Google AI Overviews: Everything you need to know

Optimizing for Google AI Overviews is becoming less about following specific content formats and more about providing useful information Google can retrieve, synthesize, and cite.

After two and a half years of experimenting with AI Overviews across industries, my agency has found that some early optimization practices still hold up, while others have turned out to be AI folklore.

Meanwhile, AI Overviews continue to expand. In June, they appeared in 39.4% of U.S. desktop Google searches, up from 25.8% in July 2025, according to Comscore. Their reach has also expanded deeper into commercial research and consideration.

How AI Overviews work: What we know now

In its latest guidance for optimizing for generative AI features, published in May, Google confirmed that AI search remains deeply connected to Search. Its generative features rely on the same core Search ranking and quality systems.

When AI Overviews appear

AI Overviews appear when Google’s systems determine that a searcher wants to quickly understand information from a range of sources, according to Google.

There isn’t a published list of queries guaranteed to trigger one, but some types of searches are much more likely to generate AI Overviews than others. Seer Interactive’s April analysis of 5.47 million tracked queries found AI Overviews appeared in: 

• 85.9% of question-format queries.
• 95.4% of comparison queries.

Query length alone appears to tell us less. Even single-word informational queries generated an AI Overview 27.3% of the time in Seer’s analysis.

We also know that AI Overviews are expanding deeper into commercial search. In July, Semrush analyzed more than 600,000 U.S. desktop keywords and found that commercial-intent SERPs containing an AI Overview increased 71% over six months, while transactional SERPs containing one fell 5%.

All of this changes the way I evaluate AI Overview opportunities. I’m relying less on broad intent labels and looking more closely at the work a query is asking Google to do. Searches that require an explanation, comparison, recommendation, or several pieces of information to reach a useful answer are stronger candidates for synthesis.

That lens has also pushed me further down the decision journey, toward searches where someone is comparing options, evaluating tradeoffs, researching pricing, or deciding what fits a particular use case. Those queries can give Google plenty to synthesize while putting a brand in front of someone much closer to a decision.

How Google finds information for AI Overviews

In its May guide to generative AI optimization, Google confirmed it can use retrieval-augmented generation (RAG) to retrieve relevant information from pages in its Search index. AI Overviews and AI Mode can also use query fan-out, issuing multiple related searches across subtopics and data sources to develop a response.

Google gives the example of a search for “how to fix a lawn that’s full of weeds” generating related searches about herbicides, removing weeds without chemicals, and preventing weeds.

A simplified version looks something like this:

• Original query → related information needs → multiple searches → relevant information retrieved → generated response → supporting sources

That broader retrieval process helps explain why the sources cited in an AI Overview don’t always match the pages ranking highest for the original query.

In a March analysis of 863,000 SERPs and 4 million AI Overview URLs, Ahrefs found:

• 38% appeared within the first 10 result blocks for the same query.
• 31.2% appeared between positions 11 and 100.
• 31% weren’t found within the first 100.

Research into query fan-out helps explain some of that disconnect. In an analysis of 173,902 URLs, pages ranking for both the main query and at least one fan-out query were 161% more likely to earn an AI Overview citation than pages ranking only for the main query.

The study shows correlation, not causation. But it still supports a broader shift in how I think about the scope of an AI Overview opportunity.

The target query is really a starting point. Query fan-out gives us insight into the related questions, comparisons, considerations, and subtopics that Google may explore around it. I use that broader landscape to consider how our content can provide useful information beyond the exact query we started with.

What counts as AI Overview visibility?

Traditional rankings still matter. I want pages to perform well in organic search, and strong search visibility gives Google more opportunities to discover and evaluate our content.

But “ranking” gets fuzzier in an AI Overview. A citation is one form of visibility, and simply earning one doesn’t tell me the whole story.

I want to know where the citation appears and how consistently a site earns citations across related searches. I also care whether the brand itself appears in the generated answer. Google can use information from a page to support its response without necessarily making that source or brand especially visible to the searcher.

So I think about AI Overview visibility across a few dimensions: 

• Are we being cited? 
• How visible is that citation? 
• Does the brand appear in the answer? 
• Are we showing up consistently across the searches that matter?

That gives me a much more useful picture than asking whether we “rank” in AI Overviews.

How to optimize for AI Overviews: What we’ve learned about earning citations

There’s no set formula for earning a citation. But the patterns we’ve observed, combined with what we now know about how Google retrieves and synthesizes information, give us a useful framework for deciding what to optimize for.

1. Map the information need

Once I’ve identified an AI Overview opportunity, I map out what someone actually needs to understand or decide: 

• What needs explaining? 
• What might they compare? 
• What questions or concerns are likely to follow?

Those information needs give me a much more useful starting point for planning the content than the target query alone.

2. Study what Google is already retrieving

I still believe competitive analysis matters, but I look at more than which competitors rank for the target query.

Look at which pages Google is citing and, more importantly, what information those pages contribute to the generated response.

Pay attention to the language and jargon those sources use, where they assume too much knowledge, and where a clearer explanation could make the information more useful.

I’m looking for gaps as much as patterns: 

• What questions are those sources answering? 
• What evidence are they providing? 
• Where are they vague, generic, or incomplete? 

Those gaps can tell us where our content has an opportunity to contribute something more useful.

3. Cover the dimensions that actually matter

Query fan-out makes it important to consider the relevant dimensions beyond a single answer. That might mean covering costs, benefits, implementation, ROI, comparisons, risks, requirements, standards, or other information that helps someone understand the topic or make a decision.

Those dimensions will look different by industry. In software, that might mean implementation, integrations, use cases, and ROI. In manufacturing, it could mean capabilities, specifications, processes, and applications. The goal is to identify the dimensions that actually shape understanding and decisions in your market.

Relevance still sets the boundaries. Comprehensive content should cover the dimensions that genuinely contribute to the searcher’s understanding, without adding adjacent topics simply to cover more ground.

4. Write clearly and directly

In my early testing, conversational language appeared frequently in content earning AI Overview visibility. I now think clarity and directness explain much of that pattern.

Google’s May guidance says publishers don’t need to write in a specific way for generative AI Search. Its systems can understand synonyms and general meaning without a page reproducing every long-tail variation of how someone might phrase a question.

Write clearly, directly, and naturally for your audience. 

If someone asks a straightforward question, give them a straightforward answer. If you’re defining something, make the definition explicit. If the answer depends on three conditions, identify them.

5. Let the information shape the structure

Process-oriented content stood out in my early testing, but I now think the information those processes surface matters more than the numbered format itself.

Processes naturally raise questions about requirements, decisions, costs, exceptions, and outcomes. Organizing those answers into steps can help readers, as can descriptive headings, logical sections, and concise explanations.

Google’s guidance also says its systems can understand multiple topics and nuances without publishers breaking information into tiny standalone “AI-friendly” chunks. Structure content in the way that makes the subject easiest for readers to understand.

6. Add detail people can’t get everywhere else

The amount of commodity content available has exploded. Think basic definitions, generic explanations, familiar tips, and summaries of information already available elsewhere, all relatively easy to reproduce without original knowledge or experience.

Google’s guidance specifically recommends unique viewpoints, firsthand reviews, and in-depth experience, and encourages publishers to provide information beyond what’s already widely available online.

For brands, some of the best sources of that information already exist within the organization:

• Proprietary datasets.
• Customer outcomes.
• Original research.
• Internal subject-matter experts.
• Product usage data.
• Testing.
• Implementation experience.
• Benchmarks.
• Expert analysis.
• A point of view developed from actually doing the work.

The goal is to identify what your organization knows, has seen, or can prove that another page couldn’t simply reproduce.

That can also change how we approach otherwise familiar content. A definition becomes more useful when it leads into real implications, examples, data, tradeoffs, or experience. A case study can explain the work itself while detailing what happened, how it was done, and what was learned along the way.

I think of content value as a progression:

• Commodity answer → clear explanation → dimensional understanding → unique evidence and experience.

Each step gives Google something more specific to retrieve and gives the reader something harder to find elsewhere.

How to measure AI Overview visibility

Measuring AI Overview performance has gotten easier, but no single metric tells the whole story.

In June, Google launched dedicated Generative AI performance reporting in Search Console, giving publishers first-party visibility into how their content appears across Google’s generative AI experiences. I’d use that data as the foundation, then supplement it with third-party monitoring for details Google doesn’t expose.

Start with first-party visibility data

Search Console’s Generative AI reporting gives us direct visibility into Google’s AI experiences, including impressions over time and the pages earning that visibility.

That gives me a baseline for answering questions like:

• Is our generative AI visibility growing or declining?
• Which pages are appearing most often?
• Is visibility concentrated on a handful of pages or distributed across the site?
• Are there meaningful changes by country or device?

I’m especially interested in trends over time. AI Overview results can change, so I wouldn’t draw a major conclusion from a single appearance or disappearance.

Add query and citation monitoring

First-party reporting doesn’t answer every question I care about. Third-party tools can help fill in the gaps by tracking specific queries, citations, competitors, and changes in the AI Overview itself.

For the searches that matter most, I want to know:

• Whether an AI Overview appears.
• Whether our site is cited.
• Where and how prominently that citation appears.
• Whether our brand is named in the generated answer.
• Which competitors or other sources appear alongside us.
• How consistently we earn visibility across related searches and over time.

That last point matters. I’m much more interested in a pattern of visibility across an important topic than a citation that appears once and disappears the next time we check.

Connect visibility to search performance

A citation is valuable information about visibility. It doesn’t tell us what happened next.

I also watch the familiar search metrics around the pages and queries earning AI Overview visibility, including impressions, clicks, and CTR.

Seer Interactive’s analysis illustrates why those measures need context. In its dataset, being cited in an AI Overview was associated with 120% more organic clicks per impression than appearing on an AI Overview SERP without a citation. But cited queries still generated 38% fewer clicks per impression than queries without an AI Overview.

That makes citation visibility and traffic useful measures to track side by side. A page can gain exposure within AI Overviews without producing a corresponding increase in clicks.

Tie AI Overview visibility to business outcomes

Ultimately, I want to know what that visibility contributes beyond the SERP.

Depending on the business, that could include qualified organic traffic, conversions, leads, branded search demand, assisted conversions, or another outcome tied to the role search plays in the customer journey.

That gives me a measurement stack that moves from AI visibility → search engagement → business impact.

AI Overview visibility is worth measuring in its own right. Its value becomes much clearer when we can connect that visibility to what happens afterward.

Where AI Overview optimization goes from here

After two and a half years of experimenting with AI Overviews, I’ve stopped looking for an AI-specific formula and started focusing on what makes a source genuinely useful to these systems.

SEO fundamentals still matter. But knowing that Google can retrieve information across related searches and synthesize it into an answer has changed how I approach optimization beyond them.

Rather than trying to replicate the characteristics of pages earning citations, I’m asking a different question: Across everything Google may need to understand about a topic, what does our content have to uniquely contribute?

That’s a more durable approach to AI Overview visibility than trying to reverse-engineer the latest citation patterns.