Some of Your Direct Traffic Is Organic. The Problem Is You Can't Say How Much.

If Direct traffic has been climbing in your reports this year, you have probably already built a story around it. Brand awareness is up. People are typing us in. The podcast is working.

Some of that story is probably true. Some of it is organic traffic wearing a disguise, and a study published this month puts a shape on how it happens.

How much AI Overview traffic gets logged as Direct?

Search Engine Land tracked traffic arriving from Google AI Overview citations for a single brand over nine months, September 2025 through June 2026. The method was clever and simple: when someone clicks a cited source inside an AI Overview, Google often appends a text fragment to the URL. The team built a GA4 dimension that fires on that fragment. Nine months later they had 51,200 tracked events across 1,661 cited snippets.

Three findings matter here.

First, AI Overviews drove 7.53 percent of the brand's organic sessions across the period. Not trivial, not dominant.

Second, 22.4 percent of that AI Overview traffic was logged as Direct instead of organic search. The mechanism is mundane. Someone clicks a citation, the referrer does not always survive the trip, the session still has to be counted somewhere, so it lands in Direct. Traffic that organic content earned gets credited to nothing.

Third, and this is the one I want you to sit with: the misattributed share was not stable. It ranged from 16.8 percent in the best month to 29.3 percent in the worst. That is a 13-point swing.

Before you go looking for 22.4 percent in your own portal: don't. This is one brand in one industry. The study documents a mechanism, not a benchmark. The authors are also upfront that the text fragment signal is not exclusive to AI Overviews, since featured snippets and People Also Ask results use it too. Your number is different, and you cannot currently know what it is. That last part is the point.

A steady bias is fixable. A moving one is not.

A steady bias is a solvable problem. If Direct always absorbed a fixed slice of organic, you could footnote it and move on. A swinging bias is different. It means your month-over-month organic trend line contains noise you cannot separate from signal, and the trend line is what clients budget against. "Organic is flat, Direct is up 20 percent" might describe a strategy shift. It might describe nothing.

Why is the HubSpot exposure worse than GA4?

HubSpot users have a second exposure. Original Source is a first-touch property. When a contact's first visit arrives from an AI Overview citation with no referrer, that contact can get stamped Direct Traffic and carry it forever, through every lifecycle report and every attribution model you run afterward. The session-level miscount washes out over time. The contact-level one compounds.

One more scope note, because my audience will catch it if I skip it. This is specifically an AI Overview problem. Clicks from ChatGPT or Perplexity usually pass a referrer and show up as referral traffic, where you can see them. And this is not "AEO traffic" or "GEO traffic." Those terms name the practice of getting cited, not a traffic source.

How should you report Direct traffic that's really AI Overview clicks?

Not build a new attribution model. The instinct to add machinery is strong, and it is wrong here. No model fixes a missing referrer.

The fix is cheaper and more uncomfortable: be honest about the confidence interval. When you hand a client an organic trend, say plainly that some unknown, moving share of Direct is organic in disguise. Before anyone reallocates spend on the assumption that Direct means "people who already knew us," check what those Direct sessions are landing on. New contacts entering as Direct on a deep blog URL did not type it from memory.

The teams that get burned by this will not be the ones with the biggest misattribution rate. They will be the ones who presented a trend line as more certain than it was, and got asked about it later.