AI Traffic Attribution: How to Credit ChatGPT, Perplexity, and Gemini for Your Visits

What AI Traffic Attribution Means

AI traffic attribution is the practice of assigning site visits, engagement, and conversions to the AI platform that produced them: ChatGPT, Perplexity, Gemini, Claude, Copilot, and the growing list of assistants that answer questions with links. When a reader asks ChatGPT for a product comparison, clicks through to your pricing page, and signs up a week later, attribution is what connects that signup back to ChatGPT instead of letting it dissolve into "Direct" or "Referral."

The discipline matters for the same reason channel attribution has always mattered. You cannot decide whether AI visibility work is paying off, or which platform deserves more of your content effort, without knowing what each one actually sends you. The difference from classic channels is that AI attribution is only partly observable. Some AI traffic arrives clearly labeled, some arrives disguised as another channel, and some arrives with no label at all. A working attribution setup accounts for all three.

How AI Referrals Show Up in GA4

Clicks out of AI assistants carry referrer headers like any other web click. ChatGPT sends chatgpt.com, Perplexity sends perplexity.ai, and Gemini sends gemini.google.com. In GA4 those appear as session sources, and for years the only way to see them together was a manually built report or a custom channel group filtering on those domains.

That changed in May 2026, when Google added a dedicated default channel for assistant traffic. GA4's channel group documentation defines the AI Assistant channel as users arriving "from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok." If your property uses the default channel group, recognized assistant referrers now roll up into that channel automatically, no configuration required.

The default channel is a genuine improvement, but treat it as a floor rather than the full picture. Google maintains the list of recognized sources, so a newer or smaller assistant may sit in plain Referral until Google classifies it. And two large slices of AI-driven traffic never reach the AI Assistant channel at all, which is where attribution gets interesting.

The Two Big Attribution Gaps

Gap one is Google's own AI surfaces. Clicks from AI Overviews and AI Mode land in Organic Search, not in the AI Assistant channel, because the referrer is simply google.com. From GA4's point of view a visitor who clicked your citation inside an AI-generated answer looks identical to one who clicked your blue link three results down. If AI Overviews are reshaping your search traffic, GA4 will show it only as a change in Organic Search volume and behavior, never as a separate line. Proxy techniques for isolating that effect are covered in our guide to tracking traffic from Google AI Overviews.

Gap two is missing referrers. Clicks from AI mobile apps and some in-app browsers often arrive with no referrer header, and links that users copy out of an AI answer and paste into a browser never had one. All of it lands in Direct, joining the long-standing dark traffic problem. The share varies by audience: a site whose readers live in the ChatGPT mobile app will undercount AI traffic far more than one whose readers click through on desktop. You cannot recover these sessions individually, but you can estimate the aggregate, which is what the workflow below is for.

A third, smaller gap is worth naming: AI influence without a click. An assistant can summarize your content, name your brand, and convince a user, who then searches for you by name or types your URL. That visit shows up as Organic Search on a branded query or as Direct. No analytics configuration surfaces it; the closest measurement is tracking whether assistants mention you at all, which is a visibility question rather than a traffic question. Our article on AI results tracking covers that side, and MeasureBoard's AI Rank Tracker automates it by re-running your prompts against ChatGPT, Gemini, and Claude and logging who gets cited.

Building an AI Attribution Workflow

Start with the traffic that identifies itself. If your property predates the AI Assistant default channel or you want finer control, build a custom channel group with a channel matching session sources chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, and claude.ai. Custom channel groups apply from creation onward and also work retroactively in reports, so this one change gives you a clean AI line across your reporting.

Next, open an Exploration. A free-form table with session source as the dimension and sessions, engaged sessions, and conversions as metrics, filtered to your AI sources, answers the per-platform questions the channel rollup hides: which assistant sends the most visits, which sends visitors who actually engage, and which sends buyers. Platforms differ more than most people expect, and the differences are worth reading before deciding where to invest. We compared behavior patterns across assistants in ChatGPT vs Perplexity vs Gemini traffic.

Then correlate landing pages. AI assistants link to specific pages, usually deep content rather than homepages. Cross session source with landing page in the same Exploration and you learn which of your pages assistants treat as citable. Those pages are your GEO assets; a page that draws Perplexity clicks month after month is telling you what that platform considers authoritative, and the pattern often predicts which of your other pages could earn citations with similar treatment.

A note on UTM parameters: they solve less here than in other channels. UTMs work when you control the link, and you do not control the links AI assistants generate. Tagging has a role in the slivers you do control, such as links in a custom GPT you publish or in content you syndicate to an AI platform directly, but for organic assistant citations the referrer plus channel grouping is the whole toolkit. Do not tag your own site's canonical URLs in the hope assistants will copy them; when they cite clean URLs instead, your tagged and untagged traffic splits into two lines.

Finally, estimate the dark portion. Watch Direct traffic to deep content pages, the kind of URL nobody types by hand. When Direct sessions to a specific article climb in step with labeled AI referrals to that article, the overflow is most plausibly the same phenomenon arriving without a referrer. This is inference rather than measurement, so state it as a range, but a consistent method applied monthly turns it into a usable trend line.

If you would rather not maintain the explorations yourself, MeasureBoard's AI Traffic Intelligence does the identification automatically from your connected Google Analytics property: it recognizes sessions from each AI platform, shows AI traffic as a share of your total, and trends it over time. It is included on the free plan.

Attributing Conversions, Not Just Visits

Sessions are the easy half. The harder question is whether AI traffic converts, and answering it takes care with attribution models. GA4's default data-driven attribution distributes conversion credit across a user's touchpoints, so an AI-referred first visit followed by a branded-search return visit may credit the conversion partly or wholly to search. That is not wrong, but it systematically understates channels that introduce people to you, and AI referrals are often exactly that: a first touch on a research question, days or weeks before purchase intent.

Two practical responses work without changing your model. First, compare first-user source with session source in an Exploration: first-user attribution shows how many of your converting users were originally acquired through an AI platform, whatever channel closed them. Second, judge AI traffic by engagement quality as an early proxy. Engagement rate, pages per session, and key-event completion for AI-referred sessions tell you within weeks whether these visitors behave like buyers, long before conversion volume is large enough to be statistically interesting.

Keep the volumes honest. For most sites, AI referrals remain a small share of total traffic, and small denominators make percentages swing wildly. A channel that doubled from twenty sessions to forty did not necessarily change; a channel that holds a rising trend across two quarters did. The point of attribution here is not to prove AI traffic is big. The point is to notice, earlier than your competitors, which platforms are starting to matter for your audience and which pages they reward.