AI Traffic Attribution

GA4 uses source and medium dimensions to describe where a session came from when that information is observable. Evidence for this claim GA4 session source and medium dimensions describe where a new session originated, using referrer and campaign information when available. Scope: Google Analytics acquisition dimensions; missing referrers and cross-device behavior limit attribution. Confidence: high · Verified: 2026-07-14 Google Analytics: Traffic-source dimensions Campaign parameters can make links you control easier to attribute. Evidence for this claim Google Analytics supports manually collected campaign parameters such as utm_source and utm_medium for traffic attribution. Scope: Tagged inbound links; publishers cannot add parameters to links controlled by external AI platforms. Confidence: high · Verified: 2026-07-14 Google Analytics: URL builders and campaign parameters

TL;DR — AI tools like ChatGPT and Perplexity send you traffic, but a lot of it shows up in Google Analytics as Direct — as if people typed your URL in by hand — instead of as AI traffic. That’s because many AI tools don’t tell your site where the visitor came from. So the AI traffic you can see is the minimum, not the real total.

What AI traffic attribution means

When someone clicks a link, their browser usually tells the destination site where they came from. That little hand-off is called the referrer, and it’s how analytics tools like GA4 know to label a visit as “came from Google” or “came from Facebook.”

AI traffic attributionAI traffic attribution is the practice of correctly identifying and measuring website visits that come from AI tools — ChatGPT, Perplexity, Gemini, Claude, AI Overviews, and AI browsers. It's hard because many of those tools strip the referrer header, so the visits land in your analytics as Direct traffic with no source. is just the job of correctly counting visits that came from AI tools — ChatGPT, Perplexity, Gemini, Claude, Copilot, and Google’s AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index.. The trouble is that a lot of those tools don’t pass the referrer along. When the referrer is missing, GA4 has nothing to go on, so it dumps the visit into Direct traffic — the same bucket as someone typing your address straight into the browser.

Why so much AI traffic hides

A few things cause it:

• Mobile apps. When you tap a link inside the ChatGPT or Perplexity phone app, the app opens it in a way that drops the referrer entirely. There’s no fix for this on your end.
• “noreferrer” links. Some AI tools deliberately add a tag to their links that strips the source.
• Copy-paste. If someone copies a URL out of an AI answer and pastes it into their browser instead of clicking, there’s no referrer at all.

The result: published studies estimate that somewhere between 35% and 70% of all AI-driven visits land in Direct instead of being correctly labeled as AI.

Why it matters even if the number looks tiny

It’s tempting to look at “AI is 0.2% of my traffic” and ignore it. Don’t. Two reasons:

1. The real number is bigger than what you see, because of all that hidden traffic.
2. AI traffic converts really well. For Ahrefs, AI searchAI search uses large language models and retrieval-augmented generation (RAG) to synthesize an answer from multiple sources rather than returning a ranked list of links. Examples include Google AI Overviews, ChatGPT Search, and Perplexity. visitors were only 0.5% of traffic but produced 12.1% of signups — a 23x higher conversion rate than regular Google traffic. A small slice can punch way above its weight.

What you can actually do

• Set up a custom “AI” channel in GA4 so the trackable AI visits get labeled correctly (the Advanced tab has the exact recipe).
• Keep an eye on your Direct traffic — if it’s growing and converting better than it used to, a chunk of that is probably hidden AI traffic.
• Accept that Google’s own AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. can’t be cleanly separated from normal search — nobody has solved that one yet.

Want the full picture — GA4’s new AI Assistant channel, the platform-by-platform referrer breakdown, the exact regex, and how to estimate your real AI traffic? Switch to the Advanced tab.

Session source and medium reflect the acquisition metadata GA4 receives, so they cannot recover a referrer that was never passed. Evidence for this claim GA4 session source and medium dimensions describe where a new session originated, using referrer and campaign information when available. Scope: Google Analytics acquisition dimensions; missing referrers and cross-device behavior limit attribution. Confidence: high · Verified: 2026-07-14 Google Analytics: Traffic-source dimensions Manual UTM parameters are useful only on inbound links whose URLs you can control. Evidence for this claim Google Analytics supports manually collected campaign parameters such as utm_source and utm_medium for traffic attribution. Scope: Tagged inbound links; publishers cannot add parameters to links controlled by external AI platforms. Confidence: high · Verified: 2026-07-14 Google Analytics: URL builders and campaign parameters

TL;DR — AI traffic attributionAI traffic attribution is the practice of correctly identifying and measuring website visits that come from AI tools — ChatGPT, Perplexity, Gemini, Claude, AI Overviews, and AI browsers. It's hard because many of those tools strip the referrer header, so the visits land in your analytics as Direct traffic with no source. is hard because AI tools frequently strip the referrer header, so 35–70% of AI-sourced sessions land in GA4 as Direct. GA4’s native AI Assistant channel (added May 13, 2026) and a custom channel group reclassify the referred portion — but neither sees the dark sessions, the AI Assistant channel doesn’t apply retroactively and eats one of your two custom channel slots, and Google AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. / AI Mode clicks are permanently merged into regular Organic Search in both GA4 and GSCA free Google service that reports how a site performs in Google Search and surfaces problems with how Google crawls, indexes, and serves it. It's first-party data straight from Google — but you don't need it to appear in results.. Referrer behavior varies wildly by platform (Claude and Mistral pass cleanly; Grok standalone, the Copilot Windows app, and the mobile apps don’t). Treat the visible number as the floor, triangulate with your Direct baseline and branded search, and weight it by conversion quality — AI traffic converts far above average.

Why AI traffic is harder to attribute than anything else

Analytics knows where a visit came from because the browser sends a Referer header on the way in. GA4’s whole channel system leans on it: a referrer of perplexity.ai becomes a Referral; no referrer becomes Direct ((direct) / (none)). Every other source you care about — organic search, social, email, referral — plays along with this contract.

AI tools frequently don’t. And there’s no UTM workaround like you’d use for a campaign, because you don’t control the URLs AI tools cite — they link to the canonical URLsHow search engines pick one canonical URL among duplicates and consolidate signals onto it. they foundA 302 (\"Found\") is a temporary redirect: it forwards users to a new URL while telling search engines the original URL should stay in the index. It's a weak canonicalization signal, not the zero-equity dead end of SEO folklore. while crawlingCrawling is how search engines use automated bots (like Googlebot and Bingbot) to discover URLs and download pages. A page has to be crawlable to be indexed, but crawling on its own isn't a ranking factor., without your tracking parameters. This is the part people miss: you can tag your own links, but you can’t tag the link ChatGPT decided to show.

Be precise about what’s actually observable. Analytics attribution is built from a narrow set of signals — the referrer header, any campaign (UTM) parameters on the inbound URL, redirect chainsA → B → C instead of A → C. Each hop loses link equity and adds latency. the request passed through, and the landing URL itself — captured by whatever your analytics tag sees at collection time. If a signal was never transmitted (referrer stripped, no UTM present, a redirectA redirect sends browsers and crawlers from a requested URL to a different one. An HTTP redirect specifically is a 3xx status code paired with a Location header; meta refresh and JavaScript redirects achieve a similar navigation without being a 3xx response themselves. Permanent redirects (301/308) are Google's signal the target should be canonical; temporary ones (302/303/307) aren't. that drops query parameters), no amount of analytics configuration recovers it after the fact; a missing signal is a gap in the record, not a value analytics can infer.

That also sets the real boundary on UTM parameters: they only help on URLs whose querystring survives to your landing page. You can add utm_source to links you control — your own social posts, email, ads — and GA4 will read them as long as a redirect in the chain doesn’t strip the parameters and no other campaign source takes precedence in GA4’s channel-matching order. You cannot add UTMs to a URL an AI tool cites on its own, because that URL was chosen and rendered by the AI product, not by you. ChatGPT Search’s automatic utm_source=chatgpt.com is the exception precisely because ChatGPT — not you — is the one appending it.

I wrote about this in AI Assistants Are Breaking Web Analytics and Hurting Their Future: “Search is moving from traditional search engines to AI assistants, but traffic from many of these sites isn’t being tracked properly in analytics.” There’s a real social-contract angle here too — “there’s a social contract between search engines and website owners, where search engines add value by sending referral traffic to websites.” AI tools crawl heavily and, too often, send back traffic you can’t even see.

The dark traffic problem

“Dark traffic” is AI-sourced traffic that arrives with no referrer and so gets filed as Direct. Published estimates of how much AI traffic this swallows range from 35% to 70%. The most-cited single figure comes from Loamly’s analysis of 446,405 visits: 70.6% of verifiable AI traffic landed as Direct in GA4 by default. Treat that number as a vendor-sourced upper bound, not gospel — but the direction is not in dispute.

Important nuance, because it’s easy to overstate: dark traffic is not simply “AI traffic = Direct.” Plenty of AI-referred traffic does show up correctly in Referral or the new AI Assistant channel. The problem is the slice that arrives with no referrer at all — and that slice is large enough to make your visible AI number a serious undercount.

Four mechanisms create it:

1. Referrer policy. Many tools send referrer-policy: strict-origin-when-cross-origin or stricter, passing only the origin (or nothing) cross-origin.
2. rel="noreferrer" on links. ChatGPT (especially paid tiers) and others add noreferrer directly to outbound links, explicitly stripping the source.
3. Mobile app architecture. iOS WKWebView and Android Custom Tabs — used by the ChatGPT and Perplexity apps — drop the Referer header when opening external links. There is no analytics-side workaround.
4. Copy-paste. Per RFC 9110, a URL pasted into the address bar carries no referrer. Copying a link out of an AI answer instead of clicking it registers as Direct.

Those four mechanisms are the AI-specific ones, but they’re not the only way a session lands in Direct — bookmarks, privacy-focused browsers, corporate proxies, and plenty of non-AI apps strip referrers too. A missing referrer is genuinely ambiguous: it can’t distinguish an AI click from any of those other sources, and there is no analytics-side method that recovers which one actually happened. Treat “Direct” as “source unknown,” not as a bucket you can silently relabel “AI” because AI usage happens to be rising.

Which AI tools pass referrers (and which don’t)

I tested each platform with document.referrer in Chrome DevTools (Ahrefs blog, May 2025). The behavior is all over the map, which is exactly why a one-size regex doesn’t capture reality:

The fragility is the real lesson. Copilot traffic “disappeared” from Ahrefs analytics for several months — a single code change on one platform can globally break tracking with no notice. As I put it in the analytics piece: “Websites have control over what info they send. They can send the full path, just the origin, or nothing — it’s up to them. We report whatever referrer we’re told to report. If they don’t send us one, then it would go in the ‘Direct’ bucket.”

How GA4 classifies AI traffic

The native AI Assistant channel (May 13, 2026)

GA4 added a default AI Assistant channel. Per Google’s docs, it covers users who “arrive at your site from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok. It excludes Google’s AI Overviews and AI Mode.” It auto-assigns medium = ai-assistant, channel = AI Assistant.

That last sentence is the whole catch. AI Overviews and AI Mode are not in the AI Assistant channel — they’re Organic Search. Google’s own definition of Organic Search now reads: users “arrive via non-ad links in organic-search results, including Google’s AI Overviews and AI Mode.” So a click from an AI Overview, when it passes a referrer at all, shows up as google / organic, indistinguishable from a regular blue-link click.

Three more limits worth being honest about:

• It only captures referred sessions. The AI Assistant channel does nothing for dark traffic — sessions with no referrer still land in Direct. Setting it up does not “solve” tracking.
• It is not retroactive. It applies going forward only. Pre-May-2026 sessions stay classified as Referral or Direct, which dents year-over-year comparisons.
• At launch it auto-recognized a limited set (ChatGPT, Gemini, Claude, and a few others), so a custom channel group still earns its keep for fuller coverage.

Before May 2026

Historically, AI traffic only ever appeared as Referral (with a referrer: chatgpt.com / referral) or Direct (no referrer). If you’re looking at older data, there is no AI Assistant bucket to find.

Custom channel groups

You can still build a custom channel group to catch more platforms and to re-segment historical data. Admin > Channel groups > Create new channel group. The two constraints to know: standard properties get only 2 custom channel groups (360 gets 5), and the AI group will eat one of them. The scripts tab has the exact regex and the one ordering rule that makes or breaks it.

Why AI traffic totals differ across analytics tools

If you run GA4 alongside Matomo, Adobe Analytics, or Ahrefs Web Analytics, don’t expect the same AI-traffic number from each. They aren’t measuring with the same ruler:

• What each tool collects. GA4’s traffic-source dimensions, Matomo’s campaign parameters, and Adobe’s referrer dimension aren’t identical fields — each product defines its own referrer collection, persistence window, and campaign-parameter handling.
• Sessionization differs. How long a session lasts, whether a returning visit restarts source attribution, and how cross-device visits are stitched together vary by product and by your own configuration.
• Channel rules differ. GA4’s channel-matching order (the AI-above-Referral rule from the setup steps) is a GA4-specific mechanic; other tools apply their own rule order and default channel definitions.
• Consent and collection scope differ. Consent-mode defaults, IP handling, and what a tool collects when consent is declined all change how much traffic even reaches the attribution layer in the first place.

None of this means one tool is “more accurate” than another — it means totals from different platforms aren’t directly comparable unless you’ve confirmed they define source, session, and channel the same way. If you’re reconciling AI-traffic numbers across tools for a report, say explicitly which tool and configuration produced each number.

The Google AI Overviews gap

This is the part no one has solved, and where I have some standing — I’m the person who reported the AI Mode noreferrer bug and got a response from John Mueller.

At launch, AI Mode links were marked noreferrer in the page code, so clicks passed no referrer and weren’t recorded in Search ConsoleGoogle's free tool for monitoring crawling, indexing, and search performance.. As I wrote on May 21, 2025 in Google Made It So You Can’t Track Clicks From AI Mode: “clicks to links in AI mode are not being recorded at all in Google Search ConsoleA free Google service that reports how a site performs in Google Search and surfaces problems with how Google crawls, indexes, and serves it. It's first-party data straight from Google — but you don't need it to appear in results., and they’re not sending a referrer value when you click.” The noreferrer half was fixed on May 31, 2025 — clicks now register as google / organic or google / search — but that “fix” just merges them into regular organic. You still can’t isolate them.

In June 2026 Google launched Gen AI Performance ReportsThe Google Search Console report that shows how your site actually performed in Google Search, built from real impressions and clicks. It reports four metrics — clicks, impressions, average CTR, and average position — and keeps the most recent 16 months of data. in Search Console covering AI Overviews, AI Mode, and generative features in Discover. The catch: impressions only — no clicks, no CTR, no position. An impression inside an AI Overview tells you nothing about whether anyone clicked through.

So both ends are blind. My honest read, from that AI Mode piece: “Attribution is important, and we don’t have it yet. It makes me suspicious that there’s something Google doesn’t want us to know about this traffic.” The partial workarounds — GTM triggers on #:~:text= fragments, conversational-query regex in GSC, watching for rising-impressions/flat-clicks divergence — each catch a sliver. None is end-to-end.

Interpreting the data: floor, not ceiling

The most useful framing I can give you: the AI traffic you see in GA4 is the minimum, not the total. Swydo put it as “GA4 data is the floor, not the ceiling.” Because of dark traffic, your real AI-influenced number is materially higher than the AI Assistant channel and your custom group combined will ever show.

That has a concrete diagnostic payoff. If your Direct channel’s conversion rate has crept up alongside AI’s rise, a chunk of that Direct traffic may be misclassified AI sessions. Loamly estimates dark AI traffic converts at 10.21% versus 2.46% for genuine Direct — a suggestive pattern worth investigating, not proof by itself, since a rising Direct conversion rate can also come from returning customers, bookmark traffic, or other non-AI sources that also land in Direct. To estimate total AI influence, triangulate: branded-search lift, shifts in your Direct baseline, and post-conversion “how did you hear about us?” surveys.

One more distinction worth being explicit about: an attributed conversion tells you GA4’s configured channel rules assigned credit for that conversion to a given source — it does not prove that source caused the conversion, and it says nothing about the sessions and touchpoints GA4 never saw. Session source and medium reflect whatever acquisition metadata GA4 received, not a verified causal chain; treat attribution numbers as “what got credited,” and treat any conversion-lift claim (yours or a vendor’s) with the same caution you’d apply to any other correlational read.

And weight all of it by quality, because AI traffic is unusually valuable. For Ahrefs, 0.5% of traffic from AI searchAI search uses large language models and retrieval-augmented generation (RAG) to synthesize an answer from multiple sources rather than returning a ranked list of links. Examples include Google AI Overviews, ChatGPT Search, and Perplexity. produced 12.1% of signups — a 23x higher conversion rate than traditional organic, with AI visitors viewing ~50% more pages. Per-platform vendor estimates point the same way (ChatGPT referral converting around 15.9%, Perplexity ~10.5%, versus Google organic ~1.76%), though those are vendor-sourced and should be cited with caution. The independently documented 23x figure is the one I’d anchor on.

AI traffic has also grown fast — roughly 9.7x over twelve months across ~82,000 sites, and a 527% jump January–May 2025 versus the same period a year earlier. The market is shifting under you, too: as of January 2026, ChatGPT held 64.5% of measurable AI referral traffic, with Claude at 18.5%, Gemini 10.6%, and Perplexity 7.3% — so a tracking setup that only watches ChatGPT and Perplexity is already behind.

What to actually do

1. Confirm whether you already have the native AI Assistant channel (post-May 2026).
2. Build a custom channel group for fuller coverage, with the AI channel ordered above Referral (see scripts).
3. Create a segment/exploration on AI traffic and watch its conversion rate, not just its volume.
4. Baseline your Direct traffic now, then watch for a rising conversion rate as a signal worth investigating — not proof — of hidden AI traffic.
5. Triangulate the invisible part with branded search and post-conversion surveys.
6. Accept that AI Overviews / AI Mode can’t be cleanly separated — don’t burn weeks trying to force it.

AI summary

A condensed take on the Advanced version:

• The problem: AI tools often strip the referrer header, so 35–70% of AI-sourced visits land in GA4 as Direct (Loamly: 70.6%, vendor-sourced). This is “dark traffic.”
• Not simply “AI = Direct”: some AI-referred traffic does show in Referral / the AI Assistant channel; the issue is the no-referrer slice.
• GA4 AI Assistant channel (May 13, 2026): captures referred sessions from ChatGPT, Gemini, Claude, etc. — but excludes AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index./AI Mode, isn’t retroactive, doesn’t see dark traffic, and eats one of your two custom channel slots.
• AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. / AI Mode: permanently classified as google / organic in GA4 and GSCA free Google service that reports how a site performs in Google Search and surfaces problems with how Google crawls, indexes, and serves it. It's first-party data straight from Google — but you don't need it to appear in results.. The June 2026 GSC Gen AI reports show impressions only, no clicks. Patrick reported the AI Mode noreferrer bug (fixed May 31, 2025 → now merges into organic).
• Referrer behavior varies by platform: Claude, Mistral, Perplexity web pass cleanly; Grok standalone, Copilot Windows app, and the mobile apps don’t. Copilot once vanished for months.
• Only a few signals are observable at all: referrer header, campaign (UTM) parameters, redirect chainsA → B → C instead of A → C. Each hop loses link equity and adds latency., and the landing URL. A signal that was never sent can’t be reconstructed after the fact, and a missing referrer alone doesn’t prove a visit came from AI — bookmarks, privacy browsers, and other non-AI sources also land in Direct.
• Tool totals aren’t directly comparable: GA4, Matomo, Adobe Analytics, and Ahrefs Web Analytics collect, sessionize, and channel-classify traffic differently, so their AI-traffic numbers for the same site can legitimately disagree.
• Floor, not ceiling: visible AI traffic is the minimum. A rising Direct conversion rate is worth investigating as a possible sign of hidden AI traffic (dark AI converts ~10.21% vs. ~2.46% for true Direct per Loamly) — but an attributed conversion reflects configured credit assignment, not proof of causation.
• Quality beats volume: for Ahrefs, 0.5% of traffic from AI = 12.1% of signups (23x conversion premium). AI traffic grew ~9.7x in a year.

Official documentation

Primary-source documentation on how the platforms classify and report AI traffic.

Google Analytics

• GA4 Default channel group — the rules for every channel, including the AI Assistant definition and the note that Organic Search “includes Google’s AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. and AI Mode.”
• Create, edit, and apply custom channel groups — how to build a custom AI channel group, the per-property limits, and channel ordering.
• GA4 What’s New — AI Assistant channel — the May 13, 2026 announcement of the native AI Assistant default channel.

Google Search

• Gen AI Performance Reports in Search Console — the June 3, 2026 launch covering AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index., AI Mode, and Discover generative features (impressions only — no click data).

Microsoft / Bing

• Bing Webmaster Tools — Help & How-To — Bing’s reporting surfaces; Copilot-in-Bing referrer behavior varies by feature (bing.com vs. direct).

Quotes from the source

On-the-record statements from the platforms and from my own first-party testing. Each link jumps to the source.

Google Analytics (What’s New, May 13, 2026)

• “Identify how users are discovering your site through chatbots like ChatGPT, Gemini, and Claude via a fresh ‘AI Assistant’ channel in Default Channel Group reports.” Source

Google — channel definitions

• AI Assistant covers users who “arrive at your site from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok. It excludes Google’s AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. and AI Mode.” Organic Search covers users who “arrive via non-ad links in organic-search results, including Google’s AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. and AI Mode.” Source

Google Search Central (June 3, 2026)

• The new Gen AI Performance ReportsThe Google Search Console report that shows how your site actually performed in Google Search, built from real impressions and clicks. It reports four metrics — clicks, impressions, average CTR, and average position — and keeps the most recent 16 months of data. cover “AI Overviews, AI Mode, and generative AI features in Discover” — impressions only, no click data. Source

Patrick Stox (Ahrefs) — on the broken contract

• “Search is moving from traditional search engines to AI assistants, but traffic from many of these sites isn’t being tracked properly in analytics.” Source
• “Websites have control over what info they send. They can send the full path, just the origin, or nothing — it’s up to them. We report whatever referrer we’re told to report. If they don’t send us one, then it would go in the ‘Direct’ bucket.” Source

Patrick Stox (Ahrefs) — on AI Mode

• “The tl;dr is that clicks to links in AI mode are not being recorded at all in Google Search ConsoleA free Google service that reports how a site performs in Google Search and surfaces problems with how Google crawls, indexes, and serves it. It's first-party data straight from Google — but you don't need it to appear in results., and they’re not sending a referrer value when you click.” Source
• “Attribution is important, and we don’t have it yet. It makes me suspicious that there’s something Google doesn’t want us to know about this traffic.” Source

AI traffic tracking setup checklist

A step-by-step pass to get AI traffic measured as well as it currently can be:

• Baseline your Direct traffic first. Record current Direct volume and conversion rate before you change anything — that’s your before/after for the dark-traffic fingerprint.
• Check for the native AI Assistant channel. In GA4 reports (post-May 2026), confirm whether the default channel group already shows an “AI Assistant” channel. If it does, referred AI sessions from ChatGPT/Gemini/Claude/etc. are already being labeled.
• Build a custom AI channel group for fuller coverage and historical re-segmentation: Admin > Channel groups > Create new channel group. (You get only 2 custom groups on a standard property — budget accordingly.)
• Add the AI channel with the source regex (see Scripts) matching chatgpt.com, perplexity.ai, claude.ai, gemini.google.com, and the rest.
• Order the AI channel ABOVE Referral. GA4 matches on first hit; if Referral catches an AI source first, it never reaches your AI channel.
• Create an AI traffic segment in an Exploration so you can analyze behavior (pages/session, bounce, conversions) separately.
• Monitor conversion rates, not just volume. Compare AI’s conversion rate to organic and watch the Direct channel’s conversion rate for the dark-AI lift.
• Triangulate the invisible part. Track branded-search trends, your Direct baseline shift, and add a post-conversion “how did you hear about us?” survey.
• Accept the AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. gap. Don’t expect to separate AI Overview / AI Mode clicks from google / organic — note it as a known blind spot in your reporting.

The GA4 custom channel group regex

This is the core of a working AI channel. In GA4: Admin > Channel groups > Create new channel group, add a new channel named AI, and set the condition to Source matches regex with the pattern below.

Minimal, high-signal pattern (the platforms that actually send traffic):

(chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com)

Fuller coverage (add the long tail as they grow):

chatgpt\.com|chat\.openai\.com|perplexity\.ai|claude\.ai|gemini\.google\.com|copilot\.microsoft\.com|deepseek\.com|grok\.x\.com|grok\.com|x\.ai|meta\.ai|chat\.mistral\.ai|openai\.com|you\.com|phind\.com|poe\.com|pi\.ai

How to add it in GA4 (and the one rule that matters)

1. Admin > Channel groups > Create new channel group. Name it something like “AI + Default.”
2. Click Add new channel, name the channel AI.
3. Set the condition: Source → matches regex → paste a pattern above. (Optional: also add a Manual term match for utm_source containing chatgpt to catch ChatGPT Search’s utm_source=chatgpt.com.)
4. Drag the AI channel ABOVE the Referral channel in the channel order. This is the step everyone gets wrong. GA4 evaluates channel rules top-down and stops at the first match — chatgpt.com is a valid referral, so if Referral sits above your AI channel, every AI session gets caught as Referral and never reaches your AI channel. Order is everything.
5. Save. Note the limit: standard properties allow only 2 custom channel groups (GA 360 allows 5).

What this does and doesn’t do: it correctly labels AI sessions that arrive with a referrer. It does nothing for dark sessions (no referrer → still Direct), and it cannot separate Google AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. / AI Mode from google / organic. There is no regex that fixes those — they’re upstream of your analytics.

How should this AI-influenced visit be classified?

Example: visible AI referral

An illustrative session arrives with a recognized assistant hostname in the referrer and lands on a cited article. The defensible label is observable AI referral. The landing page and conversion can be analyzed, but the session does not reveal how many unclicked mentions or stripped-referrer visits occurred.

Example: direct visit after an AI answer

A user reads an AI answer, copies the brand name, and later types the site address. Analytics records Direct because no referrer survives. A post-conversion survey may reveal AI influence, but the session should not be silently rewritten as an AI referral in the deterministic channel report.

Example: Google AI feature click

A click from a Google AI feature can remain inside ordinary Organic Search reporting. Report the measurable organic session as organic and explain the product-level limitation; do not invent a separate AI source from data that the reporting surface does not expose.

Test yourself: AI traffic attribution

Resources worth your time

My writing on AI analytics and traffic

• AI Assistants Are Breaking Web Analytics and Hurting Their Future — the per-platform document.referrer testing and the social-contract argument.
• Google Made It So You Can’t Track Clicks From AI Mode (Partially Fixed) — the noreferrer bug I reported, and what the partial fix changed.
• Does AI Search Traffic Convert Better Than Traditional Search? — the 23x conversion benchmark (0.5% of traffic → 12.1% of signups).
• 80% of Our AI Search Traffic Goes to Our Homepage, Product Pages, and Free Tools — where AI traffic actually lands.
• How to Track and Analyze Your AI Traffic — the GA4 custom channel group walkthrough plus Ahrefs Web Analytics’ built-in AI channel.

Supporting Ahrefs studies

• 63% of Websites Receive AI Traffic (Study of 3,000 Sites) — Linehan & Guan.
• AI Traffic Has Increased 9.7x in the Past Year — the growth dataset (81,947 sites).

My speaking

• GEO? AEO? LLMO? What’s With All This AI Stuff? — Ahrefs Evolve 2025 slides; video.

From others

• Swydo — Track AI traffic in GA4 — the “floor, not ceiling” framing and an extended regex.
• Loamly — The AI traffic attribution crisis — the 70.6% dark-traffic analysis (vendor-sourced).
• MarTech — How GA4 records traffic from Perplexity Comet and ChatGPT Atlas — the AI-browser angle.
• Semrush — AI referral traffic guide — market-share breakdown by AI platform (ChatGPT, Gemini, Perplexity, Grok) and growth trends.
• Semrush — ChatGPT 17-month referral analysis — ChatGPT referral traffic grew 206% YoY and now reaches ~170,000 unique domains.
• Analytics Mania — AI traffic in GA4 — thorough setup guide with a 100+ platform regex and channel-ordering walkthrough.
• Orbit Media — Track AI traffic in GA4 — Andy Crestodina’s practitioner perspective on multi-platform tracking setup.
• Search Engine Land — GA4 + AI Mode tracking — news coverage of the Google AI Mode noreferrer bug and its partial fix.
• SparkToro — Who sends traffic on the web — dark-social baseline research that contextualizes AI dark traffic alongside TikTok, Slack, and WhatsApp.

Stats worth citing

• 35–70% of AI-driven visits land as Direct. Loamly’s analysis of 446,405 visits puts the figure at 70.6% misclassified as Direct in GA4 by default — the headline number for the dark-traffic problem (vendor-sourced; treat as an upper bound). Source
• Dark AI traffic converts at 4.1x true Direct. Loamly reports dark AI traffic converting at 10.21% vs. 2.46% for genuine Direct — the fingerprint to watch in your own Direct channel. Source
• AI traffic grew ~9.7x over twelve months (and 527% Jan–May 2025 vs. the prior year) across ~82,000 sites in Ahrefs Web Analytics. Source
• 23x conversion premium for Ahrefs. 0.5% of traffic from AI searchAI search uses large language models and retrieval-augmented generation (RAG) to synthesize an answer from multiple sources rather than returning a ranked list of links. Examples include Google AI Overviews, ChatGPT Search, and Perplexity. produced 12.1% of signups — the highest-confidence, independently documented AI conversion datapoint. Source
• ChatGPT held 64.5% of measurable AI referral traffic (Jan 2026), with Claude 18.5%, Gemini 10.6%, and Perplexity 7.3% — a fast-shifting mix that argues against tracking only one or two platforms.
• Per-platform conversion premiums (vendor-sourced, directional): ChatGPT referral ~15.9%, Perplexity ~10.5%, vs. Google organic ~1.76%.
• LLMA large language model (LLM) is a deep-learning model trained on massive text corpora to predict the next token and generate human-like text. LLMs use the transformer architecture and power AI search features like Google's AI Overviews (Gemini) and Bing Copilot (GPT-4). visitors convert 4.4x better than organic search visitors, per Semrush/Adobe research — corroborating direction with the Ahrefs 23x figure (different methodology, different baseline). Source

AI Traffic Attribution

AI traffic attribution is the practice of correctly identifying and measuring website visits that come from AI tools — ChatGPT, Perplexity, Gemini, Claude, AI Overviews, and AI browsers. It's hard because many of those tools strip the referrer header, so the visits land in your analytics as Direct traffic with no source.

Related: LLM Visibility, Measurement and Reporting, AI Share of Voice

AI Traffic Attribution

AI traffic attribution is the work of figuring out how much of your traffic actually came from AI tools — AI chatbots (ChatGPT, Claude, Perplexity, Gemini, Copilot), AI searchAI search uses large language models and retrieval-augmented generation (RAG) to synthesize an answer from multiple sources rather than returning a ranked list of links. Examples include Google AI Overviews, ChatGPT Search, and Perplexity. features (Google’s AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. and AI Mode, Bing Copilot), and AI browsers (Perplexity Comet, ChatGPT Atlas) — and getting your analytics to count it correctly.

The core problem is the referrer header. Normally, when someone clicks a link, their browser tells the destination site where they came from, and analytics tools like GA4 use that signal to label the source. Many AI tools strip or never send that header — by referrer policy, by adding rel="noreferrer" to links, by mobile-app architecture, or because the user copy-pasted the URL instead of clicking. When the header is missing, GA4 files the visit under Direct ((direct) / (none)), so the AI source disappears. Published estimates put the share of AI visits that land as Direct anywhere from 35% to 70%.

GA4 added a native AI Assistant channel on May 13, 2026, and you can build a custom channel group to reclassify referred AI sessions — but neither captures the “dark” sessions that arrive with no referrer at all. Google’s AI OverviewsAI Overviews are the AI-generated summary box Google shows above or within its regular search results, written by Gemini models from pages retrieved out of Google's normal Search index. It's a Search feature, not a separate platform or index. and AI Mode are a separate, harder case: those clicks are classified as regular Organic Search in both GA4 and Search ConsoleA free Google service that reports how a site performs in Google Search and surfaces problems with how Google crawls, indexes, and serves it. It's first-party data straight from Google — but you don't need it to appear in results., with no way to split them out. The practical takeaway is that the AI traffic you can see is the floor, not the ceiling — and because AI visitors tend to convert at much higher rates, that floor often matters more than its size suggests.

Related: LLM Visibility, Measurement and Reporting, AI Share of Voice

Changelog