How to Track AI Referral Traffic in GA4 (And What to Do When You Find It)

What is generative AI referral traffic?

Generative AI referral traffic is website traffic originating from AI platforms — ChatGPT, Perplexity, Gemini, Claude, and Copilot — when a user clicks a cited link within an AI-generated response. Unlike organic search traffic, it arrives with purchase or conversion intent already formed inside the AI conversation, before the visitor ever reaches your site.

Why this matters more than you think right now

Most SEO dashboards are quietly lying to you — not through bad data, but through missing categorisation.

AI platforms generated over 1.1 billion referral visits in June 2025, up 357% year-on-year, according to Similarweb’s 2025 Generative AI Report. Across 19 GA4 properties analysed in a Previsible study, total AI-sourced sessions jumped 527% year-on-year between early 2024 and May 2025. In one documented case, a single site went from 600 ChatGPT visits per month in early 2024 to over 22,000 by May 2025.

None of that traffic appears in a default GA4 channel report. It is spread invisibly across Referral, Direct, and Unassigned — which means teams are optimising blind, reporting incomplete numbers to stakeholders, and missing the highest-intent visitors their site has ever received.

There is a second problem underneath the tracking gap. GA4 itself was designed before AI search existed. Its default channel definitions have not caught up with how users actually discover content in 2026. ChatGPT only began appending utm_source=chatgpt.com to citation links in June 2025, making some attribution automatic — but Google AI Overviews, AI Mode, and mobile app referrals from most LLMs still pass no attribution at all. This means the AI traffic your analytics shows is, at best, the visible portion of a larger signal.

What AI traffic actually looks like in GA4 (before you fix it)

Before any custom setup, here is what GA4 does with each source:

Platform	How GA4 categorises it
ChatGPT (desktop, with UTM)	Referral — chatgpt.com
ChatGPT (mobile app, no UTM)	Direct or (not set)
Perplexity	Referral — perplexity.ai
Google Gemini / AI Overviews	Organic — indistinguishable from standard search
Microsoft Copilot	Referral — copilot.microsoft.com
Claude (claude.ai)	Referral — claude.ai

The practical consequence: the AI traffic you can see in a default report is ChatGPT desktop and Perplexity, when they pass referrer data. Everything else — AI Overview clicks, Gemini responses, mobile ChatGPT, and zero-click AI interactions — disappears into your existing channels without any label.

Given that mobile app referrals, AI Overview clicks, and zero-click AI interactions pass no attribution data — what Seer Interactive describes as a new category of dark traffic — the AI traffic visible in your analytics is almost certainly an undercount of actual AI-influenced visits.

Why it’s worth tracking even if the volume looks small

The objection most teams raise when they see their AI traffic numbers: “It’s less than 1% of our sessions. Why does it matter?”

Here is why.

The Previsible State of AI Discovery Report, which analysed 1,963,544 LLM-driven sessions across 12 months, found that AI traffic concentrates on the pages that matter most: industry pages (1.14% AI penetration), tools pages (0.95%), and pricing pages (0.46%) — all 4–9x higher than the 0.13% site-wide average.

These are not awareness visitors. They arrive with a defined problem and, in many cases, a vendor shortlist already formed inside the AI conversation they just had. Which is why the conversion data is so striking.

A Seer Interactive analysis of one B2B client covering October 2024 to April 2025 found:

• ChatGPT: 15.9% conversion rate
• Perplexity: 10.5%
• Claude: 5.0%
• Gemini: 3.0%
• Google Organic: 1.76%

ChatGPT visitors also viewed an average of 2.3 pages per session — nearly double the 1.2 pages for organic search — indicating deeper engagement with content before converting.

At the broader market level, Semrush data from July 2025 found LLM visitors convert 4.4x better than organic search visitors. Separately, a Microsoft Clarity study of over 1,200 websites found AI platform visitors converted to sign-ups at 1.66%, compared to 0.15% from traditional search — an 11x difference.

Small volume. Disproportionately high value. And currently invisible in most dashboards.

How to isolate AI referral traffic in GA4: step by step

GA4 does not have a native AI channel. You will create one using a custom channel group. This takes approximately 15 minutes.

Step 1: Open your GA4 property

Go to Admin → Data display → Channel groups in your GA4 property. You will see Google’s default channel group. Do not edit this — create a new one.

Step 2: Create a new channel group

Click Create new channel group. Name it something clear: “AI Traffic (2026)” works well, since you will need to update the regex as new platforms emerge.

Step 3: Add a new channel definition

Click Add new channel, name it AI Search, and define it with the following condition:

Condition type: Source matches regex

Regex pattern:

chatgpt\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|chat\.openai\.com|bard\.google\.com|meta\.ai

This captures the eight most significant AI referral sources currently passing attribution data to GA4. You can expand it as new platforms emerge.

Step 4: Save and apply

Save the channel group. Note that custom channel groups apply from the date you create them — they do not backfill historical data. If you want to analyse historical AI traffic, use the Exploration reports instead (see Step 6).

Step 5: Check your Traffic Acquisition report

Go to Reports → Acquisition → Traffic Acquisition. In the channel group selector at the top of the table, switch from the default group to your new “AI Traffic (2026)” group. You should now see an AI Search row with session data.

If the row is empty, that does not mean you have no AI traffic — it means no AI traffic has passed referrer data to GA4 since you created the group. Check the Referral channel in the default view and look manually for chatgpt.com and perplexity.ai entries.

Step 6: Backfill analysis using Explore

To analyse AI traffic from before you created the custom group, go to Explore → Free-form exploration. Set the following:

• Dimension: Session source / medium
• Filter: Session source contains chatgpt OR perplexity OR claude OR gemini OR copilot
• Metrics: Sessions, Engaged sessions, Conversions, Engagement rate

This will surface historical AI referral data for any period where ChatGPT or Perplexity passed referrer information.

Step 7: Add a secondary dimension for landing pages

In the Traffic Acquisition report (or your Exploration), add Landing page as a secondary dimension. This tells you which specific pages on your site are being cited by AI platforms — which is the most actionable finding in the entire exercise.

What to look for once you have the data

Once your AI channel is live and populated, you are looking for four things.

1. Volume trend, not absolute volume. The absolute number will likely be small. What matters is whether it is growing month-on-month. Flat or declining AI referrals on a growing site is a GEO signal: your content is not being cited. Growing AI referrals on a site with weak traditional SEO is a strong positive sign — it means AI engines are finding your content useful even before Google does.

2. Which pages are getting cited. The landing page dimension will show you exactly which URLs are appearing in AI responses. Cross-reference these against your key commercial pages. If AI is sending traffic to your blog posts but not to your services or pricing pages, that is a content gap: your commercial pages are not structured in a way that makes AI engines confident enough to recommend them.

3. Conversion rate versus organic. Compare your AI channel’s conversion rate against your Organic Search channel in the same date range. If your AI conversion rate is already higher — even with tiny volume — it validates that the citation strategy is working. If it is lower, the issue is usually that the AI is citing informational content and users arriving at blog posts have no clear next step.

4. Which platforms are sending traffic. ChatGPT currently dominates AI referrals with 84.2% share across analysed sites. But only 12% of sources cited across ChatGPT, Perplexity, and Google AI Overviews overlap — meaning each platform is citing different content. If you are only getting ChatGPT referrals and nothing from Perplexity, your content may not be structured for the citation patterns Perplexity uses (which skew toward more technical, citation-dense content).

Perplexity traffic in GA4: what makes it different and what to do with it

Perplexity is the easiest AI referral source to track in GA4 — and the one most worth isolating.

Unlike ChatGPT, which only began passing UTM parameters in June 2025 and still drops attribution entirely from mobile app traffic, Perplexity consistently passes perplexity.ai as a referrer across both desktop and mobile. This means GA4 captures it reliably in your Referral channel without any workaround. If you have set up the custom channel group from the walkthrough above, your Perplexity sessions will appear cleanly within the AI Search channel from day one.

What makes this traffic strategically distinct from other AI referrals is the intent signal it carries. Perplexity searches the live web before generating every response and displays source links inline — users see which sites contributed to the answer. A click-through from Perplexity means someone read an AI-synthesised summary that named your site, evaluated the attribution, and chose to visit. That is a fundamentally different entry point from an organic Google click.

The conversion data reflects this. The Seer Interactive B2B analysis found Perplexity referral traffic converting at 10.5% — six times the 1.76% rate for Google organic search over the same period. The volume is small relative to ChatGPT (which currently accounts for 84.2% of AI referrals across analysed sites), but the per-session value is disproportionately high.

How to read your Perplexity data in GA4

Once your AI Search channel group is live, add Landing page as a secondary dimension and filter to sessions where the source is perplexity.ai. You are looking for two things:

Which pages Perplexity is citing. If the landing pages are blog posts and not your commercial or services pages, that is a content structure issue — not a volume issue. Perplexity’s retrieval system favours pages with clear definitions, structured data, and quotable standalone statements. Your informational content likely has these; your commercial pages likely do not.

Whether Perplexity traffic is converting differently from ChatGPT traffic. In most B2B cases, Perplexity visitors engage more deeply (higher pages per session) but arrive in lower volume. If your Perplexity conversion rate is significantly lower than expected, check whether the cited pages have a clear next step — Perplexity visitors have already had the problem explained to them by the AI, so awareness-stage CTAs underperform.

If you are getting zero Perplexity referrals

Three things to check before assuming your content is not being cited. First, confirm PerplexityBot can actually crawl your site — many security plugins and WAF configurations block AI crawlers by default. Check your robots.txt for an explicit Allow directive for PerplexityBot. Second, run your priority queries directly in Perplexity to see whether your domain appears in responses even without click-through — citation without clicks is common and indicates your content is being used but not linked prominently enough. Third, review whether your content structure matches what Perplexity’s retrieval system rewards: definition-first paragraphs, FAQPage schema, and explicit author attribution with linked credentials.

If you want the full technical breakdown of what drives Perplexity citation decisions — from schema markup specifics to content extraction patterns — I cover all seven strategies in How to Get Cited by Perplexity: 7 Proven Strategies.

What low AI referral numbers actually mean

If you set up the tracking and find near-zero AI referral traffic, there are three likely causes — each with a different fix.

Cause 1: Your content is not structured for citation. AI engines consistently cite content that has clear definitions, direct answers to specific questions, and structured formatting. A page with long flowing prose, no subheadings, and no FAQ section is difficult for an AI to extract a quotable answer from. The fix is content restructuring — not more content.

The fix is content restructuring — not more content. For a full guide on which schema types to implement for AI citation readiness, see Structured Data for AI Search.

Cause 2: Your entity is not established. AI engines are more likely to cite sources they recognise as authorities on a topic. If your domain has no Knowledge Panel, limited external mentions, and no consistent entity signals across the web, your content may be accurate but unverifiable from the model’s perspective. The fix is E-E-A-T and entity building — author bylines, external press mentions, structured About page data.

Cause 3: You are not targeting the right queries. AI traffic concentrates on decision and comparison queries, not awareness queries. Previsible’s analysis found pricing pages have 3.5x higher AI presence than the site average — because AI users arrive with defined problems and use AI to compare solutions. If your content strategy is 80% TOFU awareness content and 20% decision content, you are optimising for a channel that AI is not the strongest at driving.

Three things to do this week, depending on what you find

If you find AI traffic and it is converting well: Document which pages are being cited, analyse their structure, and replicate that structure across your highest-value commercial pages. The pages already getting cited are your GEO templates.

If you find AI traffic and it is not converting: The issue is CTA misalignment. The pages being cited are likely informational — they answer a question but do not offer a clear next step. Add a contextual CTA to each cited page that reflects the funnel stage of someone arriving from an AI response: they already understand the problem, so skip the awareness pitch.

If you find very little or no AI traffic: Run your priority content pages through a GEO readiness check. Specifically: does each page have a definition block in the opening paragraph? Does it directly answer the question implied by its primary keyword? Are there structured FAQ questions using natural language? If not, those are your first fixes before any broader content strategy changes.

The measurement gap is the strategy gap

The teams who will build the strongest GEO presence over the next 18 months are not the ones who wait until AI traffic is large. They are the ones setting up measurement now — when the data is thin enough to act on cleanly — and using what they find to shape content decisions before competitors are even looking.

The regex filter above takes 15 minutes. The insight it surfaces — which of your pages AI engines are already citing, at what conversion rate, from which platforms — is the most actionable signal available in SEO right now.

If you want to understand what your site looks like from the perspective of an AI engine before you run the GA4 setup, run your key pages through the AEO Article Analyzer first. It scores content against the 10 structural criteria AI engines use to decide what to cite, in under 30 seconds.

Or if you want a complete picture of your site’s GEO readiness — including entity signals, citation gaps, and on-page structure across your priority pages — the free audit covers all of it.

---

Frequently Asked Questions