Somebody in the room asks whether your ChatGPT visitors behave any differently than your Perplexity visitors. You open GA4 to check, and the honest answer lives in two different places at once. Two of the three assistants in the question get sorted automatically. The third does not, and nothing your team configured caused that gap.
Nobody on your team did this wrong. Google decides which assistants get named in GA4’s native channel, and its own documentation is not even consistent with itself about which names those are. That inconsistency matters more than it sounds like it should, because it is the whole reason a clean answer is not available yet.
Can You Actually Tell Them Apart?
Yes for the assistants GA4 recognizes, no for the rest, verified against Google’s docs on August 18, 2026. GA4’s native AI Assistant channel matches ChatGPT and Gemini automatically when the visit carries a referrer[1]. Perplexity is absent from Google’s lists, so it lands in Referral, or in Direct if the referrer never arrives, unless you build a rule[3].
GA4 ships a default channel called AI Assistant. It sets a session’s medium to “ai-assistant” automatically whenever the referrer matches a list Google maintains internally[1]. According to the page that defines the channel, that list names ChatGPT, Gemini, Deepseek, Copilot and Grok.
Google’s own release notes for the same feature tell a slightly different story. The entry announcing the AI Assistant channel names ChatGPT, Gemini and Claude instead, and does not mention Deepseek, Copilot or Grok[2]. Both pages agree on one thing. Perplexity is not on either list.
That is not a minor documentation typo. Two official Google pages describing the same feature give two different rosters, and the honest reading is that ChatGPT and Gemini are the only names confirmed twice. Claude, Deepseek, Copilot and Grok are probably covered. Perplexity, checked against both pages, is not.
| Assistant | Named on the channel-definition page | Named on the release notes | What that means today |
|---|---|---|---|
| ChatGPT | Yes | Yes | Matched automatically when a referrer is present |
| Gemini | Yes | Yes | Matched automatically when a referrer is present |
| Claude | No | Yes | Probably matched, unconfirmed on both pages |
| Deepseek, Copilot, Grok | Yes | No | Probably matched, unconfirmed on both pages |
| Perplexity | No | No | Needs a custom channel group, or it sits in the generic Referral bucket |
One more mechanical detail matters here. GA4’s channel matching only fires when a referrer exists in the first place. A visitor from an assistant that strips its referrer, which happens on some mobile in-app browsers, will not land in the AI Assistant channel no matter which list it is on. It lands in Direct regardless of the name. Google’s own AI Mode and AI Overviews sit outside this channel entirely, and what GA4 actually shows for Google AI Mode traces where that traffic reports instead.
Setting Up the Split
Splitting out an assistant GA4 already recognizes costs nothing. Splitting out Perplexity, or any assistant Google has not added yet, takes one custom channel group matching the assistant’s source pattern, ranked above the default Referrals rule so it wins the match[3].
That configuration deserves its own dedicated walkthrough, and it has one, with rule ordering and the exact fields GA4 exposes. What matters here is the shape of the decision, not the click path. The fuller setup walkthrough already covers the underlying custom-channel mechanics, if you want to build the rule today rather than wait.
Two things are worth knowing before you build it. Custom channel groups match on the Source field GA4 already derived, not on the raw HTTP referrer header, so the rule has to target the domain string GA4 recorded rather than the header itself[3]. And ordering matters. Place the new rule above Referrals, or GA4 files a matching session under the generic Referral channel first and never reaches your rule.
None of this retroactively fixes sessions that already happened. GA4 attributes on a last-non-direct-click basis, and excluding or adding a source going forward does not reclassify a session that already landed somewhere else[4]. Build the rule for what happens next, not for a clean history.
Why the Numbers Will Look Small and Why That Is Not the Point
In June 2025, AI assistants sent 1.13 billion referral visits to the top 1,000 sites worldwide, up 357% from June 2024[5]. Spread across one B2B SaaS marketing site, even a correctly configured split usually shows a small number. The volume was never the reason this channel matters.
A small number from this channel is not evidence the configuration failed. The assistant already did the work a search results page used to leave to the visitor: comparing vendors, narrowing the shortlist, forming an opinion. Someone arriving from that conversation is not browsing. They are checking whether what the assistant told them is true.
That changes what the visit is worth reading for. A landing page that gets three sessions a week from ChatGPT is not a rounding error if those three sessions are further along than three hundred sessions from a generic search term. The number stays small. The read on it should not.
This is also where it is easiest to overclaim, so the boundary is worth stating plainly. None of this says AI-referred visitors convert at a different rate than anyone else. Nobody has a defensible sample size for that claim yet, on this site or most others. What is defensible is the mechanism: they arrive later in the process, and reporting them as top-of-funnel traffic undercounts what they already know.
What to Report Upward
The honest sentence for a marketing meeting is short. Some of your AI-assisted demand is now visible in GA4, and some of it is still invisible, and treating either number as the whole picture would be a mistake in both directions.
Report what the AI Assistant channel shows for ChatGPT and Gemini as a real number, sourced from GA4’s own attribution logic. Report anything you have built a custom rule for, including Perplexity, the same way, and say plainly that it started on a specific date rather than implying it always existed.
Then say what you cannot report. Some share of your Direct traffic almost certainly includes assistant visits that arrived with no referrer at all, and GA4 has no field that recovers them after the fact[4]. Landing-page patterns and entrance behavior can suggest where that share is concentrated. They cannot turn it into a number you should put in a slide.
That is a smaller claim than most reporting decks make about this channel, and it holds up under a follow-up question, which the bigger version usually does not. It connects to the Analytics Trust Gap running through this whole pillar: the number that is wrong by omission is more dangerous than the number that is simply small, because a small number gets questioned and a confidently wrong one does not.
If your reporting already leans on GA4’s default assumptions elsewhere, the same caution applies more broadly than just this channel. The assumptions GA4 quietly makes by default cover the rest of that ground.
Sources
- Google Analytics Help, Default Channel Group Definitions – Checked 2026-08-18; defines the AI Assistant channel as traffic from sources like ChatGPT, Gemini, Deepseek, Copilot or Grok, excluding Google’s own AI Overviews and AI Mode; medium is set to “ai-assistant” automatically when the referrer matches Google’s internal list ↩
- Google Analytics Help, Release Notes – The May 13, 2026 entry announcing the AI Assistant channel names ChatGPT, Gemini and Claude as covered chatbots, a different roster than the channel-definition page; does not mention Deepseek, Copilot or Grok ↩
- Google Analytics Help, Custom Channel Groups – Checked 2026-08-18; ships a worked “Example: AI assistants” walkthrough with a source-matching regular expression that explicitly names Perplexity alongside ChatGPT, Gemini and Copilot; instructs matching on the Source field and ranking the new rule above Referrals ↩
- Google Analytics Help, Referral Exclusion List – Confirms GA4 attributes on a last-non-direct-click basis; excluding or adding a source going forward does not retroactively reclassify a session that already landed under a different channel ↩
- Similarweb (via TechCrunch), AI referrals to top websites up 357% year over year – Similarweb clickstream data, top 1,000 domains, June 2024 vs June 2025; 1.13 billion AI-referral visits recorded in June 2025 alone ↩
Seeing these patterns at your company?
Book a free WebOps Diagnostic. I'll review your site before the call and share specific observations.
Book a Free Call →Frequently Asked Questions
Partly. GA4's native AI Assistant channel matches ChatGPT and Gemini automatically whenever the visit carries a referrer, confirmed on Google's own channel-definition page as of August 18, 2026. Perplexity does not appear on either of Google's official lists, so it sits in the generic Referral bucket until you build a custom channel group for it.
Because Perplexity is not on Google's own list of recognized AI assistants, on either of the two official pages that describe the feature. A custom channel group matching Perplexity's source pattern, ranked above the default Referrals rule, is the only way to pull those sessions out of the generic Referral bucket today.
No. Google's channel-definition page names ChatGPT, Gemini, Deepseek, Copilot and Grok. Its release notes for the same feature instead name ChatGPT, Gemini and Claude, dated May 13, 2026. Both pages agree only that ChatGPT and Gemini are covered and that Perplexity is not, as of August 18, 2026.
Less than the conversation about it would suggest. In June 2025 alone, AI assistants sent 1.13 billion referral visits to the top 1,000 sites worldwide, up 357% from June 2024, aggregated across the largest properties on the internet. On a single B2B SaaS marketing site, expect a small but growing slice, not a headline number.
