AI referral traffic GA4 segments: how to isolate and read a behaviorally distinct audience

If you're lumping ChatGPT and Perplexity referrals into the same bucket as Google organic, you're probably throwing away one of the more interesting signals in your analytics. Blended acquisition reporting averages the two cohorts together — and whatever is distinct about AI-referred behavior disappears into that average.

Visitors arriving from ChatGPT, Perplexity, or Google AI Overviews have already consumed an AI-synthesized answer before they click through to your site. Some of them arrive later in the decision process because the assistant has already supplied context, comparisons, or a recommendation. Where that difference exists, it shows up in the data — but most GA4 setups don't separate this cohort at all, so the signal gets diluted into organic or direct and disappears.

Here's how to surface it.

How we know this

The behavioral claims in this article draw on named, published datasets: Semrush's June 2025 study of 500+ digital-marketing and SEO topics, Previsible's 2026 AI traffic report covering 6.77 million LLM-driven sessions across 166 GA4 properties, and Jason Tabeling's 13-month GA4 analysis published on Search Engine Land. Implementation details are cross-checked against Google's own GA4 channel documentation, and the AI Overview click data comes from Seer Interactive's ongoing CTR research. Where our own instrumentation via RankSage's rs.js beacon produces comparable signals, we note it. One rule we hold ourselves to throughout: when a widely-circulated number can't be traced to a primary study, we say so rather than citing it.

Four kinds of "AI traffic" — and what GA4 can actually see

Before building anything, be precise about which measurement problem you're solving. Four very different things get blurred together as "AI traffic," and they land in completely different places in GA4:

This article shows how to isolate the first cohort cleanly, infer the second, and watch for the third. The fourth is why AI citation tracking exists as a separate discipline.

What the behavioral gap actually looks like

AI referrals are still tiny in most GA4 accounts. That's the frustrating part. In Previsible's dataset — 6.77 million LLM-driven sessions across 166 GA4 properties — they peaked at 1.71% of total sessions in the best-covered vertical (SMB) as of May 2026, and sat well under 1% in most others. But you can have 50 sessions from ChatGPT in a month and still see a conversion pattern that looks meaningfully different from your organic traffic, which is exactly why the cohort is worth isolating.

In Semrush's June 2025 study of 500+ digital-marketing and SEO topics, the average AI search visitor was 4.4 times as valuable as the average organic search visitor, measured by conversion rate. That dataset and baseline are specific to one industry cluster, so treat the multiplier as directional rather than a universal benchmark. Growth is broader-based: Previsible measured monthly LLM-driven sessions growing nearly 10× between November 2024 and May 2026 across its 166 properties (an earlier cut of the same research recorded 527% growth in just the first five months of 2025). And in Jason Tabeling's separate 13-month GA4 analysis on Search Engine Land, LLM referral traffic converted at approximately 18% — the highest of any acquisition channel in that dataset — despite being under 2% of referral sessions. The industry mix and conversion definition behind that 18% are specific to one dataset, so read it as a reason to measure your own cohort rather than a number to expect.

Per-platform behavior differs too — practitioner writeups consistently report Perplexity referrals showing longer sessions and deeper page-depth than Google-referred visits. But the specific engagement figures circulating for this claim (10.3 minutes and 13 pages per session) trace only to vendor roundups quoting each other, not to a primary study we could verify — so we won't reproduce them as fact. The segment setup below makes measuring per-platform engagement on your own property a five-minute job, which beats borrowing anyone's benchmark.

These published gaps are large enough to take seriously. If your own AI-referred cohort shows materially different conversion behavior — and the published data says that's likely — it deserves its own segment, even at low volume.

Why GA4's defaults obscure this traffic

On May 13, 2026, GA4 added a native AI Assistant channel that automatically classifies sessions from recognized assistants — Google's documentation currently lists ChatGPT, Gemini, DeepSeek, Copilot, and Grok — setting medium to ai-assistant and campaign to (ai-assistant), no UTM tagging required. Classification is forward-only: traffic from before the launch stays wherever it originally landed. And it triggers only when the referrer matches a recognized assistant, so it covers only known platforms and only when referrer headers survive.

The structural problem: many AI surfaces and in-app browsers strip referrer headers entirely, so a meaningful share of AI-influenced visits lands as Direct or (not set). Important: Google's documentation is explicit that the AI Assistant channel excludes Google AI Overviews and AI Mode — clicks from both are classified as Organic Search, in GA4 and Search Console alike, blended invisibly into standard organic metrics. There is no clean source tag to filter on.

We wouldn't use the native AI Assistant channel as your definition of AI traffic. It's useful, but it's fundamentally a referral classifier — if the referrer disappears, GA4 has nothing to classify. That distinction matters because AI influence is bigger than AI referral traffic, and you need custom segments on top of the native channel to see any of the rest.

How to build GA4 segments for AI referral traffic

Step 1: Create a custom channel group

In GA4, go to Admin → Data Display → Channel Groups → Create new channel group. Add a rule set that matches Session source against a regex covering known AI referrers:

chatgpt\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|you\.com|phind\.com

Name this channel "AI Assistants" or "LLM Referrals." Place this rule above the default Referral rule in the priority order — GA4 evaluates channel rules top-down, and if AI sessions hit the Referral rule first, they disappear into the generic referral bucket. This isn't just practitioner lore: Google's own custom channel group documentation uses an AI-assistants example with a considerably broader regex (covering openai, gpt, bard and more) and gives exactly this instruction — reorder the list so AI Assistants sits above Referrals. And check first whether the native AI Assistant channel already captures the sources you need: build the custom group when you want additional platforms, historical comparability (the native channel is forward-only from May 2026), or your own classification logic.

Step 2: Build a segment in Explorations

In GA4 Explorations, create a user segment with the condition: Session source contains (use the same domain list). This lets you run free-form analysis — funnel exploration, path analysis, cohort comparison — on just the AI-referred cohort.

Split by platform if volume allows:

We'd still split this by assistant. ChatGPT and Perplexity may both show up as "AI traffic" in your dashboard, but that doesn't mean they send the same kind of visitor — different assistants cite different content types and pull people in at different stages of a decision. If you have the volume, keep the platforms separate.

Step 3: Infer AI Overview impact separately

Since AI Overview clicks register as organic, you can't segment them by source. The practical workaround: monitor queries in Search Console where impressions are rising but CTR is falling. Seer Interactive's analysis of 3,119 informational queries across 42 organizations (June 2024 – September 2025) measured organic CTR falling from 1.76% to 0.61% — a 61% decline — on queries where an AI Overview appears, with paid CTR down 68%. The magnitude varies by position, query type, and window; the same study found a 41% organic CTR decline even on queries without AI Overviews, so treat the numbers as directional. The same research found brands cited inside AI Overviews earned 35% more organic clicks and 91% more paid clicks than non-cited brands on the same queries. That is an observational relationship — it doesn't prove citation causes the lift — but the divergence between rising impressions, falling CTR, and cited-brand performance is your signal.

What behavioral signals are worth tracking at low volume

Don't wait for thousands of sessions before you start looking at this. If 50 ChatGPT referrals are converting noticeably differently from your organic visitors, that's already worth investigating — you won't get a statistically strong benchmark out of it, but you can see which pages are getting cited, where those visitors go next, and whether they reach a meaningful conversion event. The signals that matter most early:

Landing page mix. Which pages are AI assistants citing? AI traffic tends to concentrate on content with structured evidence — comparison tables, original data, clearly cited statistics. If your AI-referred visitors land disproportionately on one or two pages, those pages are your citation anchors. Protect and expand them.

Funnel depth. Do AI-referred visitors skip top-of-funnel content and land directly on product, pricing, or comparison pages? If so, they're arriving later in the decision cycle than organic visitors. A funnel exploration in GA4 comparing the two cohorts will show this directly.

Scroll depth and time-on-page by content type. Use GA4's scroll events (or RankSage's rs.js beacon for more granular instrumentation) to compare engagement on evidence-heavy pages versus narrative pages. Whether AI-referred visitors read cited pages more carefully is a testable hypothesis, not a given — run the comparison and let your own data answer it. A visitor who arrived through a citation may also leave faster if the page doesn't deliver what the citation promised; both outcomes are informative.

Conversion path length. Do AI-referred visitors convert in fewer sessions? The ~18% conversion rate in Tabeling's 13-month dataset suggests many arrive close to a decision. A shorter path-to-conversion for your own cohort would confirm that and change how you think about what content to put in front of them.

Branded search lift. Because referrer stripping hides a share of AI-influenced visits as Direct or branded organic, watch for correlated growth in branded search volume alongside AI citation volume. If your AI Overview or ChatGPT citations increase and branded direct traffic rises in the same period, that's consistent with a dark-funnel effect — but correlation isn't confirmation. Strengthen the inference with first-touch surveys ("how did you hear about us?") or other attribution signals before reallocating budget on it.

Where this breaks down

Referrer stripping is the biggest limitation. Custom channel groups built on domain matching only capture sessions where the referrer header survived. The true volume of AI-influenced visits is larger than what any GA4 segment will show.

The conversion-rate figures (Semrush's 4.4× multiplier, the ~18% in Tabeling's dataset) come from datasets weighted toward specific industries and early adopters — Semrush's is literally a study of marketing and SEO topics. Extrapolating them to every vertical or business model is premature. Measure your own cohort before reorienting content strategy around aggregate benchmarks.

GA4's native AI Assistant channel is also constrained to platforms Google has recognized and depends on correct referrer detection. New AI interfaces — browser-native assistants, embedded copilots, voice-to-search — may not be classified correctly for months after launch. Custom channel groups need periodic maintenance as the AI referrer landscape shifts.

Finally, higher engagement from any single platform's visitors may partly reflect user self-selection: people who choose a research-first assistant like Perplexity tend to be more deliberate researchers generally. Segment the data, but don't assume the engagement is purely a function of the referral source — distinguish the channel effect from the audience-composition effect.

FAQ: adjacent questions analytics teams are asking

Does GA4 automatically track ChatGPT referrals without any setup?

Out of the box it covers only the recognized platforms. GA4's native AI Assistant channel (added May 13, 2026) classifies sessions from recognized platforms — currently ChatGPT, Gemini, DeepSeek, Copilot, and Grok per Google's documentation — automatically when referrer headers are present. But referrer stripping means a significant share of AI-influenced visits still lands as Direct, classification is forward-only from the launch date, and emerging AI tools won't be classified until Google adds them. Custom channel groups and segments remain necessary for complete coverage.

How do I track Google AI Overview traffic separately in GA4?

You can't, cleanly. GA4 and Search Console both classify AI Overview clicks as organic search. The practical approach is to monitor Search Console for queries where impressions rise but CTR falls, and separately track branded direct traffic for correlated growth. Text fragment parameters (#:~:text=) have been used to tag snippet-driven visits via GTM, but this technique now applies mainly to Featured Snippets rather than AI Overviews.

What conversion rate should I expect from AI referral traffic?

The strongest published benchmark is approximately 18%, from Jason Tabeling's 13-month GA4 dataset, with Semrush's June 2025 study separately measuring AI search visitors at 4.4× the value of organic search visitors by conversion rate. Both datasets skew toward specific verticals and toward brands already being cited in AI answers. Treat them as directional signals and measure your own cohort before planning around either number.

Should I create separate segments for each AI platform or one combined segment?

Run both. A combined "AI Assistants" segment gives you aggregate volume for trend analysis, and per-platform segments (ChatGPT, Perplexity, Claude, Gemini) show you which assistant is citing which content. Practitioner datasets frequently report Perplexity referrals engaging more deeply than other sources — but the circulating per-platform numbers are poorly sourced, which is exactly why measuring the difference on your own property matters. In a combined segment, that difference is invisible.

What if my AI referral volume is too low to be statistically meaningful?

Don't throw the segment away — just don't pretend 30 sessions can establish a conversion-rate benchmark. Use them to learn which pages are getting cited, where those visitors land, and whether they reach meaningful events. GA4 Explorations helps you inspect paths and landing pages at that volume, though it won't make a small sample statistically significant. As the cohort grows, the pattern you established becomes the baseline you compare against.

Sources

• Semrush — AI search traffic study: the average AI search visitor is 4.4× as valuable as organic by conversion rate (June 2025; 500+ digital-marketing and SEO topics)
• Previsible — 2026 AI Traffic Report: 6.77M LLM sessions, 166 GA4 properties, ~9.9× growth Nov 2024 → May 2026; ChatGPT 92.4% of trackable LLM referrals
• Jason Tabeling, Search Engine Land — What 13 months of data reveals about LLM traffic, growth, and conversions (~18% conversion; Feb 2026)
• Google Analytics Help — Default channel group: AI Assistant definition; explicitly excludes Google AI Overviews and AI Mode
• Google Analytics Help — Custom channel groups: AI-assistants example and rule-ordering guidance
• Seer Interactive — AIO impact on Google CTR, September 2025 update: −61% organic / −68% paid CTR on AI Overview queries; cited brands +35% organic / +91% paid clicks (3,119 queries, 42 orgs)

RankSage tracks AI citations across ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews and connects that citation data to GA4, Search Console, and first-party behavioral signals in one attribution view — so you can see not just that an AI assistant cited you, but which pages it cited and whether those visitors converted. Join Waitlist