Your AI Traffic Is Invisible in Google Analytics: Here's How to Find It

On September 15, 2026, AI-visibility tracker Peec AI announced a new feature that connects its citation data to real Google Analytics sessions. Its reasoning fits in one sentence: Google's own AI answers get counted as regular search, visits from mobile AI apps and referrer-less links land as "Direct" traffic, and Google Analytics alone doesn't see all of it.

This isn't a niche problem reserved for large data teams. If your site gets cited by ChatGPT or Perplexity, a real chunk of that traffic is probably already sitting in your Analytics reports, under a label that says nothing about where it actually came from. Here's what the most solid studies show about the size of the gap, what Google Analytics' latest update does and doesn't fix, and how to find that hidden traffic without buying another tool.

Why doesn't Google Analytics see all the traffic AI engines send?

Google Analytics identifies a visit using a technical signal the browser passes along: the "referrer," meaning the page or app the click came from. The problem is that generative AI engines don't always pass that signal — especially on mobile. Without it, Analytics has no way to tell a visit came from ChatGPT rather than someone typing your address straight into their browser.

A detailed breakdown of the phenomenon, published April 30, 2026 by The GEO Community, names three distinct technical causes (thegeocommunity.com, "Why GA4 underreports AI search traffic"):

• The drop from secure (HTTPS) to non-secure (HTTP). A web security rule blocks the referrer from being passed whenever a link leads to a page that isn't served over HTTPS. Any site or redirect chain that touches HTTP loses this signal.
• Mobile apps' built-in browsers. When ChatGPT on iPhone, Perplexity on Android, Claude, or Copilot open a link, they often use a mini browser component baked into the app rather than the phone's actual browser. That component frequently fails to pass the referrer along.
• A deliberate choice by some AI engines. Some send an explicit instruction, called Referrer-Policy: no-referrer, that strips this signal outright, regardless of how the receiving site is configured.

The measured impact, per The GEO Community's comparison of Analytics reports against citation-tracking tools: Google Analytics shows 30–50% fewer AI-attributed sessions than citation volume would predict. Claude and Gemini appear worst-affected (roughly 25–50% of visits correctly identified), desktop ChatGPT least affected (80–90%).

How much AI traffic is really hiding inside your "Direct" bucket?

A Loamly study of 446,405 visits found that 70.6% of traffic confirmed as AI-referred arrives with no referrer at all, and gets classified as "Direct." In that sample, only 6,015 visits were directly identifiable as AI-referred; another 14,413 — invisible at first glance — were traced back to an AI source through a more advanced detection method than a plain referrer check.

The study, published in November 2025 and updated in February 2026, lays out its method: cross-referencing technical fingerprints, browsing-behavior analysis, and pattern recognition specific to AI bots and apps, rather than relying on referrer data alone (loamly.ai, "The AI Traffic Attribution Crisis").

In practice: if your Analytics report shows 100 AI-attributed visits this month, it's reasonable to assume roughly 240 more AI-referred visits already exist in your "Direct" bucket, invisible unless you dig further.

Does this hidden traffic actually convert better than the rest?

Yes, and by a wide margin — several independent studies, on different samples, measure AI-referred traffic converting well above regular traffic. Ahrefs published a figure in June 2025 that has since become a widely cited benchmark: over a 30-day window, AI-referred traffic made up 0.5% of their total visits but drove 12.1% of signups (ahrefs.com/blog/ai-search-traffic-conversions-ahrefs) — a conversion rate 23 times higher than classic organic search.

This traffic also behaves differently, per the same Ahrefs data: AI-referred visitors view roughly 50% more pages per session than visitors from classic organic search, and bounce less often. They spend less time on each individual page, though — consistent with a visitor who arrives already partly informed by the AI's answer, and comes to verify or dig into one specific point rather than discover the topic from scratch.

The Loamly study, which measures the AI traffic hidden inside "Direct" specifically, found it converting at 10.21%, versus 2.46% for the rest of direct traffic — 4.1 times higher.

Ahrefs does flag an important caveat, though: by its own estimate, visitors coming from AI search click roughly 75% fewer links than those coming from classic search. Some of today's conversion gap comes from a traffic source that's still small and highly qualified; if AI traffic becomes a much larger share of the total, that ratio could move closer to average. That's not a reason to ignore the signal — it's a reason not to extrapolate a fixed ROI number from it too fast.

How do you report this number without overselling it to a client or your team?

By framing it as an order-of-magnitude range, never as an exact figure: say "probably two to three times more AI traffic than Analytics shows," never "exactly X extra visits." The studies cited in this article come from different samples (446,405 visits at Loamly, one month of Ahrefs' own traffic) using different detection methods. They agree on the existence and scale of the problem, not on a universal multiplier you can apply to any site as-is.

For an agency or marketing team that has to report on this internally, three habits keep the number honest:

• Always separate measured traffic from estimated traffic. An honest report distinguishes "X confirmed AI-referred visits" (AI Assistant channel plus identifiable referrer) from "likely Y additional unidentified visits, based on published industry studies."
• Attach every estimate to its source, with a date. A number like "70.6%" cited with no source goes stale fast and loses credibility the moment someone checks it. Cited with its source and publication date, it stays verifiable and open to debate.
• Don't turn a correlation into proof of causation. A "Direct" traffic bump on a page that was just cited by an AI engine is a signal worth watching, not formal proof of attribution. It's the same principle behind our Track, Fix, Prove method: a promising signal gets checked before it gets celebrated.

What GA4's new "AI Assistant" channel changes — and what it doesn't

Since May 13, 2026, Google Analytics automatically routes part of your AI traffic into a new dedicated channel called "AI Assistant," with zero setup required. The official release note states that visits from ChatGPT, Gemini, and Claude now get a specific medium value (ai-assistant), grouped under this new channel in the acquisition report (support.google.com, May 13, 2026 release note).

That's a real improvement: before this date, this traffic blended into the general "Referral" bucket with no label of its own. But the update has three documented limits, per Semrush's May 20, 2026 breakdown (semrush.com/blog/ga4-adds-ai-assistant-channel):

1. Perplexity wasn't included at launch. Its traffic still lands in the generic "Referral" bucket unless you build a custom channel for it.
2. Google's own AI answers (AI Overviews, AI Mode) still count as "Organic Search." They don't join the "AI Assistant" channel, which targets external conversational assistants, not Google's in-house answer boxes.
3. Referrer-less traffic still lands in "Direct." The new channel only catches visits where the referrer signal is present. It doesn't touch the 70.6% hidden-traffic problem Loamly measured.

In short: the new channel shrinks the share of misclassified traffic — it doesn't make the invisible traffic disappear. That's exactly the gap Peec AI points to in the announcement referenced in the intro, right before pitching its own complementary detection method.

How do you find that hidden traffic without buying another tool?

Four free checks, all doable inside tools you already have, are enough to estimate how much AI traffic your site is really getting. None of them replace a dedicated tool once your AI traffic volume gets large, but all four give you a reliable first signal.

1. Treat the "AI Assistant" channel as a floor, not a total. Check it monthly in your acquisition report, but remember it only captures the fraction of real AI traffic that kept its referrer.

2. Compare "Direct" traffic on your GEO target pages against your AI citation trend on those same pages. If a specific page, optimized to get cited by AI engines, sees its "Direct" traffic climb with no campaign or press mention to explain it, over the same period your citation tracking shows rising visibility on questions tied to that page, the overlap is probably not a coincidence.

3. Cross-reference Google Search Console's AI reports against unexplained direct-traffic spikes. Search Console shows how often you show up in Google's AI answers, without breaking out clicks; Analytics shows the clicks, without always saying they came from an AI. Side by side, on the same pages and the same weeks, these two partial views fill in each other's gaps.

4. Test how plausible your "Direct" sessions actually are, based on their landing page. A genuine direct visit usually lands on your homepage or some page the visitor bookmarked or remembers. A "Direct" session that lands straight on a deep, specific, hard-to-type page — a blog post, one particular product page — looks a lot more like a click from an AI answer than a real direct visit. It's not proof on its own, but applied across all your "Direct" landing pages, this one filter often reveals a clear pattern.

This whole approach follows the same logic as Track, Fix, Prove: a rising AI citation curve proves nothing until it shows up as measured traffic, even approximately. And a "Direct" traffic bump with no obvious explanation deserves a check, not blind praise or blind dismissal.

An invisible number is still a number that counts

AI traffic isn't the top source of visits for most sites yet. But the studies agree on one point: what Google Analytics shows you is a systematic undercount, not an accurate reading. Ignoring that gap means under-investing in the channel that converts best, as both Ahrefs and Loamly show, each with its own method and its own sample.

The good news is that closing the gap doesn't require a bigger budget. GA4's new "AI Assistant" channel, live since May 2026 with zero setup, already covers part of the problem. The rest can be found with simple cross-checks between tools you already run, until a dedicated tool becomes worth the cost.

Before chasing where this traffic goes, it's worth confirming your pages actually get cited by the AI engines that could be sending it in the first place. Run a free AI visibility audit to see which questions your site already shows up for — and which ones are still up for grabs before attribution even enters the picture. If you're then deciding where to focus across engines, our 2026 AI traffic breakdown by engine picks up right where this one leaves off.