Study Finds Google AI Mode Alters 91% of URLs

A recent study has uncovered that Google’s new AI Mode tends to return different URLs for the same search query around 91% of the time. This sheds light on how the system selects and rotates its sources.

The analysis, which looked at 10,000 keywords, found that there’s only a 9.2% overlap in exact URLs across repeat searches for identical queries. This suggests a notably low level of consistency within AI-generated results.

Interestingly, the system often pulls in different pages, even when they come from the same website or domain. This inconsistency means that the specific content presented to users can shift significantly between searches.

Researchers pointed out that this volatility could create new opportunities for publishers. Since AI Mode doesn’t strictly rely on traditional ranking signals, it might surface a wider variety of pages over time.

The study was carried out by SE Ranking, which compared the AI Mode results against both Google’s AI Overviews and the classic top 10 organic listings.

Their findings show that the average overlap of exact URLs across these three datasets is just 9.2%. This figure highlights just how much AI Mode deviates from standard organic search.

This pattern also means that sites which might not consistently appear in the top organic spots could still gain traffic through AI Mode’s rotation of sources.

Despite the potential benefits for publishers, the lack of consistency could also create confusion for users who might see different answers each time they search.

The research further revealed that while AI Mode often draws on the same domains, it tends to highlight different articles or subpages from those sites.

Experts suggest this approach by Google could help reduce over-reliance on a few major sites and instead spread visibility more evenly across the web.

At the same time, the minimal crossover between AI Mode and traditional organic rankings raises questions about how these AI results are curated.

It remains unclear whether this volatility is intentional or simply a byproduct of how the AI system gathers and ranks information.

The findings add to growing discussions about the transparency of Google’s AI-driven features and how they influence user experience.

Overall, the study offers an early glimpse into how AI Mode could reshape search behaviour and content discovery.

As AI-generated answers become more common, both users and publishers will need to adapt to this new, less predictable landscape.

For now, the research suggests that the AI Mode experience on Google search may remain dynamic, offering fresh content each time but with only limited overlap to the familiar top organic results.

 

Highlights From The Study

AI Mode on Google has been shown to deliver highly varied results, even for the same search queries.

To test how consistent it really is, researchers ran the same 10,000 keywords through AI Mode three separate times in a single day. Interestingly, the results changed most of the time.

In over a fifth of cases—about 21.2%—there were absolutely no overlapping URLs across the three sets of responses. This means that, for many queries, the AI system picked entirely different pages each time.

When the researchers looked instead at the domain level, the consistency improved slightly. Around 14.7% of the time, AI Mode did still choose pages from the same websites, even if they were different individual URLs.

The study also explored how AI Mode results compared to the classic top 10 organic search listings. Only 14% of URLs shown in AI Mode were the same as those found in the organic results for the same keywords.

When the overlap was measured at the domain level rather than the exact page, this figure went up to 21.9%.

Yet in nearly 18% of the queries tested, there was no overlap at all between the AI Mode results and the organic URLs. This suggests AI Mode may rely on selection processes quite separate from Google’s normal ranking algorithms.

On average, each AI Mode answer included about 12.6 citations. These references came mainly in the form of block links, which made up 90.8% of all citations.

Smaller proportions were in-text links, which accounted for 8.9%, and AIM SERP-style links—designed to look more like regular search results—which made up just 0.3%.

Despite this wide variation, the study found that certain domains did appear regularly across multiple AI Mode runs.

Some of the most frequently cited sites included Indeed at 1.8%, Wikipedia at 1.6%, Reddit at 1.5%, YouTube at 1.4%, and NerdWallet at 1.2%.

Interestingly, Google’s own properties also featured heavily. Around 5.7% of all links came from Google-owned pages, with many of these pointing to Google Maps business listings.

The researchers also compared AI Mode directly to another of Google’s AI features: AI Overviews. Here, they discovered an average overlap of just 10.7% at the URL level.

At the domain level, the overlap was a bit higher, reaching 16%.

These findings point to the likelihood that AI Mode and AI Overviews rely on different underlying logic, even though both are powered by artificial intelligence.

Overall, the research illustrates how Google’s AI-driven features may highlight a broader mix of content—often differing from traditional search rankings—while still tending to draw from trusted, well-known domains.

 

What This Means For Search Marketers

The unpredictable nature of AI Mode results is proving to be both a challenge and an opportunity for website owners and marketers.

Since the results can change even when users type in the same search query, it becomes harder to track and measure consistent visibility over time.

Yet, this very unpredictability might actually be a good thing for some publishers. Instead of the same few high-ranking pages dominating the results day after day, AI Mode seems to refresh its citations regularly.

This means there’s a greater chance for different sites to appear in AI answers, even if they don’t manage to break into the traditional organic top 10.

In this environment, publishers who produce relevant, well-written content could see more chances to be featured—even if they’ve previously struggled to compete with bigger brands in classic search.

To keep up, SEO specialists and content creators might want to focus on strengthening their overall domain authority and ensuring their content thoroughly covers key topics.

It could also be wise to spread content out across a mix of trusted platforms, rather than relying on a single website to carry all the weight.

Another useful step is to pay closer attention to local SEO tactics, such as building up a strong Google Maps profile or local business listings, as these may influence AI Mode results too.

Watching how often a website or brand appears in AI Mode answers could reveal useful patterns over time, helping teams refine their strategy.

The key takeaway is that AI Mode is more dynamic than traditional search results, and this fluidity could favour publishers who remain flexible and proactive.

Those who stay alert to changes and keep creating content that’s both helpful and relevant stand a better chance of gaining exposure.

As Google continues to develop and tweak AI Mode, it’s likely the way citations are chosen will evolve too.

Being ready to adjust strategies in response to these shifts will be crucial for staying visible in AI-powered search features.

Overall, while AI Mode brings complexity, it also offers fresh opportunities for websites willing to adapt and diversify their presence.

By combining quality content, strong domain reputation and smart local SEO, publishers may benefit from the changing landscape rather than being left behind.

Keeping an eye on ongoing developments and experimenting with different approaches will be essential as AI-driven search becomes a bigger part of how users discover information online.