Most brands still don’t person a clear image of really overmuch postulation AI Overviews are sending them. Google hasn’t fixed america a cleanable awesome for AI Overview postulation successful Search Console, making it difficult to cognize really overmuch integrated postulation is coming from AI Overviews, which contented is driving it, and really accurately that postulation is being reported.
So we built our ain tracking.
Since September 2025, I’ve been capturing AI Overview referral information for 1 of the brands I negociate successful the proscription industry. From September 2025 to June 2026, we recorded 51,200 tracked events crossed 1,661 cited snippets.
The information shows that AI Overview prominence successful the SERP isn’t stable. It besides reveals clear patterns successful the types of contented Google cites, really those citations alteration complete time, and really overmuch of that postulation gets misattributed successful analytics.
Here’s what 9 months of search revealed — and what it intends for SEO reporting, contented prioritization, and GEO.
The search setup — and why it works
The methodology is straightforward, but astir teams haven’t implemented it.
When a personification clicks connected a cited snippet wrong a Google AI Overview, Google sometimes appends a #:~:text= part to the destination URL. We created a civilization magnitude successful GA4 that fires whenever a convention lands pinch that part present.
It’s not a cleanable signal, but it’s the astir reliable 1 disposable without waiting for Google to expose this natively successful Search Console. The part is already sitting successful your GA4 data. You conscionable request to aboveground it.
Once we had it captured, we grouped snippets into thematic categories and started analyzing patterns. A fewer things stood retired immediately.
- Concentration is high: The top-performing snippet unsocial drove 2,276 events. The mean crossed each 1,661 snippets is 31. Like astir things successful SEO, a mini number of pages are doing the bulk of the work.
- Snippets person lifecycles: Some highest during a circumstantial play and past slice — usually erstwhile query intent shifts seasonally aliases erstwhile the contented becomes old comparative to what Google is now preferring to cite. Others look months aft publication and support climbing. This makes freshness and specificity much important than you’d initially assume.
- Content prioritization gets sharper: The astir useful output of having this information is that it makes decisions much grounded. When you representation snippet measurement and maturation trajectory against your existent contented investment, the gaps go visible.
Track, grow, and measurement your visibility crossed Google, AI search, social, local, and each transmission that influences buying decisions.
Certain taxable types sewage cited
In our case:
- Transfer clip content and pricing content are already being cited often pinch clear upward momentum. The play location is to refresh and grow the existing content.
- Destination guides are underperforming comparative to what the information suggests they could beryllium driving.
- Structured carrier comparison tables, formatted arsenic existent HTML tables, are punching good supra their weight successful position of citation wave — which tells you thing astir really AI Overviews are selecting contented to surface.
The broader shape is accordant pinch what astir group moving successful GEO are finding: AI Overview citations favour contented that is specific, structured, and straight answers a well-defined question. Times, prices, named routes, comparison formats. Not vague editorial.
22.4% of AI Overview postulation is being misattributed to Direct
This was the astir striking finding.
When we started plotting the dataset connected a GA4 exploration report, we noticed that a meaningful information of postulation arriving via AI Overviews was being attributed to the Direct transmission alternatively than Organic Search.
We cognize AI Overviews only beryllium wrong Google Search. Traffic from location should beryllium attributed to organic. So we built misattribution search into our AI Overview dashboard and started measuring it connected a monthly and play basis.
Across the afloat dataset — much than 50,000 AI Overview events complete 9 months — the mean misattribution complaint was 22.4%.
That’s 11,468 AI Overview events attributed to Direct alternatively of Organic. The scope varied by month:
- Worst month: May 2026 – 29.3% of AI Overview sessions were misattributed to Direct.
- Best month: April 2026 – 16.8% were misattributed.
The implications for SEO reporting are significant. Depending connected your full integrated postulation measurement and really overmuch AI Overview referral postulation you’re generating, you could beryllium materially underreporting integrated capacity without realizing it.
AI Overviews are driving 7.53% of integrated sessions — but it’s volatile
Using the aforesaid first-party dataset, we built a percent metric to way what stock of our full integrated sessions came from AI Overviews each period.
From September 2025 to June 2026, 7.53% of integrated sessions came from AI Overviews.
But the distribution is not flat:
- At its highest successful February-March 2026, it reached 16-17% — meaning astir 1 successful six integrated visitors arrived via an AI Overview.
- More precocious (as of writing), it has trended down to astir 2-4%.
That volatility is notable. It suggests AI Overview prominence successful the SERP is not unchangeable — it fluctuates based connected query type, Google’s assurance successful disposable content, and perchance broader algorithmic changes. Brands treating AI Overview postulation arsenic a fixed percent of integrated are apt miscalibrating their models.
The caveats you should know
Two important limitations pinch this approach:
The #:~:text= identifier isn’t exclusive to AI Overviews
The aforesaid part is utilized by Featured Snippets and People Also Ask results. So immoderate information of what we’re capturing whitethorn bleed from those formats.
We did tally the workout successful Ahrefs to cheque our Featured Snippet exposure, and astatine the clip of checking, we were included successful less than 30 — mostly because we don’t target position that typically trigger Featured Snippets.
In practice, we judge AI Overviews are the ascendant driver fixed the measurement and the play of information collection, but it’s not a perfectly cleanable signal.
The metric compares events to sessions
Our civilization magnitude successful GA4 is event-scoped, not session-scoped. The AI Overview stock percent truthful compares events against sessions, which isn’t an perfect comparison.
It’s amended than thing — and it’s directionally meticulous — but worthy knowing if you replicate this.
Track your visibility crossed AI search, uncover missed opportunities, and turn your beingness wherever customers are asking questions.
What this intends practically
If you’re not search this yet, start. The setup isn’t complex, and the awesome — imperfect arsenic it is — is the champion first-party information you’ll get until Google decides to aboveground this natively.
A fewer things to return distant from 9 months of data:
- Structured, circumstantial contented wins. Times, prices, routes, named comparisons. Not editorial padding.
- Freshness matters much than you think. Snippets person lifecycles. Stale contented loses citation share.
- Your integrated postulation is apt underreported. If ~22% of AI Overview events are going to Direct, your transmission reporting has a systematic gap.
- The stock is volatile. Don’t dainty a highest period arsenic a caller baseline.
The #:~:text= part is sitting location successful your GA4 information correct now. You conscionable request to aboveground it.
Special acknowledgment to Simant Sah for helping collate the information and put together the study wherever the screenshots travel from.
Contributing authors are invited to create contented for MCP and are chosen for their expertise and publication to the hunt community. Our contributors activity nether the oversight of the editorial staff and contributions are checked for value and relevance to our readers. MCP is owned by Semrush. Contributor was not asked to make immoderate nonstop aliases indirect mentions of Semrush. The opinions they definitive are their own.
About the Author
