AI Mode volatility: Results from our local search test

In our first round of AI Mode research, we found that volatility was one of its defining features. This naturally raised another question: how does that volatility impact local searches?

For example: if you search “restaurants near me” in New York and repeat the same query a few minutes later, do you get the same results? If someone in Denver searches at the same time, will their results match yours? And what happens if the query specifies a city name instead of just “near me”?

What we learned gave us a clearer picture of how AI Mode behaves in real searches and also helped us improve the way we approach positions in the AI Mode Tracker.

Now, let’s move on to our findings.

Domain-level results

At the domain level (looking at whether the same websites were cited, regardless of the exact page), AIM’s behavior splits sharply depending on query phrasing.

General local queries 

If you search for “restaurants near me” several times in New York, you might expect most of the same sites to appear each time. Instead, AIM reshuffled results almost every run. The same happened with “gas station nearby” in LA, “coffee shops near me” in Chicago, and other general local queries.

The numbers below show the average overlap (how many of the same domains repeated across runs) and volatility (how often results changed) for a sample of cities:

That means more than 60% of the domains disappear between runs, even with the same user, city, and query.

Volatility grew when we ran the same general local queries across multiple cities:

Here, less than one in four domains carried over. A New Yorker and a Houstonian searching “restaurants near me” will likely see AIM results based on data from different websites.

This means the user’s geographic location has a significant impact on AIM results for general local queries.

Location-specific queries 

When we added a city name (e.g., “restaurants in Denver”), the pattern flipped. Overlap nearly doubled compared to general local search queries:

So, by adding the city name to the query, you can expect AIM to show a more consistent set of domains in its responses.

Even across locations, city-specific phrasing held up well:

So while AIM still adjusts its results based on the user’s location, the effect is much smaller once the query includes a location reference.

Distribution of domain overlap

When we look at how domains are distributed across AIM responses, the difference is striking. 

For general searches with local intent (say, “best coffee shops near me”), only about 12-16% of the keywords show over half the same websites when the query is repeated in the same city. In other words, the results shift around quite a bit if the location isn’t spelled out.

But when the query is tied to a specific place (e.g., “best coffee shops in Denver”), that overlap climbs to 40-44%. 

Even more remarkably, in more than a third of these city-specific searches (35-37%), the overlap goes beyond 90%. That means AIM is essentially giving you the same set of domains each time.

AIM can deliver consistent, stable results, but only when the query makes the location explicit. For example, users who specify a city in their query (Denver, LA, Arizona, etc.) tend to see more consistent overlaps than those who leave the location implicit. If the city isn’t specified, the system has more room for interpretation, and the domains it surfaces are much more likely to change from one run to the next.

URL-level results

At the URL level, volatility becomes even more visible. Here, AIM isn’t just swapping websites but entire pages.

General local queries 

Within the same city, URL overlap hovered around 18–20%:

That translates to volatility rates of 80–82%. Simply put, four out of five URLs disappeared between runs for the same user, location, and search query.

Across cities, stability all but collapsed:

In other words, AIM rebuilt its answer set from scratch. The chance of two people in different cities seeing the same URLs was close to zero.

This makes sense when you think about it. If you’re in Los Angeles, you’d expect local suggestions, not tips from a Denver or Atlanta guide. To ensure this, AIM places local context at the center of its results. So, this approach feels entirely expected.

Location-specific queries 

Once again, adding a city name doubled stability:

Now, nearly half the URLs are repeated from run to run.

Even across user locations, city-specific queries retained relatively strong stability:

This is far from the near-zero overlap of general local queries across cities.

Distribution of URL overlap

At the URL level, the patterns are also interesting. 

For general local queries, only about 1.3-1.8% of keywords reached 90% or higher overlap. In practice, that means the exact pages AIM surfaces usually shift almost completely between runs.

When the query references a specific location, however, the stability improves dramatically. In these cases, about 34-36% of keywords hit the 90%+ overlap mark. This means AIM is far more likely to return the same set of pages consistently when the location is explicit.

AIM doesn’t just shuffle which domains you see. It often changes the exact pages as well. But once the city is part of the query, the results become far more predictable.

What drives AIM volatility (and what doesn’t)?

Let’s first take a look at the overlap of sources across AIM answers. The table below shows how much results overlap at the URL and domain levels, depending on whether queries are general or location-specific, and whether they’re run in the same location or across different ones. 

As you can see, general local queries are highly volatile: when repeated in the same location, they share only 19% of URLs and 35% of domains, and across different locations, the overlap drops to just 2% of URLs and 24% of domains. In contrast, location-specific queries are far more stable: within the same location, they share nearly 48% of URLs and 55% of domains, and even across locations, the overlap remains relatively high at 43% of URLs and 49% of domains.

Now, let’s look at the bigger picture: what these overlaps tell us about when AIM gives consistent results and when it doesn’t.

1. Query phrasing is the biggest driver.
Specifying the location makes AIM far more predictable. Broad prompts like “restaurants near me” produce unstable results, with overlap dropping into the 30% range. Adding a location reference (whether it’s a city, a neighborhood, or even a specific place like “restaurants near the White House”) nearly doubles stability, both at the domain and URL level. 

2. User location only matters for general local queries.
Geography plays a role when the query is vague. For “near me” searches and similar, moving between cities reduces overlap by 10–12% (from the mid-30s to the low-20s). But once the query includes a city name, the effect of user location nearly disappears, shifting results by only 2–6%.

3. Domains are steadier than URLs.
Even when AIM repeats the same websites, it often cycles through different pages. For general local queries, more than 80% of URLs change between runs, compared to roughly 60–65% turnover at the domain level. That’s only natural, since AIM shifts sources for local queries depending on local context (what’s nearby, region-specific content, or what’s most relevant in that area at the time).

4. Patterns are consistent across cities.
The results were strikingly similar in New York, Los Angeles, Washington, Houston, and Denver. No city proved more stable than another, which means the volatility patterns aren’t tied to local markets. They’re built into how AIM handles local queries.

5. Distribution of overlap reinforces the trend.
Looking at overlap distributions highlights just how different the two query types are. For general local queries (without a specific location mentioned), only 12–16% of keywords returned more than half the same domains, and just 1–2% reached 90%+ overlap at the URL level. In contrast, city-specific queries were far steadier: 40–44% of keywords crossed the 50% domain threshold, and 34–36% hit 90%+ overlap at the URL level.

In short: volatility depends more on how the query is phrased than on where it is run. General local queries leave AIM with a broad context to interpret, so the results vary widely. Once a city name is included, however, the search space narrows. This reduces ambiguity, anchors AIM to a consistent pool of sources, and minimizes the impact of user location.

Research methodology

The dataset of 5,000 keywords was divided into two equal groups:

Each group contained 2,500 queries. For city-specific queries, we selected five cities and assigned 500 queries to each:

Each query was parsed 15 times (three runs per city) to capture geographical and temporal variation.

We compared the results under four conditions:

Overlap was measured at two levels:

• Domain level: whether the same websites appeared in the AIM results.
• URL level: whether the exact same pages appeared in the AIM results.

This approach allowed us to isolate the effects of query phrasing and user geography while controlling for timing and repetition.

Disclaimer: While we aim to provide the most objective interpretations, we recognize that other valid perspectives may also exist.

Conclusion

Our tests show that AIM volatility is real, but there are clear factors that shape these AI responses.

When queries are vague, the system has too much freedom, and the results shift dramatically. But when the query clearly names a location (and especially when that location matches the user’s own), the results settle into a stable pattern.

Written by

Yulia Deda

Yulia is an SEO and Content Marketing Expert at SE Ranking, where she dives deep into the worlds of SEO, AI, and digital marketing. Her passion is making industry research easy to understand and practical for everyday marketing. Outside of work, Yulia enjoys long walks, cozy evenings with friends, and exploring different cultures through food and travel.