Set up to track AI search traffic in GA4 and you'll hit a bad surprise fast: most of it isn't there. ChatGPT, Perplexity, and Copilot readers land on your site, but GA4 files them under Direct because the referrer header never arrives. That's not a rounding error. Some properties miss 35% to 70% of AI sessions this way. This piece covers what GA4's native AI Assistant channel catches, the regex segments that catch what it misses, and a conservative way to put a revenue range on traffic you can't fully see. If your budget meeting needs a defensible number instead of a guess, start here.
Why AI Traffic Disappears Into Direct Traffic
Two behaviors cause this, and neither is a bug you can patch on your own site.
Most people don't click the small citation link inside an AI answer. They copy the URL and paste it into a new tab, or they type the brand name straight into a search bar. Either way, the browser opens a fresh session with no referrer, and GA4 has nothing to tag it with beyond Direct.
Mobile apps make it worse. ChatGPT's mobile apps typically don't pass a referrer when a user taps a link inside the app. Perplexity is the exception: its mobile app passes perplexity.ai as the referrer in roughly 17% of sessions, the best rate among the major AI platforms. That's still fewer than one in five sessions labeled correctly.
A visitor who arrived because ChatGPT recommended your product gets counted the same as someone who typed your domain from memory. You can't fix this with better tagging. You can only work around it with segmentation and estimation, which is what the rest of this piece covers.
Setting Up GA4's Native AI Assistant Channel
GA4 added a default channel group called AI Assistant in May 2026. It automatically recognizes referrals from ChatGPT, Gemini, and Claude, and files them separately from Direct and Organic Search.
To check it, go to Reports, then Acquisition, then Traffic acquisition, and look for AI Assistant under the Session default channel group dimension. If it's missing, your property may be running an older channel grouping definition. Check Admin, then Data display, then Channel groups, and confirm you're on the current default.
Here's the catch: this channel only catches sessions that arrive with an intact referrer header. It still misses 35% to 70% of actual AI-referred sessions, because copy-paste behavior and mobile apps strip the header before the session even starts. Treat the AI Assistant number as a floor, not a ceiling. It shows you the AI traffic you can prove exists, not the AI traffic that actually exists.
Custom Regex Segments to Catch What GA4 Misses
The native channel doesn't cover Perplexity, Copilot, or newer platforms by default, and it can lag on updates. Build a custom channel group or exploration segment with a regex match on the session source dimension instead.
A working pattern to start with:
chatgpt\.com|chat\.openai\.com|perplexity\.ai|copilot\.microsoft\.com|claude\.ai|gemini\.google\.com
Steps: 1. In GA4, open Explore, then Free form. 2. Add "Session source / medium" as a dimension. 3. Create a segment with a condition: session source matches regex, and paste the pattern above. 4. Apply it alongside your existing channel groups and compare totals.
Prioritize ChatGPT patterns first. A review of 166 GA4 properties found 6.77 million LLM-driven sessions, and ChatGPT accounted for 92.4% of all trackable LLM referral traffic in that sample. Perplexity, Claude, and Copilot make up the rest, but ChatGPT is where the regex work pays off fastest.
Estimating AI-Influenced Sessions From Landing Page Patterns
Even with regex segments running, you're still only catching sessions that kept a referrer. To estimate the rest, look at behavior instead of source.
Watch for Direct sessions landing on deep, specific URLs rather than the homepage. Someone who types your domain from memory usually lands on the homepage. Someone who copy-pasted a link from an AI answer usually lands on the exact article or product page that answer cited.
Cross-reference timing too. If Direct traffic to one specific page spikes the same week you know you were cited by ChatGPT or Perplexity, that's a strong signal the spike is AI-driven rather than brand recall.
Neither method gives you a precise count. Both beat assuming every Direct session already knew your brand.
Calculating True AI Conversion Lift on Unattributed Sessions
Here's where the incomplete data still earns its keep. Even on the sessions you can label, the conversion difference is large enough to matter.
One agency dataset shows 73% of visitors from AI referrals convert during their first session, against 23% for Google organic visitors on the same properties. A separate sample of 200 Stripe-connected sites found a similar pattern at the payment level: AI-referred visitors convert to a Stripe payment at 2.7%, versus 1.4% for Google organic on the same landing pages.
To turn that into your own lift estimate: 1. Take your measured AI-referred conversion rate, from the AI Assistant channel plus your regex segment. 2. Take your organic conversion rate on the same landing pages. 3. Divide the two to get your own multiplier — don't borrow someone else's 2.7x, since it shifts by vertical. 4. Apply your multiplier only to traffic you can actually measure. Resist stacking it onto dark-traffic guesses; that compounds one estimate on top of another.
Revenue Benchmarks by Platform: Why Perplexity Beats Gemini
Once you can segment by platform, revenue per visitor tells you where to put effort next. The numbers vary almost 5x across major AI referral sources, based on an analysis of 200 Stripe-connected sites.
| Platform | Revenue per visitor (RPV) |
|---|---|
| Perplexity | $1.42 |
| Claude | $1.18 |
| ChatGPT | $0.87 |
| Gemini | $0.41 |
| Google AI Overviews | $0.29 |
The gap between Perplexity and Gemini is close to 3.5x, and Perplexity outperforms Google's own AI Overviews by roughly 4.9x. If you're deciding where to spend content or outreach effort first, this is the order to work in — not "AI search" treated as one bucket.
Google AI Overviews sits at the bottom partly because it's largely a zero-click surface. Around 83% of searches that trigger an AI Overview end without any click at all, which caps how much revenue that channel can ever return per query.
Building a Conservative Attribution Model for Budget Planning
Now put it together into one model you can defend in a budget meeting.
Start with your measured AI sessions: native channel plus regex segment. Add a conservative estimate for dark traffic. One B2B SaaS sample put a median of 41% of Direct/(none) traffic as actually AI-referred, but that's a median across a specific, Stripe-connected sample, not a universal rate for every business. Use a lower bound, such as half that share, until you've checked it against the landing-page and timing patterns described earlier.
Multiply your total estimated AI sessions by your platform-specific revenue-per-visitor figure or your own measured conversion lift, whichever you trust more for your business. Report a range, not a point estimate. "$14,000 to $31,000 in AI-influenced revenue this quarter" is honest. A single tidy figure implies a precision the data doesn't support.
Revisit the model every quarter. Perplexity's referrer-passing rate, GA4's channel definitions, and the mix of AI platforms your buyers use will all shift faster than an annual review can catch.
Frequently asked questions
How much traffic am I actually getting from AI search if most visitors show up as Direct?
More than GA4 shows you, but nobody has an exact number. One analysis puts the gap at 35% to 70% of AI sessions missing referrer data entirely, and a separate B2B SaaS sample found a median 41% of Direct/(none) traffic was actually AI-referred. Use both figures as a range, not a single answer, until you validate against your own landing pages.
How should I set up GA4 to track AI referral traffic accurately?
Start with the native AI Assistant channel under Traffic acquisition, then layer a custom regex segment covering chatgpt.com, perplexity.ai, claude.ai, and copilot.microsoft.com. Neither source is complete alone; together they catch more than either does by itself.
Which AI platform should I focus on first — ChatGPT, Perplexity, or Google AI Overviews?
On volume, ChatGPT wins: it accounted for 92.4% of trackable LLM referral traffic in one 166-property sample. On revenue per visitor, Perplexity wins at $1.42 versus ChatGPT's $0.87 and Google AI Overviews' $0.29. Track both metrics before you pick a single priority.
How do I measure ROI from AI search optimization when conversion rates are high but traffic volume is still small?
Multiply your own measured AI conversion rate, not a borrowed industry figure, by your own AI session count, then treat any dark-traffic estimate as a separate, clearly labeled line. A small, provable number holds up better in a budget conversation than a large, unprovable one.
Measure What You Can, Estimate the Rest Honestly
Set up the AI Assistant channel and a regex segment this week — it's roughly a 20-minute job inside GA4 Explore. Then build the conservative dark-traffic range from the last section and bring both numbers, the measured floor and the estimated range, into your next budget conversation. Don't wait for a perfect attribution model; it isn't coming. GA4 will keep undercounting AI traffic for the foreseeable future, and the teams that plan around that fact will out-argue the ones still reporting a single made-up number.
Sources
- The Agency Guide to Tracking AI Traffic in GA4 — Setup, Regex Patterns, and More - Swydo — Copy-paste/referrer-stripping mechanics and the 73% vs 23% first-session conversion comparison
- The 2026 AI Search Revenue Benchmark: Real Data From 200 Stripe-Connected Sites | Attrifast — Stripe payment conversion rates, per-platform RPV benchmarks, and the 41% Direct/(none) AI-referred median
- How to Track AI Referral Traffic in GA4 | GrowthX — Perplexity's mobile app referrer-passing rate
- Referral traffic in Google Analytics: how to track AI and direct visits in GA4 | Catalog — 166 GA4 properties, 6.77 million LLM-driven sessions, and ChatGPT's 92.4% share of trackable traffic
- PPC vs SEO vs AI Search 2026: Which Channel Drives Better ROI? — GA4's native AI Assistant channel launch date and its 35-70% attribution gap
- AI Overviews and Organic Traffic: What the 2026 Data Actually Shows - Contently — The 83% zero-click rate on searches that trigger an AI Overview