Mapping the Volatility of Local AI Overviews: A 32-Day Case Study

Local AI Overviews tracking remains a moving target for search professionals, as recent longitudinal data suggests that brand citations within these summaries are far from stable. Last updated January 29, 2025, a month-long analysis of a single local entity shows that the presence of a business in AI-generated responses can fluctuate wildly even when the underlying website content remains static.

Recent data shared by an independent search analyst on Reddit provides a rare look at the raw frequency of these AI summaries. By monitoring 1,814 specific AI-generated answers for a single local business over a 32-day window, the data revealed a landscape defined by inconsistency. Rather than a steady climb or a fixed position, the business saw its inclusion in AI Overviews toggle between presence and absence without changes to its Google Business Profile or core site architecture.

For a 12-location HVAC operator, this level of volatility makes traditional monthly reporting nearly impossible. We have observed that if a brand appears in an AI Overview on Tuesday but vanishes by Thursday for the exact same query, a single-day ranking snapshot becomes a liability rather than an asset. The citation frequency does not yet follow the linear path established by traditional organic blue links, where positions generally move in smaller, more predictable increments. A dental practice in Leeds might find itself featured as the primary recommendation for an "emergency dentist" query at 9:00 AM, only to be entirely omitted by noon, even if its star rating and proximity remain superior to the competitors that replaced it.

The challenge of Local AI Overviews tracking

The primary difficulty in modern measurement is that generative results are not static assets. Traditional search results are generated from an index that is crawled and updated; AI Overviews, however, are synthesized dynamically. This creates a situation where the "truth" of a ranking depends entirely on the moment the query is executed. In the 32-day study, the business was cited in approximately 40% of the 1,814 instances, but those citations were not clustered. They were scattered across the month in a pattern that looks more like noise than a signal of authority.

We see this as a fundamental shift in how local visibility is earned. In the past, Google’s local algorithm relied heavily on the "distance, prominence, and relevance" triad. While those factors still matter for the traditional Local Pack, the AI Overviews seem to be pulling from a broader and less stable set of data points. A dental practice in Leeds, for instance, may be displaced by a competitor simply because the competitor recently published a blog post that the LLM perceives as a more "fresh" answer to a specific long-tail question, even if that competitor's physical office is further away from the searcher.

Why is local search generative AI so unstable?

The instability appears rooted in the tension between Google's retrieval-augmented generation (RAG) and the underlying LLM. According to Google Search Central's guidance on AI-generated content, the system prioritizes high-quality information that demonstrates experience and expertise. However, in practice, the real-time synthesis of these answers allows for "hallucination" or omission of specific brands based on subtle shifts in query phrasing or minor updates to the model's weights.

For the 12-location HVAC operator archetype, this means their visibility is at the mercy of how well the AI interprets their service area across different sessions. Unlike the traditional Local Pack, which has a relatively fixed geographic boundary, the AI Overview might decide a business is relevant for a query one moment and irrelevant the next. This lack of a persistent "cache" for AI answers means that tracking tools are often capturing a single, potentially anomalous state of the SERP rather than a stable reality.

How does this compare to traditional Local Pack rankings?

Before the rollout of AI Overviews, a business could reasonably predict its traffic based on its position in the Map Pack. The data from this 32-day case study indicates that AI Overviews act more like a "rolling carousel" than a fixed leaderboard. While the Local Pack remains relatively consistent for a given geographic center, the AI-generated text above it may cite a completely different set of businesses from one hour to the next.

Google Business Profile documentation emphasizes that relevance and prominence are key to improving local ranking. However, AI Overviews seem to introduce a fourth, invisible variable: "synthesized context." The AI is not just looking for the best business; it is looking for the business that best fits the sentence it is currently constructing. This is a significant departure from how the Local Pack functions, where the business is an entity in a list rather than a component of a paragraph. The frequency of "ghosting" — where the AI provides a general answer but fails to cite any specific local business — was also notably high in the 1,814 search instances, occurring even for queries with high local intent.

What this means for local businesses

Operators and agencies must adjust their expectations regarding AI-driven lead generation. The data suggests that we are currently in a transition period where AI visibility cannot be treated as a reliable KPI. We recommend the following actions for managing this uncertainty:

1. Shift to aggregate reporting: Instead of tracking specific "wins," focus on the percentage of time your brand appears across a 30-day window to smooth out daily volatility. A 40% citation rate is a more honest metric than a single-day #1 ranking.
2. Prioritize diverse citations: Because the AI draws from various sources, ensuring your brand information is consistent across third-party directories remains the best defense against being omitted. The AI often cross-references data to verify a business's existence before citing it.
3. Monitor specific AI-triggering queries: Identify which long-tail questions (e.g., "how much does a new furnace cost in Chicago?") trigger AI Overviews and ensure your site has direct, authoritative answers to those questions. This helps the LLM associate your brand with the specific topic.
4. Maintain traditional SEO foundations: Since AI Overviews are still erratic, the stable traffic from the traditional Local Pack should remain the primary focus of your optimization efforts. Use Google Business Profile features like posts and photos to maintain the prominence signals that the AI still uses as a secondary verification.

Is it time for a new reporting standard?

The sheer volume of fluctuations noted in this study suggests that the industry may need to move toward a "share of voice" metric for AI Overviews rather than a simple ranking number. If a brand is cited in 40% of AI Overviews for a specific keyword set, that is a more honest reflection of performance than claiming a top position that only exists for a fraction of users.

We believe that until Google stabilizes the retrieval mechanisms for local data in AI summaries, agencies should be cautious about promising consistent visibility in these modules. The erratic nature of the 1,814 tracked answers serves as a reminder that being "AI-optimized" is currently a pursuit of a moving target. The goal should be to provide the most comprehensive and verifiable data about a business so that when the AI does its real-time synthesis, your brand is the most logical choice to include.

Sources

• I tracked 1,814 AI answers for one local business - Reddit
• Google Search Central: AI-generated content guidance
• Google Business Profile Help: How to improve local ranking