How to Track ChatGPT Traffic and Leads (GA4, utm_source=chatgpt.com, Brand Search)

You track ChatGPT traffic in GA4 by capturing two things: sessions whose source is chatgpt.com (via the utm_source=chatgpt.com parameter ChatGPT appends to many links, or via the referrer) and the much larger indirect effect that never carries a ChatGPT label, which shows up as branded search, direct visits and “heard about you from ChatGPT” in your forms and sales calls. If you only do the first, you will conclude ChatGPT is irrelevant. If you do both, you will see what it is actually doing to your pipeline. This guide sets up both layers, step by step.

Why the referral line understates everything

Start with the honest numbers. In SE Ranking’s study of 101,574 sites, AI platforms accounted for 0.32% of all website traffic in 2026, up from 0.02% in 2024, a 16x increase; organic search accounted for 42.75%. Within that AI slice, ChatGPT held 74.78% of referrals in 2026, and its referrals grew 27% year on year. It is the dominant AI referrer and still a rounding error next to Google.

Now the behavior behind the numbers. A 2026 Idea Grove survey of 1,000 US consumers found only 2% would buy from an unfamiliar brand on an AI recommendation alone; 45% immediately Google the brand; 18% go to review sites; 71% rely on Google rankings and 78% say reviews raise trust. The recommendation happens in ChatGPT. The click happens somewhere else.

So a brand that ChatGPT names in most answers for its category can see a few dozen ChatGPT sessions a month and a sizable, unlabeled rise in branded search. The attribution job is to connect those two. For how buyers move through this differently from a Google journey, see the ChatGPT buyer journey vs the Google buyer journey.

Layer one: capture direct ChatGPT referrals in GA4

What arrives and how it is labeled

When ChatGPT includes a link to your site and a user clicks it, the visit arrives in one of two forms. Many links carry utm_source=chatgpt.com appended by ChatGPT, which GA4 reads as session source “chatgpt.com” with medium “(not set)” unless a utm_medium is also present. Others arrive with an HTTP referrer of chatgpt.com and no UTM, which GA4 files as source “chatgpt.com”, medium “referral”. A few arrive with no referrer at all because of privacy settings or app-to-browser handoffs, and those land in direct traffic where you cannot recover them.

Step 1: verify you are receiving it

In GA4, open Reports, then Acquisition, then Traffic acquisition, and search the table for “chatgpt”. If you see nothing, check Explore with a free-form report using Session source as the dimension and a filter containing “chatgpt”. Zero sessions over 90 days for an established brand usually means you are rarely cited, not that tracking is broken; see the next section.

Step 2: build an “AI assistants” channel

GA4’s default channel grouping files chatgpt.com under Referral, mixed in with every other referring site. Create a custom channel group: Admin, Data display, Channel groups, then create a new group and add a channel named “AI assistants” with conditions on Session source matching a regular expression such as chatgpt\.com|openai\.com|perplexity\.ai|gemini\.google\.com|copilot\.microsoft\.com|claude\.ai. Place it above Referral in the order so it takes precedence. Make it the default in your reports.

Step 3: register landing page and source as conversion dimensions

Mark your lead events (form submissions, demo requests, trial signups) as key events if they are not already, then segment them by the new channel. The question you want answered monthly is simple: how many key events came from AI assistants, from which landing pages, and at what rate compared with organic search.

Step 4: pass the source into your CRM

Capture utm_source, utm_medium, referrer and landing page in hidden form fields and write them to the lead record. GA4 tells you about sessions; your CRM tells you about revenue. Without this join you will never know whether ChatGPT leads close.

Why you may be getting mentions and no clicks

A click requires a citation: a link in the answer pointing to your domain. A mention does not come with a link unless retrieval happened to pull your page. In practice a brand can be named in most answers and linked in very few, because the model’s confidence in you comes from consensus across third-party pages while the sources it lists are the pages it retrieved that day.

Those sources also rotate heavily. A 2026 study found only 10.6% of cited URLs persisted across 28 days and 40 to 60% of cited sources rotated monthly. Your referral line will therefore be lumpy: a page gets cited for a few weeks, traffic appears, then it rotates out. Do not read a drop in ChatGPT referrals as a drop in recommendations. Measure the two separately; the distinction is the subject of mentions vs citations in ChatGPT.

Layer two: measure the indirect effect

Branded search as the primary outcome metric

Since nearly half of consumers Google the brand rather than click, branded search volume is the second signal to watch, and for many brands the more important one. Pull it from Search Console (impressions and clicks on queries containing your brand name, including misspellings), and from Google Ads brand campaign impression volume if you run one.

Plot monthly branded impressions against your ChatGPT mention rate and top-three share from a monthly audit. Position matters here as much as presence, because a brand named first is the one the buyer searches next; the reasoning is in why position 1 to 3 in a ChatGPT answer is the whole game. Suppose your audit shows mention rate across a cluster of 40 prompts rising from 30% to 65% over three months while branded impressions rise over the same period with no other campaign change; that correlation is not proof, but it is the kind of evidence a finance team accepts when nothing else explains the lift. The measurement side is covered in how mention rate and position are measured properly.

Isolating ChatGPT’s contribution to brand search

Brand search moves for many reasons: PR, paid social, events, seasonality. Three ways to tighten the attribution:

• Segment branded queries by modifier. Queries like “[brand] reviews”, “[brand] pricing”, “[brand] vs [competitor]” and “is [brand] legit” are the ones a freshly recommended buyer types. Track them as a group.
• Compare geographies. If your ChatGPT work targets one market, branded search should move there and not elsewhere.
• Hold a control period. Run the audit and brand-search tracking for a month before changing anything, so you know the baseline noise.

Direct traffic and new-visitor share

Buyers who hear a brand name and type the URL land in direct traffic. A rise in new-user direct sessions to your homepage and pricing page, alongside a rise in mention rate, is corroborating evidence. It is weak on its own and useful in combination.

Self-reported attribution

Add “ChatGPT or another AI assistant” as an option in the “How did you hear about us?” field on every lead form, and ask the question on sales calls. Self-reporting under-counts, because people pick the last touch they remember, which is often Google. Treat it as a floor. When even the floor shows AI assistants at a meaningful share of leads, the real figure is higher.

Attribution models and ChatGPT

GA4’s data-driven attribution can only assign credit to touchpoints it saw. A ChatGPT recommendation followed by a branded Google search looks to GA4 like an organic search conversion with no prior touch. No attribution model inside GA4 will fix this, because the first touch happened in a window GA4 never observes.

The practical response is to treat ChatGPT as an upper-funnel channel measured by lift rather than by last-click credit: mention rate and position as the leading indicators, branded search and direct as the mid-funnel indicators, and CRM close rates on AI-sourced leads as the lagging one. This is the same logic you would apply to podcast advertising or out-of-home, where the exposure cannot be clicked.

Reading the small direct channel properly

The sessions you do capture are worth studying closely, precisely because there are few of them. In our observation, visitors arriving with utm_source=chatgpt.com behave differently from organic visitors: they land on deeper pages (pricing, comparison, specific product pages) rather than the homepage, they arrive with a specific question already framed, and they are further along in a decision. Measure their conversion rate, pages per session and the pages they land on, but compare over quarters, not weeks; a channel with forty sessions a month cannot support weekly conclusions.

Use the landing pages as a diagnostic too. The pages ChatGPT links to are the pages it is retrieving and grounding on. If they are out of date, unclear or missing a pricing table, that is where to fix first, because those are the pages a recommended buyer sees.

A monthly attribution report that fits on one page

Suppose you report monthly. The report has four blocks:

1. Visibility: mention rate, average position and top-three share per prompt cluster from the audit, with month-on-month change.
2. Direct channel: AI-assistants channel sessions, landing pages, key events and conversion rate from GA4; leads and pipeline from the CRM where the source was captured.
3. Indirect channel: branded search impressions and clicks (total and the “reviews, pricing, vs, legit” subset), new-user direct sessions, and self-reported AI attribution share.
4. Interpretation: one paragraph on whether the leading indicator moved and whether the lagging ones followed.

Keep the same definitions every month. Attribution that changes its rules cannot show a trend.

Ads in ChatGPT and attribution

Since early 2026, some ChatGPT users see ads. Third-party reporting indicates ads launched for logged-in US adults on the Free and Go tiers, with the first paid ads on February 9, 2026, shown as a “Sponsored” card under the answer’s sources and sold by OpenAI’s sales team rather than through a self-serve manager as of July 2026. If your brand ever runs such ads, tag their links with a distinct utm_medium so paid clicks do not contaminate your organic ChatGPT channel. The organic recommendation inside the answer remains separate from the ad and is not for sale.

Where visibility work fits

Attribution tells you what happened. It does not change what ChatGPT says. If the audit shows you are rarely named, no amount of GA4 configuration will produce traffic or brand lift. The levers that move the answer are covered in the complete guide to ChatGPT brand visibility.

The Blue Ocean GPT service reports at exactly the level this guide asks for: per-prompt mention rate and position from about 100 answers per prompt per market, reproducible by your team in ChatGPT, re-audited monthly, with no lock-in. We also log whether a link to your domain was present in each answer, so you can see the citation side alongside your GA4 channel. Targets, never guarantees; how we do it is explained on your call. To see your current baseline, find out what ChatGPT says about your brand.

What to do next

• Create an “AI assistants” channel in GA4 today and pass utm_source and landing page into your CRM as hidden fields.
• Start tracking branded search, including the “reviews, pricing, vs, legit” subset, before you change anything, so you have a baseline.
• Run a ChatGPT visibility audit on the same monthly cadence as the attribution report, and plot the two together.