The gap between Search Console and GA4
I compared 28 days of data from one site across two tools, and a clear discrepancy emerged.
- Google Search Console clicks: 96
- GA4 sessions (excluding self-visits and not set): 277
That’s a gap of 181 sessions. Roughly 65% of all traffic couldn’t be explained by Google Search.
Even adding Yahoo organic, search accounted for just 101 sessions. The remaining 176 came from somewhere else.
It’s not social either
Breaking down the source/medium tells a more specific story.
| Source | Sessions | Share |
|---|---|---|
| direct / (none) | 117 | 42% |
| Google + Yahoo search | 101 | 36% |
| X (t.co) | 19 | 7% |
| 10 | 4% | |
| AI referrals (ChatGPT, NotebookLM, etc.) | 3 | 1% |
| Other | 27 | 10% |
Search makes up just over a third. Social accounts for about 11%. And the single largest channel is direct at 42%.
The 19 sessions from X were largely the result of active posting — what came back was roughly what was put out. And both X and Facebook now suppress impressions on posts containing external links, making social a structurally unreliable acquisition channel.
The problem with direct
So what’s actually inside that 42%?
Direct traffic is a catch-all. It includes bookmarks, typed URLs, clicks from email or messaging apps where the referrer is stripped — and increasingly, visits from AI search services that don’t pass a referrer.
The complication is that each AI platform handles referrers differently.
Perplexity and ChatGPT (desktop) do pass referrer data, so they appear as identifiable referral sources in GA4. ChatGPT’s mobile app and some AI agents don’t — those sessions land in GA4 as direct.
Google AI Mode introduces yet another problem. Clicks from AI Mode are counted in Search Console and recorded as google / organic in GA4. They don’t fall into direct, but they’re indistinguishable from regular search clicks. A user who clicked through from an AI-generated answer looks exactly the same as one who clicked a traditional blue link.
In this dataset, several pages had GA4 sessions but zero clicks in Search Console — meaning no traffic from Google Search at all.
One case stood out: an English-language article about AI search behavior. Zero GSC clicks. Zero social referrals. Yet it received a direct session with an engagement time of 137 seconds — over two minutes of reading.
It’s hard to imagine someone arriving at a specialized English article via bookmark or typed URL. More likely, the URL appeared in an AI-generated response and was either copied into a browser or clicked through a service that strips referrer data. But in GA4, it’s simply labeled “direct.”
Beyond session-level tracking: reading the wave
This is where the approach needs to shift.
To be fair, the trend is moving in the right direction. AI services are increasingly passing referrer data. Perplexity and ChatGPT desktop clicks now show up as referrals in GA4. That’s a welcome development.
But it’s not enough to see the full picture. ChatGPT mobile traffic still falls into direct. Users who copy-paste URLs from AI answers into their browsers generate direct sessions. Google AI Mode clicks blend into google / organic with no way to separate them. Counting only what shows up as AI referral traffic captures the tip of the iceberg.
So how do you measure what referrers can’t show?
One approach: AI crawler logs.
Before an AI search service can cite your content in a response, it needs to crawl the page. These crawls don’t appear in GA4, but they do show up in server or CDN logs.
If the frequency of AI crawler visits to a page increases around the same time that page’s direct sessions rise, that correlation could be a signal — an indication that AI citations are driving site visits.
This isn’t session-level attribution. It’s reading the overlap of time-series waves. The same logic behind Marketing Mix Modeling — inferring effect from the correlation between ad spend waves and revenue waves — applied to AI crawl activity and direct traffic patterns.
No conclusions yet
To be honest, this is still a hypothesis.
Whether the 42% direct traffic contains meaningful AI-driven visits can’t be confirmed with current data. Whether AI crawler frequency correlates with direct session trends requires sustained observation. This kind of time-series correlation analysis only becomes meaningful with at least a year of accumulated data.
But one thing is clear: 42% of this site’s traffic sits in a channel where the contents are almost entirely invisible. And there’s good reason to believe AI-driven visits are part of what’s hiding there.
Making that visible requires going beyond GA4 — capturing AI crawler activity from CDN or server logs, accumulating it over time, and reading the patterns as waves.
You can’t optimize what you can’t measure. The first step is learning to see the wave.
The figures in this article are based on 28 days of data from a single site. Traffic composition varies significantly depending on site size, industry, and acquisition structure.