For every AI visit GA4 shows you, 2.4 more are hiding in your "Direct" bucket. I proved this across 446,405 visits: 6,015 AI visits with referrer headers vs 14,413 dark AI visits with no referrer. GA4 catches the first group. It dumps the second into "Direct."
GA4 cannot fix this. It depends on HTTP referrer headers, and ChatGPT mobile, Claude, and most AI surfaces strip them. But you can estimate how much dark AI traffic hides in your own data using four methods that work inside GA4 today.
Want to see your actual dark AI numbers instead of estimating? Loamly detects dark AI automatically.
The Detection Gap
If you already set up a custom AI Traffic channel in GA4, you're catching AI visits that carry referrer headers. That's real progress. But it only covers part of the picture.
| Traffic Type | Visits (Loamly dataset) | What GA4 Shows |
|---|---|---|
| AI with referrer (visible) | 6,015 | "AI Traffic" channel (if configured) |
| Dark AI (no referrer) | 14,413 | "Direct" |
| Non-AI | 425,977 | Correctly classified |
Source: 446,405 visits across Loamly customer websites
The 14,413 dark AI visits convert at 10.21% transactional rate, vs 2.46% for non-AI traffic. That is 4.1x higher conversion hiding in a bucket GA4 treats as undifferentiated.
The methods below won't identify individual dark AI visits. GA4 simply can't do that. What they give you is a reasonable estimate of how much dark AI you have, which is enough to justify proper detection or shift budget allocation.
Method 1: Build a "Suspicious Direct" Segment
GA4's Direct bucket contains two groups: people who typed your URL and people who copied it from an AI chat. These groups behave differently.
Create a custom segment in GA4 Explorations:
- Go to Explore and create a new exploration
- Click Segments then the + button, then Session segment
- Add these conditions:
- Session default channel group exactly matches "Direct"
- Session source exactly matches "(direct)"
- New / returning exactly matches "new"
- Add a second condition group (AND):
- Engagement: Average session duration > 60 seconds
- OR Views per session > 2
Name this segment "Suspicious Direct (Possible AI)."
This isolates first-time visitors who arrived without a referrer but showed high engagement. Genuine direct visitors who typed your URL are usually returning users or land-and-leave. New visitors with high engagement and no referrer match the dark AI pattern.
| Visitor Type | Typical Behavior |
|---|---|
| Genuine direct (returning) | Knows the site, goes to dashboard or login |
| Genuine direct (new, typed URL) | Low engagement, high bounce |
| Dark AI (new, pasted URL) | High engagement, multi-page, content-first landing |
Compare this segment's size against your total Direct traffic. In Loamly's data, roughly 3.9% of Direct visitors are actually dark AI. Your percentage depends on your industry and how often AI models recommend your brand.
Method 2: Direct Growth vs Brand Search Divergence
If your Direct traffic is growing but your brand search volume is flat, the gap is likely AI traffic.
The logic: Direct traffic traditionally comes from brand awareness. People know your URL and type it. Brand search volume (people searching your company name in Google) tracks the same awareness signal. When the two diverge, something else is sending no-referrer traffic.
Steps:
- GA4: Pull monthly Direct sessions for the last 6-12 months (Reports, Acquisition, Traffic Acquisition, filter to "Direct")
- Google Search Console: Pull monthly clicks for your brand name queries
- Compare the growth rates
| Month | Direct Sessions (GA4) | Brand Clicks (GSC) | Gap |
|---|---|---|---|
| Sep 2025 | 1,200 | 900 | 300 |
| Oct 2025 | 1,400 | 920 | 480 |
| Nov 2025 | 1,800 | 950 | 850 |
| Dec 2025 | 2,500 | 980 | 1,520 |
Illustrative example. Your numbers will differ.
If Direct grows significantly faster than brand search, the excess is likely AI-referred traffic arriving without referrer headers. ChatGPT alone has 900M+ weekly active users as of early 2026. The copy-paste behavior from AI chats is growing faster than any brand campaign.
This was my original signal. I saw Loamly's Direct traffic grow 126% year-over-year while brand awareness campaigns hadn't changed. That gap is what convinced me to build dark AI detection in the first place.
Method 3: Landing Page Pattern Analysis
Dark AI visitors land on different pages than genuine direct visitors.
In GA4:
- Go to Reports, then Engagement, then Landing page
- Add a secondary dimension: "Session default channel group"
- Filter to "Direct" only
- Sort by landing page
Flag informational content pages with unusually high Direct traffic. Blog posts, guides, product comparison pages. Nobody memorizes /blog/how-to-track-chatgpt-traffic-in-ga4. If a deep content page gets significant Direct traffic, someone pasted that URL from an AI conversation or social platform.
| Page Type | Direct Traffic Likely Source |
|---|---|
| Homepage | Genuine direct (brand awareness) |
| Login or dashboard | Genuine direct (returning users) |
| Blog posts | Suspicious: likely AI or social copy-paste |
| Feature or comparison pages | Suspicious: likely AI recommendations |
| /check/ report pages | Suspicious: likely AI citations |
Tally the Direct traffic to suspicious pages. That is your minimum dark AI estimate. The real number is higher because some AI visitors land on your homepage too, when AI says "check out [brand].com" without a deep link.
Method 4: GSC Clicks vs GA4 Sessions Gap
This method gives directional insight into AI Overviews traffic specifically.
Google Search Console counts clicks from search results. GA4 counts organic sessions. These numbers are never identical (different measurement models), but the gap between them has been growing since AI Overviews expanded to 13% of Google searches.
Steps:
- GSC: Go to Performance, Search results. Note total clicks for a 28-day period
- GA4: Go to Reports, Acquisition, Traffic Acquisition. Filter to "Organic Search" for the same 28-day period. Note sessions
- Compare the ratio over time
If the GA4-to-GSC ratio has increased since mid-2024, part of that increase likely includes AI Overview clicks that GA4 counts as organic sessions but GSC may report differently. I cover this in more detail in How to Track AI Overviews Traffic.
This method is the least precise of the four. Treat it as directional context alongside the other methods.
What You'll Find
Most sites I've seen show dark AI at 2-5% of total Direct traffic. For B2B SaaS and tech companies, it's often higher. For local businesses and e-commerce, it's lower.
Microsoft Clarity's study on AI traffic across 1,200 publisher websites found AI visitors convert at 3x the rate of other channels. Loamly's data shows dark AI specifically converts at 4.1x. This is revenue hiding in your analytics.
The four methods above won't give you exact dark AI visitor counts. But they'll give you enough signal to answer the important question: is AI driving meaningful traffic that my analytics can't see?
The Honest Limitation
These methods give estimates, not detection. You cannot definitively identify an individual dark AI visit in GA4. There is no GA4 dimension, filter, or API that distinguishes "pasted from ChatGPT" from "typed the URL."
What you get is a reasonable upper bound on dark AI volume. That's useful for:
- Showing stakeholders that AI traffic matters
- Justifying investment in proper detection
- Contextualizing why your "Direct" bucket keeps growing
What you don't get: per-visit attribution, conversion tracking by AI platform, or the ability to optimize for specific AI sources. For that, you need server-side detection that goes beyond referrer headers.
I built Loamly's dark AI detection because GA4 can't do this. The AI traffic detection docs explain the multi-signal approach in detail. If GA4 could detect dark AI, I wouldn't have spent four months building a detection engine. That's not a sales pitch. That's the reality of how HTTP referrer headers work.
FAQ
Can I use Google Tag Manager to detect dark AI traffic?
GTM fires after page load, so it inherits the same referrer limitation as GA4. No referrer header means no AI detection, regardless of which tag management system you use.
Does GA4 plan to add AI traffic detection?
Google has not announced plans for dark AI detection. They added an "AI Overviews" appearance type in Search Console, but that only covers visible AI Overview clicks in Google Search. It doesn't cover traffic from ChatGPT, Claude, or Perplexity.
Why does ChatGPT strip referrer headers?
ChatGPT's mobile app uses a WebView that doesn't pass referrer headers. ChatGPT Operator uses headless browsing. And when users copy-paste a URL from any AI chat, the browser has no referrer to send. This isn't a bug. It's how HTTP works when navigation doesn't start from a link click.
How accurate are these estimation methods?
Each method has a different accuracy profile. Method 1 (behavioral segment) tends to overcount because some genuine new visitors also show high engagement. Method 2 (brand search divergence) is the most reliable signal but requires 6+ months of baseline data. Methods 3 and 4 are directional only. Use all four together for triangulation.
Want to skip the estimation and see real numbers? Loamly detects dark AI traffic automatically. No GA4 workarounds required. Real data. No marketing spin.
Stay Updated on AI Visibility
Get weekly insights on GEO, AI traffic trends, and how to optimize for AI search engines.
No spam. Unsubscribe anytime.
Related Articles
80% of Your AI Traffic Is Invisible. Here's What GA4 Misses.
14,413 dark AI visits vs 6,015 visible. Dark AI converts at 10.21%. GA4 dumps it into 'Direct.' Real detection data.
ChatGPT, Claude, and Gemini Recommend Different Companies. Here's the Data.
2,014 companies across 3 platforms. ChatGPT cites 70.4% of the time. Gemini only 57.6%. The divergence matters.
Check Your AI Visibility
See what ChatGPT, Claude, and Perplexity say about your brand. Free, no signup.
Get Free Report