10 Observations about AI Overviews after a full Month

It’s been over a month since Google officially rolled out AI Overviews to broader users.  The transition from SGE to AI Overviews yielding some interesting changes that we anticipated and tracked with BrightEdge Generative Parser™. As with any new technology, there were some challenges right out of the gate that Google swiftly addressed and today we’re seeing some of the effects of those changes.

It’s still very early days, but we can already see that AI Overviews are evolving quickly. Some changes are subtle, while others impact entire industries. AI Overviews are increasingly focusing on specific queries and delivery methods that effectively help users obtain the information they need to make the best decisions.

To help you better understand what AI Overviews are and how to optimize content for them, we’ve published our Ultimate Guide to AI Overviews, which you can read here. We encourage you to bookmark it, as we will be making regular updates as new data becomes available.

Here are ten findings we observed with BrightEdge Generative Parser™ over the month of June that highlight how AI Overviews are evolving:

1. AI Overviews presence went from 11% to 7% of queries in June

A month before Google officially rolled out AI Overviews in mid May, we observed a rollback on AI results in the Google labs. That trajectory continued through May, and throughout June, we saw that they scaled back even further.

We saw a slight uptick in AI Overviews around the 17th that continued until the 25th. This increase was almost entirely related to education and e-commerce queries. The data points to Google continuing to refine and deploy AI Overviews with greater precision to help users most effectively.

2. Some Industries had big swings in their AI Overview presence.

In some industries, AI Overviews were consistent throughout June. However, as the month went on, we observed dramatic drops in visibility for keywords related to certain industries. This suggests that, for these industries, users were not finding the AI Overviews as helpful.

AIO Presence goes from 26% to 13% of Education keywords in June:

In education, we observed the most significant changes pertaining to broader informational queries related to degrees and specific tests. However, there is a higher likelihood of AI Overviews appearing when users search for specific skills, such as business development skills or prompt engineering skills.

AIOs Presence for Entertainment went from 14% to nearly 0% in June
 

Just before the end of May, we observed roughly 14% of entertainment queries showing AI Overviews. Interestingly, there was a positive correlation when users searched for golden age Hollywood celebrities. However, by the beginning of June, these AI Overviews disappeared. This is likely because many of these queries also featured knowledge graphs and universal results, where AI Overviews did not add significant value to the user.

Traditional search result of a golden age Hollywood actor that already features extensive

enhancements and summaries.  AI Overviews are no longer appearing for results like these.

AIO presence on E-Commerce keywords went from 26% to just 9%
 

We saw significant reductions in queries specific to e-commerce in June, although there was a brief uptick from the 17th to the 25th, which, as mentioned above, impacted overall visibility.

3. AI Overviews became 13% smaller during the month of June

Google’s test with showing more education and e-commerce AI Overviews from the 17th to the 25th also resulted in slightly larger AI Overviews—presumably due to product viewers and carousels taking up more space on the page. However, once that test completed, we saw the overall size reduce again. This continues a trend we’ve been observing with BrightEdge Generative Parser™ since March, when AI results were still in the lab.

In fact, at the end of June, AI Overviews are half the size on average compared to Google’s SGE results at the beginning of March! This refinement suggests AI Overviews are getting better at distilling content down to what is needed and eliminating unnecessary or unhelpful content from the results. For marketers, this also demonstrates how AI Overviews do not crowd out other important aspects of a search result, such as traditional organic or paid listings.

4. AI Overviews are continuing to reduce information that overlaps with traditional results

When AI Overviews were launched, one of our findings was  that the overlap in AI Overview citations and traditional organic results was significantly reduced. We correlated this with Google’s description of AI Overviews as being able to do the searching for you, which would eliminate the need to cite the same sources you’d already see in traditional results. Instead, AI Overviews anticipate the second and third searches you may do and source information from those places. In June, AI Overviews appeared to get better at this, reducing the overlap in results by an additional 3% across all queries.

Note that during Google’s test on the 17th, we observed a slight uptick in citations and traditional rank, which corresponded to the e-commerce lift. As Gemini—the LLM that powers AI Overviews—gets better at anticipating queries and AI Overviews are deployed with greater precision, we expect this disparity in overlap to continue on its current trajectory.

5. There are patterns where AI Overviews appear for certain keywords more than others

Over the course of June, we observed that AI Overviews were either increased or decreased for specific search behaviors. These changes are particularly helpful in understanding the role AI is intended to play in a user’s purchase journey. Here are some noteworthy changes observed throughout the month of June:

Increases:

• For “best” keywords: There was a 50% uptick in the presence of AI Overviews for keywords such as “best men’s shoes”, “best watches”, etc. (this could then be followed by a potential explanation).
• For “what is” questions: There was approximately a 20% AIO increase in keywords starting with “what is” (e.g., “what is data analytics”, “what is EDR”, “what is a mutual fund”). This suggests an increase in informational queries.
• For “how to” queries: There was about a 15% AIO increase in keywords beginning with “how to” (e.g., “how does cloud networking work”, “how to prevent kidney failure”). This indicates a rise in process-oriented searches.
• For “symptoms of” searches: There was roughly a 12% AIO increase in keywords containing “symptoms” or “signs of” (e.g., “syphilis symptoms”, “signs of Crohn’s disease”). This shows a growing interest in identifying health conditions.
• For keywords related to data governance, analytics, and cloud technologies: There was approximately a 10% AIO increase in keywords such as “data governance”, “what is data analytics”, “how does cloud networking work”.
• For “treatment” related queries: There was approximately a 10% AIO increase in keywords including “treatment” (e.g., “shoulder tendonitis treatment”, “hip tendonitis treatment”). This suggests increased interest in medical interventions.

Decreases:

• For “Vs” comparisons: Similarly, there’s approximately a 20% AIO decrease in keywords containing “vs” for product or concept comparisons. This could be a result of avoiding hallucinations.
• For Brand-specific queries: AIOs appeared for 15% fewer keywords contain specific brand names (e.g., fewer mentions of companies like “Samsung”, “Google”, or “KitchenAid”). Similar to vs, this could be an area where Google is concerned about hallucinations
• For general product categories: AIOs were present for about 14% fewer keywords relate to broad product categories (e.g., fewer mentions of “shoes”, “furniture”, “appliances”).
• For lifestyle-related queries: Approximately 12% less keywords related to lifestyle, fashion, and home decor included AI Overviews.
• For Product model numbers: Roughly 10% less keywords containing specific product model numbers or technical specifications displayed AI Overviews. These are purchase intent queries so AI may not be adding value given all the other features of the search result that are present.
• For Seeking general tech support: AIOs presence was reduced by approximately 25% for basic tech support queries. It could be that AI Overviews have difficulty adding value to these queries when Featured Snippets and other features are already present. For instance, examples where AIO no longer appears are keywords such as:
  •    – “how to reset router”
  •    – “troubleshoot printer issues”
• For Seeking basic health advice: AIOs were removed in approximately 15% of general wellness queries.  Similar to above, these queries may likely not see value from an AI Overview due to volatility of the answers. Examples of keywords where AIO’s do not appear at the end of June include:
  •    – “how to lose weight”
  •    – “natural remedies for common cold”

These changes suggest that AI overviews are more likely to be triggered by users seeking in-depth, specialized information, particularly in medical, technological, and financial domains.

This shift indicates a trend towards using AI for complex, knowledge-intensive topics rather than for everyday consumer decisions or basic troubleshooting that could be solved by elements in universal results.

6. Reputable sites are sourced by AI, with much less reliance on UGC content.

One of the challenges in May that Google addressed was the reliance on trustworthy content for AI citations. In June, the data shows that, particularly for medical content, AI Overviews are relying on authoritative and trustworthy sources to help users understand symptoms and other healthcare queries that show AI Overviews.

• Citations for cdc.gov show an overall increase of approximately 35% from the beginning to the end of the month. This suggests a growing reliance on official public health guidance in AI-generated responses.
• Citations for wikipedia.org declined 28% from the start to the end of the month. This may indicate a shift away from general knowledge sources towards more specialized, authoritative references.
• Citations for medicalnewstoday.com decreased by approximately 40% from the beginning to the end of the month. This could suggest a reduced focus on current health news in favor of more established medical sources.
• Throughout June, there were day-to-day fluctuations of up to 30% in citations for individual domains, even though the top sources remained consistent. This suggests that AI Overviews adapt their sources based on daily query patterns while still prioritizing established, reputable sources.

Concurrently, we’ve witnessed a steep decline in the citations from UGC sources that might contain content that is hard for the LLM to distinguish from authoritative ones. In fact, near the end of May, we saw Reddit and Quora citations fall off almost entirely in AI Overviews. This drop corresponds to statements Google has made comments about some difficulties AI has in detecting sarcasm and other linguistic nuances that make it difficult to understand when a source doesn’t necessarily mean what it says. On May 30th, Google’s Head of Search Liz Reid made a post on the official Google Product blog stating that they were limiting the use of user-generated content in responses. The data shows that this limitation resulted in a near complete drop-off for Reddit and Quora.

We saw a similar trajectory for citations for Quora occur:

For marketers, this should mean AI Overviews are less prone to hallucinations and giving customers false information.  It also reinforces the need for trustworthy content that demonstrates E-E-A-T. 

7. Comparison tables appear for half as many queries as they did on June 1

As noted above, one area where we see AI Overviews appearing less is around specific brand queries. This could be to avoid the possibilities of hallucinations. As a result, when users are comparing brands or specific products, AI Overviews appear less frequently, and we see fewer modules for pros/cons or versus comparisons.

Interestingly, this drop-off appears to occur around the same time we saw a reduction in e-commerce queries. We are now seeing a slight uptick in keywords that only show the “pro,” which could indicate Google is looking to avoid hallucinations that could say negative things about brands.

8. Product Viewers and Carousels are appearing significantly less

Since the beginning of June, we are seeing less keywords that feature product carousels or product viewers in the results:

As this chart shows, there was an uptick when we saw an increase in e-commerce keywords showing AI Overviews. However, when that uptick decreased, so did the carousels showing AI Overviews. This doesn’t necessarily translate 1:1 with e-commerce, as we still find instances where AI Overviews provide guidance for the user without showing them product viewers or carousels.

(example of a query Google interprets to be e-commerce but does not serve a product carousel in the AI Overview)

9. Ordered and Unordered Lists are appearing less:

One of the most common types of modules in AI Overviews are unordered lists. We describe in detail how they work and what marketers can do to help increase the likelihood of a citation for them in our Ultimate Guide to AI Overviews. However, throughout June, we observed a significant decrease in their presence.
 

This may be a result of Google preferring snippets and paragraphs for broader queries that are on the rise (e.g., symptoms) which lend themselves to narratives and bullet points as opposed to unordered lists.

AI Overview showing simple bullet points and paragraphs

Unordered List as they appear in AI Overviews

This may also contribute to the smaller space AI Overviews are taking up. As Google looks to find ways to deliver more compact results that distill information, we expect the modules they deploy to be more precise in where they are used.

10. 10% increase in Financial Warnings

For YMYL (Your Money or Your Life) queries, AI Overviews offer a warning in addition to cautions about generative AI being experimental, advising that professional advice should also be sought. These warnings have been mostly prevalent for healthcare queries, but throughout June, the data showed Google was increasing these for financial queries. After an uptick in the middle of the month, we are now settling at a rate that is 10% higher than on June 1:

These financial warnings are a snippet that is appended to the disclaimer about generative AI:

This uptick may suggest that Google is protecting AI from liability around sensitive topics.

It’s an exciting time as AI Overviews unfold, and we can learn more about how they will help customers and shape digital marketing! It’s clear from the data that AI Overviews are carving a unique role in the search landscape that doesn’t simply duplicate elements that already work well. Marketers have a great opportunity to expand their reach and leverage AI to address their customers’ needs before they even click on a website!