22.4% of AI Overview Traffic Books as 'Direct' in GA4. Your Dealership's Organic Report Is Undercounting the Channel Everyone Is Arguing About.

Yesterday Search Engine Land published nine months of first-party AI Overview tracking, and the most useful number in it is a measurement error.

From September 2025 through June 2026, Alex Galinos recorded 51,200 tracked AI Overview events across 1,661 cited snippets for a brand he manages. Across that dataset, 22.4% of the AI Overview traffic was landing in Google Analytics as Direct rather than Organic Search. In raw counts, that is 11,468 AI Overview events that originated on a Google search results page and got credited to nobody. The monthly range ran from 16.8% in April 2026 to 29.3% in May.

If you judge your store's SEO by the Organic Search line in GA4, that is the line under discussion.

Then he ran the second number. Over the full nine months, 7.53% of the brand's total organic sessions came from AI Overviews. At the February-March 2026 peak it hit 16-17%, close to one in six organic visitors. By the time he wrote it up, it had fallen back to around 2-4%.

Two findings that pull in opposite directions. The first says you have more organic traffic than your dashboard is showing you. The second says the size of the AI channel can swing by a factor of five inside a single year, so any number you build a budget on has a short shelf life.

The method costs nothing and your GA4 property is already collecting it

Here is the part worth an afternoon of somebody's time at your store.

When a shopper clicks a cited link inside a Google AI Overview, Google often appends a text fragment to the destination URL. It looks like #:~:text= followed by the exact sentence Google quoted, and its job is to tell the browser to scroll down and highlight that passage. Galinos built a custom dimension in GA4 that fires whenever a session lands with that fragment present, then grouped the results by page.

His summary of why he bothered: "It's not a perfect signal, but it's the most reliable one available without waiting for Google to expose this natively in Search Console. The fragment is already sitting in your GA4 data. You just need to surface it."

That is the whole trick. No vendor, no subscription, no dashboard with a logo on it.

Google is not going to hand you this number

The obvious question is why anyone has to do this in 2026, when Google has a reporting product sitting right there.

Google did ship one. On June 3, it announced Search Generative AI performance reports in Search Console, giving site owners a dedicated view of their visibility inside AI Overviews and AI Mode. Read what it actually contains. Google's own help documentation says the report covers impressions, defined as "how many times links to your site were shown to a user in a generative AI feature on Google Search," broken out by page, country, device and date.

No clicks. No click-through rate. No queries. No position. Google's blog post says it is "rolling these reports out to a subset of websites" and that it is working on "adding additional metrics over time."

So the report answers how often Google showed your dealership inside an AI answer. It does not answer whether a single human being clicked through, or what they did when they got there.

And since August 13, even the impressions have been wrong. Google's own data anomalies log carries this entry: "A logging error caused a decrease in impressions on the Generative AI performance report in Search for data starting on August 13, 2026. This issue affects data logging only and is ongoing." Barry Schwartz reported it on August 17, and it was still unresolved. If your agency sends you an August report showing your AI visibility falling off a cliff in the middle of the month, that dip is a logging bug in Google's pipeline.

We have run this play before

Anyone who was doing digital in 2011 recognizes the shape of this.

On October 18 that year, Google made SSL search the default for signed-in users. Overnight, a website stopped receiving the actual search query in the referrer and simply learned that a visitor came from google.com. Google's answer at the time was that aggregate query, impression, click and CTR data would remain available in Webmaster Tools. The industry called the hole "(not provided)" and spent the next decade building workarounds to fill it.

The pattern repeats. Google removes granular visibility, offers an aggregate report on its own terms and its own timeline, and the operators who get ahead are the ones who stop waiting and instrument their own site. The dealers who learned to read Search Console properly back then got years of usable insight out of it while their competitors kept demanding keyword-level reports that were never coming back.

Same decision, new channel.

What this means for your dealership

Your organic numbers are probably understated, and so is your agency's work. If roughly a fifth of AI Overview arrivals are booking as Direct, then every "organic sessions are flat" conversation you have had this year was held with a number that has a systematic hole in it. This cuts both ways for a dealer. It means an SEO program may be performing better than the report shows, and it also means the vendor you were about to fire deserves a look at the real figure first. We have written before about why your own traffic report beats Google's public claims, and this is the sharpest version of that argument yet.

Your Direct line has been flattering you. Dealers love a big Direct number because it reads as brand strength. People typed us in. Some meaningful slice of that is Google AI traffic wearing a costume, and it inflates exactly the metric general managers use to argue their store does not need to invest in search.

The content getting cited is content you already have and probably buried. Galinos found that "structured transport comparison tables, formatted as actual HTML tables, are punching well above their weight in terms of citation frequency," and that citations favor "content that is specific, structured, and directly answers a well-defined question. Times, prices, named routes, comparison formats. Not vague editorial." Translate that to a franchised store: trim-level comparison tables, service pricing, lease terms, hours, warranty coverage windows, what a specific package includes. Most dealership sites have all of it, rendered inside a JavaScript widget or a PDF where nothing can cite it. That matches what we found when we ran 1,100 car shopping searches through AI Overview.

Citation is not permanent. Snippets in his dataset had lifecycles. Some peaked and faded as content went stale relative to what Google now prefers to cite; others emerged months after publication and kept climbing. A page that earned citations in March is not guaranteed to hold them in September.

Do not rebase your forecast on a peak. Going from 16-17% of organic sessions down to 2-4% inside a few months is the single most important line in the piece for anyone writing a 2027 marketing plan.

The honest caveats, because they are large

This is one brand, in one industry, in one measurement setup. Galinos manages transportation brands, not car dealerships, and the query mix behind his data is transfer times and route pricing. Take his method. Do not take his percentages as yours.

The signal itself is dirty in both directions. Galinos says so plainly: the #:~:text= fragment is also used by featured snippets and People Also Ask, so some of what he counted came from those formats rather than AI Overviews. Running the other way, not every AI Overview link carries the fragment at all, so the method also misses real AI Overview clicks. He also concedes his GA4 dimension is event-scoped while sessions are session-scoped, which makes the share percentage directional rather than exact.

So this is an instrument, not a meter. It will tell you whether the number at your store is closer to 2% or 20%, and which of your pages are pulling citations. It will not survive being quoted to three decimal places in a board deck. Anyone who sells you a precise AI visibility figure is selling you a reading that cannot be that precise.

What to actually do about it

1. Turn the tracking on. Have whoever owns your GA4 create a custom dimension that captures the page location fragment, then build an exploration filtered to sessions containing #:~:text=. This is an afternoon of work, once.
2. Measure your own Direct leak before you argue about it. Compare the AI Overview events you capture against how those same sessions were channeled. Your store's misattribution rate is your number, and nobody else's applies.
3. Check whether your property has the Search Console report yet. It is rolling out to a subset of sites. If you have it, use it for impressions and page-level visibility only, and ignore mid-August movement until Google clears the logging bug.
4. Find the pages already earning citations, then feed them. Concentration in his data was extreme. The single best snippet drove 2,276 events against an average of 31 across all 1,661. A small number of pages will be doing almost all of the work at your store too, and those are the ones to refresh first.
5. Get your specifics into real HTML. Service pricing, trim comparisons, warranty terms, hours, financing eligibility. Real tables and real text on the page, crawlable without executing a widget.
6. Re-check quarterly. The volatility finding means an annual audit is useless here.

Most dealers have spent two years arguing about whether AI search is taking their traffic. The more useful question is whether their reporting can even see it, and right now for most stores the honest answer is no. Google's own report cannot tell you, and as of this morning Google still has the logging error flagged as ongoing.

Your analytics can, if somebody turns it on.

If you want us to set the AI Overview tracking up on your store's GA4 and show you what your Direct line is actually hiding, book a walkthrough and we will go through it together.