GA4 has an AI Assistant channel now. It still won't tell you how much AI traffic you get

Dilshat Rakhimov
July 2026 · 9 min read

On 13 May 2026 Google added an AI Assistant channel to GA4's Default Channel Group. Within a week I had the same question from three different clients: "so we can finally see our ChatGPT traffic?"

Partly. You can now see the slice of AI traffic that arrives with a referrer header, from a list of assistants Google doesn't fully publish. Published clickstream work puts that slice at under a third of the total. The rest lands in Direct, and no channel group - native or custom - will pull it back out.

This is a short, honest map of what the native channel does, what it misses, how to build the custom group that covers the referrer gap, and what to do about the part that can't be fixed with regex.

What the native channel actually does

The mechanism is one line in the help centre: GA4 checks the referrer against an internal list of AI assistants, and when it matches it sets medium to ai-assistant and campaign to (ai-assistant). The channel rule itself is then trivial - "The medium exactly matches ai-assistant".¹

Three consequences that matter more than they look:

You can't extend the list. The channel keys off a medium Google assigns. Nothing you configure changes which referrers get that medium. Your only move is to build a second, custom channel alongside it.

Google's own two pages disagree on the list. The Default Channel Group reference describes it as traffic "from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok".¹ The What's New announcement instead named ChatGPT, Gemini and Claude.² Perplexity appears in neither. Treat any list you read - including this one - as a snapshot, and check your own Referral report for what's still leaking through.

AI Overviews and AI Mode are excluded, deliberately. Google's documentation says so explicitly.¹ That's the right call: the click still came from google.com, so it belongs in Organic Search. But it means "AI Assistant" is not the same thing as "AI-influenced", and you should not present it that way internally.

And it isn't retroactive. GA4 processes channel assignment daily, so everything before your property's rollout stays classified the way it was. Don't try to rebuild history; draw a line and annotate the date.

Gap one: the assistants that aren't on the list

This is the easy half. Build a custom channel group and put an AI channel above Referral.

Admin → Data display → Channel groups → Create new channel group. Add a channel, condition on Source, and use partially matches regex if your property offers it:

Custom channel groups are the one place in GA4 that offers a partial-match regex operator. If you only see plain matches regex - the full-match variant used in Explorations and segments - wrap the whole alternation: .*(chatgpt|openai|perplexity|...).* Getting this wrong is the single most common reason a correct-looking pattern returns zero sessions.

Then fix the order. GA4 evaluates channel rules top to bottom and stops at the first match, so an AI channel placed below Referral will never fire - referral catches those sessions first. A working order: Direct, Cross-network, Paid Search, Organic Search, Paid Social, Organic Social, AI, Referral, Unassigned.

Two notes before you ship it. Custom channel groups apply retroactively in reports, which is genuinely useful here - unlike the native channel, this one will re-classify your history. And keep the custom channel separate from the native one rather than trying to replicate it: the native ai-assistant medium is a clean, Google-maintained signal, and you want to be able to see when its coverage changes.

Gap two: the one you can't regex away

Now the part that matters.

Loamly's State of AI Traffic 2026, built on 446,000+ visits, found 70.6% of AI-driven traffic arrives with no referrer at all; only 29.4% carries a recognisable AI referrer.³ Every one of those referrer-less sessions lands in Direct, in the native channel and in your custom one equally.

Why it happens is mundane:

• Native and desktop apps. A link opened from the ChatGPT or Claude desktop app, or from a mobile app's in-app browser, frequently arrives with no Referer header at all.
• Copy-paste. People copy the URL out of the answer and paste it into a fresh tab. That's a typed-in visit as far as any browser is concerned.
• Referrer policy. Some assistants set a restrictive referrer policy, which strips the header on the way out by design.

None of these leave a fingerprint you can pattern-match. There is no header to write a rule against. Any post promising a regex that recovers Direct-bucketed AI traffic is selling you a heuristic and calling it a measurement.

Sizing your own blind spot

You can't recover the sessions, but you can size the gap - and a sized gap is a usable number.

1. Self-reported attribution. Add one required field to the lead or signup form: How did you hear about us? with "ChatGPT / AI assistant" as an explicit option, and push the answer into the CRM alongside the GA4 client ID. Then each quarter compare self-reported AI leads against leads GA4 attributed to the native AI Assistant channel plus your custom one. The ratio between them is your blind-spot multiplier, measured on your own traffic rather than borrowed from someone else's study. It's the only method here that produces a defensible number.

2. Landing-page shape, as a directional index only. Assistants send people deep - to the specific answer page, not the homepage. A new-user Direct session landing on a deep article behaves nothing like a returning user typing your domain. Build the segment if you like, watch its trend, and label it clearly as an index. Do not put it in a channel report and do not let anyone report it as attributed traffic.

3. Crawler hits, which are a different metric entirely. GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot and friends in your server logs tell you whether your content is being read for citation. That's supply, not traffic - a leading indicator that your pages are eligible to be cited, weeks before any human clicks. Pull the user-agent strings from each vendor's published crawler documentation rather than from a blog post; the lists change.

About those conversion-rate multiples

Every write-up on this topic quotes a multiple. The published ones, as of mid-2026:

Source	Basis	AI vs organic search
WebFX	2.3B sessions, Jan 2024 – Dec 2025	~1.2×
118 B2B SaaS sites	conversion to payment, 2.7% vs 1.4%	~1.9×
Semrush	cross-industry	~4.4×
Opollo	312 B2B tech firms, 14.2% vs 2.8%	~5×

A range of 1.2× to 5× across four studies is not four measurements of the same thing. It's four different definitions of "AI traffic".

The spread has an obvious mechanical cause, and it's the same one this whole article is about. If you only count the ~30% that carries a referrer, you're measuring a selected population - people who clicked a link inside an answer, mid-task, further down the funnel than average. Small denominator, self-selected numerator, big multiple. The other 70% sitting in Direct converts like Direct converts, and nobody's multiple includes it.

So: don't take a multiple from a blog post to your CFO. Report the absolute count of AI-attributed key events, next to your self-reported ratio, next to the date the native channel went live on your property. Three honest numbers beat one impressive one.

The setup, end to end

1. Confirm the native channel is live on your property - check for ai-assistant in Session medium. Rollout reached broad availability around 7 June 2026; if you see nothing after that, you're looking at pre-rollout data.
2. Build the custom channel group for the assistants the native list misses, Perplexity first.
3. Move the AI channel above Referral and verify with a test session.
4. Mark the AI channel's key events. If nothing there is a key event, none of this measurement changes a decision.
5. Add the self-reported attribution field and wire it to the CRM. This is the step people skip, and it's the one that produces the real number.
6. Add a monthly AI-crawler report from server logs, kept separate from traffic reporting.
7. Annotate the rollout date and don't rebuild history.

The takeaway

The native AI Assistant channel is a real improvement - it's free, it needs no configuration, and it cleans AI referrals out of your Referral bucket. It is also a referrer-based classification of a traffic source that mostly doesn't send a referrer, and no amount of channel-group work changes that.

Measure the part you can see, size the part you can't with self-reported attribution, and be precise in how you label both. The failure mode here isn't under-measuring AI traffic. It's presenting 30% of it as if it were all of it, and building a content strategy on the multiple that falls out.

Dilshat Rakhimov
Growth Analytics & Digital Architecture

¹ GA4 Default channel group definitions — Analytics Help
² Google Analytics adds AI Assistant channel to measure AI traffic — Search Engine Land, 14 May 2026
³ State of AI Traffic 2026: Industry Benchmark Report — Loamly
⁴ Gen AI search trends study — WebFX
⁵ The 2026 AI Search Benchmark Report — Opollo