How to Track AI Traffic in GA4 (and Whether It Converts)

To track AI traffic in GA4, open Reports → Acquisition → Traffic acquisition and look for the AI Assistant channel. No channel there yet? Build a custom channel group with a Source matches regex rule for ChatGPT, Perplexity, Gemini, Claude, and Copilot, and you'll have AI referral traffic as one named segment. Then you can ask the question that actually matters: are those visitors buying?

AI assistants send real buyers to websites now. Someone asks ChatGPT for the best waterproof hiking boots, gets three names, clicks one, and lands on your product page already half-sold. For most of 2025 and into 2026, GA4 quietly dropped that visit into Referral or Direct, where nobody goes looking. So a channel that grows fast and tends to buy well ends up nearly invisible in a stock GA4 setup. Below: how to see it, why part of it stays hidden whatever you do, and how to find out if it converts. There's a one-prompt shortcut at the end if you'd rather not build reports by hand.

Where AI traffic hides in GA4

Start with the mental model. An AI-referred visit ends up in one of two places.

Plenty of it lands in Direct, because the assistant dropped the referrer somewhere along the way. In-app browsers do that. So do the ChatGPT desktop and mobile apps, and so does any link a user copies and pastes. If your Direct numbers have looked suspiciously healthy lately, this is part of why.

The rest lands in Referral or Organic, where the referrer survived and you can read the host: chatgpt.com, perplexity.ai, gemini.google.com, claude.ai, copilot.microsoft.com.

What you want is a single named segment for the visits GA4 can attribute, so you can hold its conversion rate up against every other channel. The stripped-referrer visits stuck in Direct are a harder problem, and I'll come back to why no GA4 trick fully solves them.

Why it deserves a real workflow

This is worth more than a weekly squint at the Referral report, for two reasons that pull the same way.

The volume keeps climbing. A year ago ChatGPT and friends were a rounding error in most acquisition reports; now they're a visible slice, and the AI browsers (Perplexity Comet, ChatGPT Atlas) are adding traffic GA4 records in yet another way again.

And these visitors arrive further along. Through 2026 a steady run of vendor and agency studies has reported AI referrals converting better than plain organic search, sometimes much better. I wouldn't bank on the headline multipliers you'll see quoted, a "3x" in one writeup or some per-platform percentage in the next; those samples are small, self-selected, and defined however suited the author. The part that survives scrutiny is the direction. A person who shows up after an assistant has already answered their question and named you is closer to buying than someone who typed a keyword into Google. Worth knowing whether that's happening on your site.

The good news: GA4 finally has an AI Assistant channel

For most of this channel's short life you had to assemble it yourself. That's no longer the default.

In 2026 Google started shipping a built-in channel group called AI Assistant that pulls generative-AI traffic (ChatGPT, Gemini, Copilot, Perplexity, Claude, and more) out of the Referral bucket on its own. It keys off the referrer: when an incoming host matches Google's list, GA4 stamps the session's medium as ai-assistant and files it under the new channel. Same referred visits a hand-built regex would catch, minus the upkeep.

Two things to know before you lean on it. The rollout has been staggered across properties with no published finish line, and Google hasn't shared the full list of hosts it recognises, so "check whether you have it" beats "assume you do." To check, go to Reports → Acquisition → Traffic acquisition, switch the dimension to Session default channel group, and see whether AI Assistant is in the list. If it is, jump to the conversion section. If it isn't, the custom group below has you covered.

I build the custom group regardless. It doesn't wait on a rollout, it matches exactly the hosts I choose, and it backfills history, which the default channel won't.

Build a custom AI channel group (every property, today)

Five minutes, and it doesn't care whether the default channel has reached you.

1. Open Admin → Data display → Channel groups.
2. Hit Create new channel group and name it something like "AI Traffic."
3. Add a channel called "AI Assistant" and drag it up the list, so it grabs sessions ahead of Referral and Organic.
4. Set the rule to Source matches regex and paste:

chatgpt\.com|chat\.openai\.com|perplexity\.ai|gemini\.google\.com|claude\.ai|copilot\.microsoft\.com|you\.com|poe\.com|phind\.com|meta\.ai

5. Save. GA4 applies custom channel groups to past data, so your history re-buckets too. Set it up early anyway, so the segment reads the same in every report.

A warning that cost me an afternoon the first time round: GA4's Source dimension only ever holds a hostname, never a path. So the bing.com/chat entry you'll find in older guides matches nothing, because the /chat part never shows up in a Source value. Copilot lands under copilot.microsoft.com (and occasionally bare bing.com), both already in the pattern. The list above is the big five plus You.com, Poe, Phind, and Meta AI; drop whatever's noise for you, and expect to add a name every few months as new tools appear.

Default channel vs. a custom group

Both share the weakness in that bottom row, which is where the next section picks up.

The catch: a lot of AI traffic never leaves Direct

Most "track AI traffic in GA4" posts skip this part, and it's the part that will burn you once you start moving budget around based on the numbers.

GA4 sorts a visit by whatever referrer the browser provides. A lot of AI clicks provide none: a link tapped inside the ChatGPT app, a URL copied out of a chat into a fresh tab, a wrapper that strips the referrer in transit. With nothing to work from, GA4 calls it Direct. The free ChatGPT tier has a reputation for not passing a referrer reliably, which alone sends a slice of the biggest AI source straight onto the Direct pile.

How big a slice? Genuinely unknown. There's no trustworthy public figure, and anyone who hands you a precise percentage is guessing. So read your AI channel as a floor. The true number is higher, you just can't size the gap from inside GA4, and no amount of channel-group cleverness changes that. A referrer that never arrived can't be reconstructed after the fact.

Triangulation is the realistic move. Genuine direct traffic mostly hits the homepage and branded URLs, so when unattributed Direct lands cold on /products/whatever, something sent it. Drop a "How did you hear about us?" box at signup or checkout and watch the ChatGPT answers stack up. Tail your server logs for the crawler agents (GPTBot, PerplexityBot, ClaudeBot), since the engines reading you this month tend to send the humans next month. Each signal is fuzzy alone. Stacked together they tell you whether Direct is hiding a real AI channel.

So does it convert?

Once the traffic is visible, the useful work starts: working out whether it earns its keep. Conversion decides that, not raw session volume.

With the AI segment in place, put a conversion metric beside it: switch the Traffic acquisition table to your AI channel and add the key-event and revenue columns. A few cuts are worth the time.

Set AI against your other channels first, session conversion rate and revenue per session next to Organic, Direct, and Paid. That one comparison settles whether the "AI converts better" chatter holds on your site, which is the only verdict that pays your bills.

Break the channel down by source next, because ChatGPT, Perplexity, and Gemini don't behave alike: different intent, different pages cited, different outcomes. An average across them can bury one source that floods you with sessions and almost no orders.

Then follow it down the funnel instead of stopping at the headline rate. Pull the AI funnel on its own, landing page through add-to-cart or signup through checkout, and watch where it sheds people. These visitors often arrive on a deep page the funnel was never built around, and the leak point tells you what to fix. For product catalogs, the GA4 ecommerce reports with AI workflow slots in right here.

Converting above your average? Go earn more of it (the GEO angle is below) and tidy up the funnel. Converting below, despite all the hype? You've found a concrete, fixable problem, usually a landing page or funnel that doesn't fit how these visitors show up, and the conversion rate drop diagnostic works on the segment as it stands.

The shortcut: ask instead of build

Everything so far is clicking around the GA4 UI. It works, but it's tedious enough that most teams do it once and never again.

The 2026 shortcut is to wire GA4 to an assistant over MCP (Model Context Protocol) and just ask. With a GA4 MCP server attached to Claude or ChatGPT, the cuts above collapse into single questions:

• "Compare conversion rate and revenue per session for my AI Assistant channel versus organic search and direct over the last 30 days."
• "Split my AI-assistant traffic by source. Which one sends the most sessions, and which one actually converts?"
• "Pull the funnel for visitors arriving from AI assistants. Which step loses the most, and how does it compare to my site-wide funnel?"
• "Has my AI Assistant channel grown over the last 90 days, and is its conversion rate trending up or down?"

Either assistant handles them; Claude vs ChatGPT for analytics gets into where each one shines, and the GA4 prompts library has more. Keep one thing in mind: the server only reads GA4, so it knows exactly what GA4 knows, Direct blind spot included. What it kills is the busywork, turning a half-hour of clicking into a question you'll actually bother to ask each week.

That's what we built ConvRadar to do (and yes, this site is ours). It's a hosted GA4 MCP server you set up in the browser, no terminal or Python, unlike Google's official server. It hands Claude and ChatGPT 22 conversion-diagnostic tools: traffic-quality scoring, funnel-drop diagnosis, anomaly detection, and a hypothesis library that turns a finding into an A/B test you can actually run. Aim it at the AI channel and "is this traffic any good, and where does it drop?" is one message, not an afternoon. Setup takes about five minutes in Claude or ChatGPT, and it's free during the open beta, email signup, no card. The one thing it can't tell you is whether ChatGPT recommends you in the first place. That's visibility, a different job.

Visibility and outcome aren't the same measurement

People keep folding two separate questions into "AI traffic," and that's where the confusion starts.

One question is visibility: do the engines mention and recommend you at all? You answer it with GEO signals, share of model (how often you turn up in answers versus rivals), citation rate, crawler hits in your logs. GA4 is the wrong tool here. By the time it logs a session, the recommendation already happened somewhere it can't see, which is why this calls for purpose-built GEO tooling that queries the models directly.

The other question is outcome: when visibility does send someone over, do they buy? That's pure GA4-plus-conversion, the analysis from a few sections up. Visibility tools prove GEO is working as publicity; the conversion view proves it's working as revenue. Keep them apart. A visibility vendor shouldn't get to claim conversions it can't observe, and your analytics shouldn't be graded on your AI ranking.

A five-minute weekly habit

It fits in a few minutes a week, the same rhythm as a standing CRO loop:

1. Open the AI channel and note the week-over-week change in sessions.
2. Compare its conversion rate and revenue per session to your site average; anything past roughly 15% is worth a look.
3. Eyeball Direct. Climbing on deep landing pages? That's AI leaking through.
4. Converting well, double down on whatever's getting you cited. Converting badly, pull the funnel and patch the leak.

Hand all four to an assistant in a single message and it takes longer to read this than to run it.

FAQ

How do I track AI traffic in GA4? Both ways, ideally. First see whether your property already has the default AI Assistant channel (Reports → Acquisition → Traffic acquisition, in the channel dimension); Google has been rolling it out through 2026 and it buckets known AI sources for you. Then add a custom channel group (Admin → Data display → Channel groups) with a Source-matches-regex rule for chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, copilot.microsoft.com and friends, ordered above Referral. The custom group runs on any property today and backfills history.

Why is my AI traffic showing up as Direct in GA4? Missing referrers. The ChatGPT app, copy-pasted links, and the free ChatGPT tier often hand GA4 no referrer, so it defaults the session to Direct. How much ends up there is anyone's guess, but enough that your labeled AI channel undercounts the real thing. No channel group recovers it, since the referrer is already gone by the time GA4 records the hit. Watch for unattributed Direct on deep internal pages, and add a "how did you hear about us?" field to corroborate.

Does AI traffic convert better than organic search? Usually, going by the 2026 studies, and sometimes by a wide margin. The exact multipliers are shakier: they come from small, self-selected datasets measured inconsistently, so don't build a plan on any single figure. What holds up is the intent, since AI referrals tend to arrive already answered and qualified. Measure your own and compare your AI channel's conversion rate to your site average directly. That's the number that counts.

What is the GA4 AI Assistant channel group? Google's built-in channel, rolling out across 2026, that separates generative-AI traffic (ChatGPT, Gemini, Copilot, Perplexity, Claude, and others) from generic Referral. It stamps qualifying sessions with the medium ai-assistant and surfaces them in the standard Acquisition reports with nothing to configure. The caveats: the rollout is staggered, the recognised-host list isn't public, and it only catches sessions carrying a referrer GA4 knows.

How do I see conversions from ChatGPT or Perplexity specifically? Isolate AI traffic with the default channel or a custom group, then add key-event and revenue metrics to the Traffic acquisition table for that channel. Break it out by Source to tell ChatGPT, Perplexity, and Gemini apart, since they convert at different rates. For funnel detail, build an Exploration with the AI channel as a dimension, or just connect GA4 to an assistant and ask in plain English.

Can I measure GEO (generative engine optimization) results in GA4? Half of it. GA4 can show the outcome of GEO, whether AI referrals convert, but not the visibility, whether engines cite and recommend you. Share of model, citation rate, and answer presence need dedicated GEO tooling. For turning that visibility into customers, GA4 plus a conversion view is the right call. Run both, and don't ask either to do the other's job.

Do I need code to analyze AI traffic in GA4? No. The channel groups are point-and-click. And once GA4 is connected to Claude or ChatGPT through a hosted MCP server, the analysis is a question you type, with no SQL, no CSV exports, and no Explorations to assemble. The GA4 MCP server overview lays out the no-code and self-hosted routes.

See whether your AI traffic is worth chasing

The default channel tells you how much AI traffic turns up. Whether it converts is the part you actually care about, and with GA4 wired to your assistant that's a single question away. Point it at the AI channel and ask: "is my AI Assistant traffic converting better or worse than the rest, and where does its funnel leak?" The answer tells you whether to chase this channel or repair it.

Connect your GA4 →