallucinations), lack of opt-out, risk of bias, major impact on referral traffic.
- Fuels rise of zero-click searches: users often get answers without visiting external sites.
- Looking Ahead
- Google is expanding personalization, planning features, and AI-organized results pages.
- Ongoing tension between user convenience vs. publisher visibility, accuracy, and fairness.
Ever wonder what exactly drives AI Overviews? Well, here’s a detailed and comprehensive guide discussing Google’s AI Overview system. We’ll explore how it works, what features it brings, its technical underpinnings, its rollout and impact, and the challenges it faces.
What Are AI Overviews?
AI Overviews are AI-generated summaries that appear at the top of Google Search results, offering a synthesized snapshot of key information on a topic or question aimed at giving you clarity at a glance while also linking to full sources to help explore further. These summaries typically occupy the top spot on the search results page and are designed to answer more complex queries in a concise and helpful way.
Below is an example of a typical AI Overview:
Notice, there are two components:
- The Overview
- The Source
The Overview
This is the section that provides you with the actual answer. It is generated by Google’s Gemini Large Language Model. Depending on the information provided, it can also be expanded to “show more” information. Additionally, the overview will provide direct links to information it pulled for individual paragraphs, as you can see below:
“Query Fan-Outs” – The Source
This is the section showing a user where the information was derived from. It will include a small amount of text, an image (if applicable), and the source site. These sources are hyperlinked, so a user can select them and visit the source website directly.
More Options:
Finally, a user can click on the three dots above the sources to reveal how Google is determining their results. More on this later.
How Did We Get Here?
Officially launched in the U.S. in May 2024 with plans to reach over a billion users globally by year-end, AI Overviews have since expanded to over 120 countries, supporting 11 languages.
Google offers an experimental version through Search Labs titled “AI Overviews and more,” allowing users access to AI-generated features and improved control over how summaries are displayed.
What Purpose Do AI Overviews Serve?
AI Overviews help users “get to the gist of a complicated topic or question more quickly” and serve as a launchpad to discover deeper content via linked sources. First introduced experimentally under the “Search Generative Experience” (SGE) and officially rebranded as “AI Overviews” in the U.S. in May 2024, the feature’s rollout stemmed from positive user feedback during trials.
Usage data suggests that AI Overviews increase user satisfaction, prompt more diverse site visits, and boost traffic to linked pages compared to traditional web listings.
That being said, there is overwhelming evidence that websites immediately below these AI Overviews suffer considerable traffic loss.
So, How Do They Work Under the Hood?
AI Overviews leverage a custom version of Google’s Gemini LLM that is integrated with Google’s search infrastructure. This LLM then uses advanced reasoning and multi-step logic to come up with a specific set of results. The system gathers and synthesizes information from multiple web sources, crafting a cohesive AI-generated answer. In other words, the system uses two components to craft its answers.
Two AI Answer Components:
The Index
The source of the information provided. This is seen to the right of the AI Answer.
The LLM
The system that interprets the data by calculating numerical representations called tokens and then utilizes the statistical relationships and patterns between them. For example, it learns what words or phrases are likely to follow others. These learned relationships essentially give the LLM its “knowledge”.
Both components working together ultimately provide the user the information seen.
Understanding The Source of AI Information
Earlier in this article, we mentioned the “More Options” next to the AI Sources. This is a telling piece of information that sheds light on this question. Let’s look at it closer. It specifically states that the information was derived from:
- Info across the web
- Google’s Knowledge Graph
- Info about People
- Info about Places
- Info about Things
Info from Across the Web
So that’s a bit ambiguous, but in the end it’s content.
- Written content – guides, case studies, stories, reviews
- Visual content – images, infographics & memes
- Audio content – panel discussions & interviews
Delivered on:
- Blogs
- Social Media
- Books
- Memes
- Videos & Podcasts
- etc
Content that is relevant to a specific subject matter.
Google Knowledge Graph (GKG)
The Google Knowledge Graph is a database that organizes factual information about people, places, and things. It then correlates them to their relationships with each other, creating an “entity”.
Quick Grammar Review:
“people, places, and things”. Where else have you heard that?
Oh yeah – 3rd grade English Class.
So the Google Knowledge Graph looks at nouns and how they are related to one another.
Let’s construct an example of this:
Stephan Boehringer (the people or person) is associated with
Get The Clicks (the thing), who has an office in Orlando (the place).
These three components drive a large portion of the Knowledge Graph and contribute to the concept of the “entity”.
The Entity = The Brand
The concept of an “entity” in the digital marketing world is also known as the Brand. It consists of:
- The brand of the person
- The brand of the place, and
- The brand of the thing.
Google Knowledge Panel
In the end, the information derived is the Google Knowledge Panel. An example of this is below.
How Small Businesses Can Gain Exposure in AI Overviews
Businesses need to get on board with Generative Engine Optimization (GEO). While it is yet another thing to optimize, it is absolutely imperative not to be left behind. Think of Generative Optimization as the Bolt-on-Service to your SEO. You can’t do GEO unless your SEO is solid.
SEO provides the foundation for GEO, the next level of optimization that opens the door to a whole new ecosystem of possibilities. This obviously includes AI Overviews, but GEO also optimizes for other LLMs such as:
- ChatGPT
- Siri / Apple Intelligence (Apple Devices)
- Copilot / OpenAI (Windows Devices)
- Gemini (Android Devices)
- Claude (Amazon)
- Perplexity
Generative Engine Optimization
To get this done, a business needs to:
- Figure out which search results you care about that are showing AI Overviews.
- Review the content that is being provided and who the sources of that content are.
- Write better content. Content should be unique, provide information that is not available from other sources, and offer information gain. The content needs to follow a concise structure that is easy for an LLM to understand. Don’t be wordy; keep it simple, but provide all the information needed.
- The entity (person, place, thing) needs to be an authoritative brand on the subject matter. So the brand behind the content needs to be top-notch.
Our GEO Program is designed to accomplish just that. It pushes your brand to new heights by leveraging an omnichannel approach to Brand development. We determine which searches trigger AIOs and then curate very specific content that is easy for the LLMs to understand. This content is provided through multiple channels, including video & social media.
Evolving Features & Capabilities
Google is developing options to let users adjust AI Overviews,making them simpler or more detailed depending on the user’s familiarity or learning goals. Future updates will enable handling of complex, multi-faceted queries like combining multiple constraints (e.g., “yoga/pilates studio popular with locals, with discounts, within a commute for me”). Google has begun integrating planning capabilities such as meal plans, trip itineraries, and more, directly into Search, with options to tweak or export (e.g., to Google Docs or Gmail). For idea-generating searches (like dining, recipes, travel), Google will roll out AI-organized results pages, grouping content under AI-generated subheadings to help users explore ideas more easily.
Benefits for Users and Publishers
Google’s AI Overviews are designed to streamline the search experience by surfacing concise summaries at the top of results pages. For users, this means faster access to key insights without needing to click through multiple links. The tool can also help simplify complex topics, breaking them down into digestible explanations. In addition, the system increasingly tailors its responses, offering planning tools and personalized query handling that adapt to individual needs. Still, the convenience comes with caveats: the generated insights can sometimes contain flaws or inaccuracies, underscoring the importance of treating them as a starting point rather than a definitive answer.
For publishers and content creators, the impact of AI Overviews is more complex. Some research indicates that these summaries may expand the range of sites that receive traffic, introducing readers to sources they may not have otherwise encountered. This could help diversify visibility across the web. At the same time, the prominence of AI Overviews at the top of search results raises concerns about click-through rates. If users find what they need within the summary itself, fewer may take the extra step of visiting the original page. This trade-off continues to be a focal point in the debate over the role of generative AI in online discovery.
Limitations and Criticisms
Despite its promise, AI Overviews faces several criticisms. One of the most persistent challenges is accuracy. Like other generative AI systems, the tool is prone to so-called “hallucinations,” producing content that is misleading, unreliable, or outright incorrect. These issues have sparked concern among users and experts alike, prompting Google to strengthen safeguards and refine the system’s quality controls.
Control and transparency also remain sticking points. At present, users cannot fully disable AI Overviews by default. Those who prefer traditional search results must manually switch to the “Web” filter on each query or turn to third-party extensions to maintain a classic search experience. This lack of permanent opt-out functionality has drawn criticism from users who feel the system prioritizes automation over choice.
Finally, questions of bias and fairness linger. By elevating certain sources within its summaries, AI Overviews may unintentionally amplify specific voices while sidelining others. Critics argue this dynamic could skew visibility, reinforcing existing biases or creating uneven opportunities for publishers. These concerns highlight the broader tension between the efficiency of AI-driven tools and the need for transparency, diversity, and balance in how information is presented.
Real-World Examples & Case Studies
Traffic Declines Among News and Content Publishers
A report from Digital Content Next (DCN), which includes media companies like The New York Times, Condé Nast, and Vox, surveyed 19 member sites in May–June 2025. The findings: most sites saw a drop in Google search referral traffic between 1% and 25%. The median year-over-year decline overall was about 10%, with news sites faring slightly better (-7%) than non-news sites (-14%).
For example, some prominent news organizations have reported steeper losses. The Guardian described “sharp declines” in traffic. Some publishers saw drops of up to 89% in traffic from Google Search after AI Overviews and related AI-Mode features were introduced
Click-Through Rate (CTR) Drops When AI Summaries Appear
According to Pew Research, when users see an AI-generated summary with search results, they are significantly less likely to click through to other websites. For searches with AI summaries, only 8% of visits resulted in someone clicking a standard result link, versus 15% when no summary was present.
In addition, a study by Ahrefs estimated that AI Overviews correspond with about a 34.5% drop in average CTR for keywords triggering those summaries
Another pattern
Queries with both Featured Snippets and AI Overviews see the most significant visibility loss, especially for non-branded or lower-ranking keywords.
Severe Impact for Top Ranking Pages Below AI Overviews
An analysis by Authoritas found that websites that used to hold the #1 spot for a query but are now pushed below an AI Overview can lose as much as 79% of their traffic for that query.
Adaptation & Growth Cases
While many sites are seeing declines, some are managing to reverse or even capitalize on AI Overviews.
In one case study, a client was able to grow “AI referral traffic” by 1,300% by optimizing content specifically to be cited in AI Overviews; before the effort, their content wasn’t appearing in any AI Overviews, but after changes, they secured citations for dozens of relevant keywords.
This suggests that while many publishers are losing clicks, there is potential upside if content is positioned and formatted in ways that align with what Google’s AI wants to use in summaries.
“Zero-Click” Searches & Changing Behavior
The rise of AI Overviews is accelerating the shift toward “zero-click” searches, with users ending sessions on Google without visiting external websites. Pew Research Center data show that users who see summaries are more likely to end their browsing session without a click, as opposed to continuing to other sites.
A Few Final Thoughts About AI Overviews
These examples show that AI Overviews are not just theoretical risks but are already reshaping referral traffic patterns:
- Publishers that traditionally ranked high in organic results may find that their visibility is reduced even if their ranking remains the same, simply because AI Overviews appear above them.
- Non-branded, informational content seems especially vulnerable. These are the kinds of queries where users are satisfied with a summary rather than navigating to an external page.
- On the other hand, publishers who are able to get their content cited in the overviews or who produce content that goes beyond what AI can summarize (e.g., deep analysis, unique data, personal stories) may still retain or regain traffic.
In essence, Google’s AI Overviews represent a significant evolution of search, transforming how information is surfaced, organized, and consumed. With sophisticated language models powering dynamic summaries and new planning tools, the system can make searching quicker and more intuitive. However, trade-offs in control, reliability, and content visibility are key areas that Google continues to refine.
For more information about this and the opportunity GEO presents, contact us, and we’ll be happy to discuss this further.
