9 months of AI Overview data: 51,000+ tracked events analyzed

For most brands, the true volume of traffic coming from AI Overviews remains a mystery. Google has yet to provide a direct signal for this traffic within Search Console, leaving marketers guessing about how many organic sessions originate from AI-generated summaries, which content pieces are fueling those visits, and whether analytics platforms are even counting them correctly.

To answer those questions, we built our own tracking system.

Since September 2025, I have been collecting AI Overview referral data for a brand I manage in the transportation sector. Over the span from September 2025 to June 2026, we logged 51,200 tracked events across 1,661 cited snippets.

What the data reveals is that AI Overview visibility in search results is anything but consistent. It also highlights clear trends in the types of content Google chooses to cite, how those citations evolve over time, and the significant portion of traffic that gets mislabeled in standard analytics.

Here is what nine months of tracking uncovered, along with the practical implications for SEO reporting, content strategy, and generative engine optimization.

The tracking method, and why it works

The approach is simple, yet few teams have put it into practice.

When a user clicks on a cited snippet inside a Google AI Overview, Google sometimes appends a #:~:text= fragment to the destination URL. We set up a custom dimension in GA4 that activates whenever a session lands with that fragment present.

It is not a flawless signal, but it is the most dependable option available until Google decides to expose this data natively in Search Console. The fragment is already sitting in your GA4 data. You just need to surface it.

Once captured, we grouped snippets into thematic categories and began examining patterns. Several findings stood out immediately.

Concentration is high. The single top-performing snippet drove 2,276 events on its own. The average across all 1,661 snippets was just 31. As with most aspects of SEO, a small number of pages account for the majority of the impact.

Snippets have lifecycles. Some peak during a specific window and then decline, typically when query intent shifts seasonally or when content becomes outdated relative to what Google now prefers to reference. Others emerge months after publication and continue to gain traction. This makes freshness and specificity more critical than you might initially assume.

Content prioritization sharpens considerably. The most valuable outcome of having this data is that it grounds decisions in evidence. When you map snippet volume and growth trajectory against your current content investment, the gaps become obvious.

Certain topic types got cited

In our case, transfer time content and pricing content are already receiving frequent citations with clear upward momentum. The obvious move there is to refresh and expand those existing pages. Destination guides, on the other hand, are underperforming relative to what the data suggests they could be driving. Structured transport comparison tables, formatted as actual HTML tables, are punching well above their weight in citation frequency. That tells you something about how AI Overviews select content to surface.

The broader pattern aligns with what most people working in GEO are discovering: AI Overview citations favor content that is specific, structured, and directly answers a well-defined question. Times, prices, named routes, comparison formats. Not vague editorial.

22.4% of AI Overview traffic is being misattributed to Direct

This was the most striking finding.

When we began plotting the dataset in a GA4 exploration report, we noticed that a meaningful portion of traffic arriving via AI Overviews was being attributed to the Direct channel rather than Organic Search.

AI Overviews only exist within Google Search. Traffic from there should be counted as organic. So we built misattribution tracking into our AI Overview dashboard and started measuring it on a monthly and weekly basis.

Across the full dataset, more than 50,000 AI Overview events over nine months, the average misattribution rate was 22.4%.

That amounts to 11,468 AI Overview events attributed to Direct instead of Organic. The range varied by month. The worst month was May 2026, when 29.3% of AI Overview sessions were misattributed to Direct. The best month was April 2026, with 16.8% misattributed.

The implications for SEO reporting are substantial. Depending on your total organic traffic volume and how much AI Overview referral traffic you generate, you could be materially underreporting organic performance without realizing it.

AI Overviews are driving 7.53% of organic sessions, but it is volatile

Using the same first-party dataset, we built a percentage metric to track what share of our total organic sessions came from AI Overviews each period.

From September 2025 to June 2026, 7.53% of organic sessions came from AI Overviews.

But the distribution is not flat. At its peak in February through March 2026, it reached 16-17%, meaning nearly one in six organic visitors arrived via an AI Overview. More recently, as of writing, it has trended down to around 2-4%.

That volatility is notable. It suggests AI Overview prominence in the search results is not stable. It fluctuates based on query type, Google’s confidence in available content, and possibly broader algorithmic changes. Brands treating AI Overview traffic as a fixed percentage of organic are likely miscalibrating their models.

The caveats you should know

Two important limitations apply to this approach.

First, the #:~:text= identifier is not exclusive to AI Overviews. The same fragment is used by Featured Snippets and People Also Ask results. So some portion of what we capture may bleed from those formats. We did run the exercise in Ahrefs to check our Featured Snippet exposure, and at the time of checking, we were included in fewer than 30, largely because we do not target terms that typically trigger Featured Snippets. In practice, we believe AI Overviews are the dominant driver given the volume and the period of data collection, but it is not a perfectly clean signal.

Second, the metric compares events to sessions. Our custom dimension in GA4 is event-scoped, not session-scoped. The AI Overview share percentage therefore compares events against sessions, which is not an ideal comparison. It is better than nothing, and it is directionally accurate, but worth knowing if you replicate this.

What this means practically

If you are not tracking this yet, start. The setup is not complex, and the signal, imperfect as it is, is the best first-party data you will get until Google decides to surface this natively.

A few takeaways from nine months of data. Structured, specific content wins. Times, prices, routes, named comparisons, not editorial padding. Freshness matters more than you think. Snippets have lifecycles, and stale content loses citation share. Your organic traffic is likely underreported. If roughly 22% of AI Overview events are going to Direct, your channel reporting has a systematic gap. The share is volatile. Do not treat a peak month as a new baseline.

The #:~:text= fragment is sitting there in your GA4 data right now. You just need to surface it.

Special thanks to Simant Sah for helping collate the data and put together the report where the screenshots come from.