We analyzed 405,576 AI overviews in Google’s search results. Three findings stand out: AI overviews are 157 words long on average, they cite an average of 5 sources per query, and 52% of the sources they mention also rank in the top 10 organic results.
Formatting is just as consistent: 78% of overviews contain either an ordered or an unordered list. AI overviews appear above traditional SERPs and featured snippets, so being cited by one is a prime spot on the page. Below are 25+ statistics from our dataset, covering answer length, sources, formatting, e-commerce and user-generated citations, and how the domains cited in AI overviews compare with top-10 domains on traffic and keyword visibility.
Study methodology
This study is based on 405,576 AI overviews from Google’s traditional search engine result pages, all from queries classified as English. For every query we captured the full top 10 organic results alongside the sources cited in the AI overview, which let us measure how often Google’s AI-generated answers draw on pages that already rank. "People Also Ask" results were stripped out so they could not be counted as search results. A separate visibility analysis used 157,005 queries for traffic data and 156,826 queries for keyword data. We found no significant difference between AI overviews with Search Labs turned on and turned off.
Our research focuses on analyzing the performance and characteristics of AI overviews [formerly Search Generative Experience or SGE] in Google’s traditional search engine result pages.
We looked at real queries users put into search.
Our final dataset contained 405,576 examples after filtering.
Here’s how we ended up with this sample.
We analyzed how often Google’s AI-generated overviews selected results from the top 10 search results only for English queries.
We reviewed over 400,000 examples where we had full data for the top 10 search results and checked how frequently Google's AI overview picked those high-ranking results when presenting information.
We took careful consideration to remove "People Also Ask" results from the dataset so they would not be mistakenly classified as search results, resulting in a clean dataset.
Here’s an overview of our data:
- We analyzed only queries classified as English.
- Our analysis focused on the top 10 results from the SERP.
- After filtering, we ended up with 405,576 examples with complete top-10 results.
Our research found no significant difference between AI overviews when Search Labs is turned on and when it’s not.
What are Google AI overviews?
AI overviews in Google Search are summaries that answer a query using generative AI. They give quick answers with links to web pages for users to explore the topic further.
These overviews also include brief explanations, lists, or highlights of the most relevant points of a topic.
So, instead of clicking on a page and reading through an article, users can extract the most important information at a glance. They also sit above the traditional organic results, and in our dataset each overview cited an average of 5 sources.
Here’s an example.
The generated answer also includes a link to its source so you can see where Google picked up the information from.
Often, these are different sources on the topic to provid e a well-rounded answer.
In the right panel, the AI overview snippet also includes sources to web pages that Google deems trustworthy and authoritative on the subject of your search.
Being picked as a source by AI overviews means you get a prime spot on SERPs, above other organic results.
But how can you show up as a source for AI overview answers?
We did some digging.
Characteristics of AI overviews
Across 405,576 examples, the typical AI overview is short, heavily sourced, and closely tied to the existing top 10. AI overviews run 157 words on average, roughly 984 characters, and 99% of them come in under 328 words. They mention an average of 5 sources per query, and 90% of the time Google lists 8 or fewer. Repetition is rare: 99% of sources are referenced only once per answer. And 52% of the sources cited also rank in the top 10 organic results for that query, at an average position of around 5. Here are the basics of AI overviews in detail.
AI overviews are, on average, 157 words long
AI overviews in Google search are, on average, 157 words long. That’s around 984 characters.
99% of answers fall under 328 words.
They focus on delivering quick, digestible information.
Yet, AI overviews expand on a topic more than featured snippets, which typically are no more than 50 words, depending on the type.
Meanwhile, the AIO in this example is around 160 words when you expand beyond “Show more” - close to our average.
And as you can see, AI overviews appear above featured snippets.
AI overviews also pull information from different sources, which users can then click on to dig deeper. You can see that from the link icon at the right of each sentence.
AI overviews contain the exact query 5.4% of the time.
The fact that AI-generated overviews contain the exact query 5.4% of the time shows that AI paraphrases or synthesizes information rather than directly replicating the exact input.
This indicates that the AI answer is designed to focus on providing context, meaning, or relevance rather than just keyword matching.
Look at this example.
The exact query is “best month to visit thailand”. Meanwhile, the AI answer does not include the word “month”, it includes “best time”.
What does this mean for your on-page optimization process?
This isn't new information since we know that Google has been using NLP processing since the Hummingbird update in 2013.
But this is further proof that you should prioritize natural language in your content rather than trying to meet a certain keyword target.
You still need to communicate to search engines what your page is about with the help of keywords.
But you don’t need to stuff every relevant keyword.
As long as you optimize your content for the right terms and topics, Google will understand the context and might include you as a source in AI overviews.
AI overviews mention an average of 5 sources per query
90% of the time, Google lists 8 or fewer sources.
This shows us that the AI overviews don’t always generate an exact number of sources every time.
Understandably, the number of sources depends on what Google determines the intent of the query to be and the availability of credible information.
In less than one per cent of the cases, 0 sources are mentioned.
Presumably, this is when a search query may be too niche and have little search behaviour for Google to determine a good response.
99% of the sources are referenced only once per answer
It is very uncommon for AIOs to reference the same URL more than once within different sections of its response.
The same URL showed up more than once in less than 1% of the time in our data.
For example, even though this answer is specific to Surfer and cites our domain, it’s taking information from different pages.
52% of sources mentioned by Google AI overviews rank in the top 10 results
We studied how often the sources mentioned in the AI overviews also appear in the top 10 results in the traditional search engine results.
More than half of the sources also rank in the top 10 SERPs for that query.
It’s quite evident that Google’s preference for AI overviews is skewed towards web pages that are already performing well for the target query.
The sources mentioned by Google that also appear in the top 10 results of the SERP rank around position 5 on average.
We found an even distribution of higher-ranked results in positions 1–5 show up in the AI overview compared to other pages in positions 5-9.
What do AI overviews look like?
AI overviews are formatted in short paragraphs and as lists, and lists dominate. In our data, 61% of AI overviews contain an unordered list, 12% contain an ordered list, and 6% contain both types, while 21% contain no list at all. In 78% of cases an overview carries one or both list types. Ordered lists tend to appear for step-by-step queries. Unordered lists show up even for "top X" searches, including when the pages Google cites use numbered lists themselves. When an overview has no list, the answer is usually broken into short paragraphs instead. You cannot control which format Google picks, but list-friendly content and clean H2 and H3 structure help.
Here’s the breakdown of how they look like.
12% of AI overviews contain an ordered list
Ordered lists, appear in only 17% of the answers.
While not always the case, AI overviews often show up as ordered lists when you ask for step-by-step instructions.
Meanwhile, there’s another story when it comes to unordered lists.
61% of AI overviews contain an unordered list
Unordered lists are quite a more common way AI responses are structured.
AI overviews use unordered lists even when answering queries like “Top X [product/tips]”.
That’s even when they pull information from sources that have top 10 included lists.
6% of AI overviews contain both an ordered and unordered list
Sometimes, it happens that AI overviews contain both types of lists in their information.
However, this is more rare than the other instances.
21% of AI overviews don’t contain any lists
And in 22% of the cases, AI responses contain neither an ordered nor an unordered list.
Yet, the answers tend to be formatted into short paragraphs, making the content more digestible.
In 78% of the cases, they contain either one or both types of lists.
It’s not surprising, as lists are skimmable and a great format for summarizing information, which is what AI overviews do.
They’re good content formatting practice.
That’s to say, there’s not much you can do to affect whether your content appears as an ordered or unordered list.
However, writing “lists,” “how-tos,” and “step-by-step guides” may help you get there. Also, properly format your content using H2 and H3 titles.
AI overviews for e-commerce
E-commerce sources are rare in AI overviews. Only 0.3% of the overviews we analyzed contain an e-commerce source, so shopping and product-listing pages are a very small slice of what Google cites overall. When an e-commerce query does involve product pages, the pattern changes: 72% of those AI overviews contain 6 sources from e-commerce web pages. We identified these sources from the shopping-specific markup Google attaches to them, covering product offers, apparel products, general shopping products, and entity feedback such as reviews. Those links point users towards buying a product rather than reading about it.
We also looked at how sources typically related to shopping or product listings are identified and categorized by Google’s AI overviews.
These are web pages with the following HTML attributes.
- "SGEEntityFeedback": This tag might indicate feedback or reviews associated with a product or e-commerce listing.
- "SGEShoppingOfferCard": Likely refers to a card-style presentation of a product offer, similar to product ads or listings commonly seen in online stores.
- "SGEApparelProduct": Indicates a product related to clothing or fashion, suggesting that SGE can identify specific types of products (like apparel).
- "SGEShoppingProduct": A more general tag for any product that is part of a shopping or e-commerce experience.
These URLs direct users to buying products and include product listings, shopping cards, or reviews associated with e-commerce platforms.
0.3% of overviews contain an eCommerce source
This means that a small proportion of overviews include such sources.
For e-commerce-related queries that do involve product pages, 72% of AI overviews contain 6 sources from e-commerce web pages
AI overviews can identify and prioritize links related to products and shopping, making it easier for users to find relevant shopping options.
User-generated sources in AI overviews
User-generated sources are just as rare as e-commerce ones. Only 0.3% of the AI overviews we analyzed contained a user-generated source, meaning reviews, feedback, or content contributed by individuals rather than an official website. In the under 1% of cases where AI overviews do include user-generated content, they include 2 user sources. We identified them through the markup Google uses to attribute content to a user and to flag blocks of user reviews. For most queries, then, AI overviews are built from publisher pages rather than community or review content.
The same is seen when it comes to user generated sources in AIOs.
Our study analyzed sources of user generated information in the form of reviews, feedback, or content contributed by individuals in the AI overview snippet, rather than official websites.
The HTML attributes we studied are:
- "SGEAttributionUser": This tag indicates that the content is attributed to a user or is user-generated. It’s likely used to flag reviews, comments, or any content created by individual users.
- "SGEReviewBlock": This likely refers to a block of content specifically structured for user reviews, such as a section of a website that compiles multiple user reviews for a product, location, or service.
Only 0.3% of overviews contained a user-generated source
In under 1% of the cases they do contain a user-generated source, AI overviews include 2 user sources.
So there's only a very small proportion of such sources.
AI overviews and domain visibility
The domains cited in AI overviews tend to be smaller than the domains ranking in the top 10 for the same query. In 69% of cases, the domains in the top-10 search results have higher average traffic than the domains cited in the AI overview, and in 71% of cases those top-10 domains rank for more keywords. The comparison covers 157,005 queries for traffic and 156,826 queries for keyword data. It does not measure clicks. What it does show is that a site outside the top 10 can still be cited in an AI overview and gain visibility there.
Next, we analyzed domain visibility by comparing traditional top-10 results and AI overview sources.
We only looked at user queries for which we could get the traffic and keyword rank data for all domains in top-10 results and AIO sources.
We ended up with:
- 157,005 queries for analyzing traffic
- 156,826 queries for analyzing keywords
Here’s what we found.
In 69% of cases, domains in top-10 have higher average traffic than those cited in AIOs
Domains in the top-10 search results typically receive more visits than the domains cited in AIOs.
That’s not to say that people click on SERPs more than on the sources cited in AIOs - that’s not what we are measuring here.
It means smaller websites not in top-10 can show up in AI overviews.
It could be a good chance for such pages to gain visibility through AIOs.
71% of the time, top-10 domains rank for more keywords
In 71% of cases, domains in traditional top-10 SERPs outperform those in SGE in terms of how many keywords they rank for.
This means that top-10 SERP domains are more frequently visible across a broader range of search queries compared to those cited in SGE summaries.
Even though in many instances, the difference is around zero. This means that the average number of ranked keywords in SGE and top-10 results is often quite close for many queries.
Ultimately, the majority of traditional top-10 results maintain better keyword rankings.
What AI overviews mean for your SEO strategy?
AI overviews reward the same content that already performs in organic search. 52% of the sources they cite also rank in the top 10 for the query, so ranking well remains the most reliable route into an overview. In 69% of cases the top-10 domains have higher average traffic than the domains cited in the overview, which means smaller sites do get pulled in. Because answers average 157 words and cite around 5 sources, the pages that win citations are the ones that answer a question cleanly and can be quoted in a sentence or two.
So there is no separate AI overview checklist to work through. Cover each topic thoroughly, optimize for the right terms, keep your pages updated, and keep formatting skimmable with clear headings and lists. Tools like Surfer help you stay ahead in the SEO content game.
.avif)
