How to Track AI Traffic in Google Analytics 4: ChatGPT, Perplexity, Gemini and Copilot
Sergey Carp from El Paso–Juárez, Co-founder / CTO
"People say customers come from ChatGPT now. Do we get any?" A business owner opens Google Analytics 4 and sees the familiar channels: Organic Search, Direct, Paid Search, Referral. There is no row called "AI". Visits from ChatGPT are scattered across a long list of referring websites, mixed in with directories, partners and spam. It is unclear whether any AI traffic exists, whether the work on AI visibility is paying off and what to show leadership.
The good news: you can see most AI referral traffic, and setting it up takes an hour or two. The bad news: analytics will never show the full picture, and you need to account for that from day one. Below we walk through how to track AI traffic in Google Analytics 4 step by step, and then look at what to do about the influence that never shows up in reports.
Key takeaway
In GA4, give AI assistants their own channel above Referral and track key events. Referrals undercount AI's impact, so also track your share of mentions.
Where AI traffic actually comes from
When someone asks ChatGPT, Gemini, Perplexity or Copilot a question, say, "where can I get my espresso machine repaired in El Paso–Juárez", they get an AI answer: a ready-made text with recommendations and links to source websites. In chat assistants, the links appear as citations or cards right in the conversation. In Google, the answer shows up as an AI Overview above the regular results or as a conversation in AI Mode.
From there, three things can happen, and analytics sees each of them differently:
- The person clicks a link in an AI assistant. They arrive as a referral from another website: chatgpt.com, perplexity.ai and so on. This is visible.
- The person clicks a link in an AI answer inside search. The visit comes from a Google domain and looks like regular organic search traffic. As a rule, standard reports cannot separate it.
- The person reads the answer and does not click. They remember the company name and later search for it or call. In analytics, that is a branded search, a direct visit or nothing at all.
So the job has two layers. Layer one: capture everything that comes from AI assistant domains. Layer two: measure your presence in AI answers separately, without relying on clicks.
Which referral domains to look for
The list changes: services move to new domains, launch apps, and new assistants appear. At the time of writing, the main domains are:
- chatgpt.com: ChatGPT; older data may also contain chat.openai.com;
- perplexity.ai: Perplexity;
- gemini.google.com: Gemini;
- copilot.microsoft.com: Microsoft Copilot;
- claude.ai: Claude;
- chat.deepseek.com: DeepSeek.
Once a quarter, review the full list of referring sources and add any new AI service domains that show up. Also note that ChatGPT often appends utm_source=chatgpt.com to its links. That parameter lets you identify the visit even when the referrer is not passed.
How to track AI traffic in Google Analytics 4: step by step
Google regularly renames and moves items in the GA4 interface, so the menu names below are approximate. Look for the closest match in your account.
Step 1. Find ChatGPT referral traffic in the acquisition reports
In the acquisition section of your reports, open the traffic acquisition report and switch the primary dimension to session source / medium. AI referrals usually look like chatgpt.com / referral, perplexity.ai / referral and so on. By default, GA4 assigns them to the Referral channel, where they are mixed in with every other website. Set the date range to at least three months: for most websites AI traffic is still small, and in a single week it may not show up at all.
Step 2. Create a custom channel group for AI assistants
To see AI assistants as a separate row, go to the property settings, find the data display or channel group settings, and create a custom channel group. The easiest way is to copy the default channel group and add a new channel to it, for example "AI Assistants". The channel condition is based on session source and lists the AI assistant domains:
- chatgpt.com;
- perplexity.ai;
- gemini.google.com;
- copilot.microsoft.com;
- claude.ai;
- chat.deepseek.com.
You can build the condition as several "source contains" rules joined with OR, or as a single regular expression that matches any of the domains. Two details matter here. First, channels in a group are evaluated in order from top to bottom, and a session lands in the first channel that matches. That means the new channel must be moved above the Referral channel, otherwise ChatGPT visits will stay in Referral. Second, if you use a regular expression, the dots in domain names must be escaped, and it is best to write and test the expression with an analyst so it does not catch anything extra. For example, gemini.google.com must not match regular google.com traffic.
After saving, select the new channel group in the traffic acquisition report. As a rule, custom channel groups in GA4 also apply to historical data, so you can see past trends right away.
Step 3. Build an exploration
Standard reports are fine for an overview, but details are easier to see in Explore. Create a free-form exploration: put session source or your "AI Assistants" channel in the rows, landing page as a second row dimension, and sessions, engaged sessions and key events in the values. Add a filter for your AI channel. This answers the main question: which pages AI assistants recommend most often, and what people do after they arrive.
Step 4. Set up key events
In GA4, conversions are called key events. Mark form submissions, phone link clicks, purchases and bookings as key events. Without them, your "AI Assistants" channel shows visits but not revenue.
Step 5. Save a monthly snapshot
Save the exploration so you do not have to rebuild it. Once a month, record three numbers: sessions from AI assistants, key events from that channel and the top landing pages.
AI traffic analytics in GA4: where to look for what
| Task | Where to find it in GA4 | Notes |
|---|---|---|
| See AI assistant domains | Traffic acquisition report, session source / medium dimension | By default they sit in the Referral channel |
| Check UTM parameters | Session source or session source / medium dimension | ChatGPT often adds utm_source=chatgpt.com |
| Separate AI traffic | Custom channel group with an "AI Assistants" channel | Place the channel above Referral |
| Landing pages | Exploration: AI channel plus landing page | Shows which pages AI recommends |
| Leads and sales | Key events in the traffic acquisition report or exploration | Mark the right events as key events first |
| Clicks from AI Overviews and AI Mode | Not visible separately | Counted as organic search from Google |
| People who never clicked | Not visible | Measure through AI visibility tracking |
If your website also targets Russia, the same approach works in Yandex Metrica: a saved segment by referring domains and UTM parameters, combined with goals.
Limitations: what analytics will not show
Honest analytics starts with understanding its blind spots. Traffic from ChatGPT and other AI assistants will always be underreported, and here is why.
- AI answers inside search. A click from a Google AI Overview or AI Mode comes from a Google domain and looks like regular organic search traffic.
- People do not click. An AI answer often settles the question completely. The person remembers the name and comes back later through a branded search, a direct visit or a phone call.
- No referrer. Mobile apps and some browsers do not always pass the referring source. Those visits end up in Direct.
- Copied links. Someone copies a URL from an answer and pastes it into the browser: that is a direct visit too.
- New services. A new assistant's domain does not land in your AI channel until you add it.
The conclusion is simple: AI referrals in analytics are a floor for the real impact, not the full picture. Judging GEO by referrals alone is like judging a billboard by the number of people who took a photo of it.
The second metric: share of mentions across tracked prompts
Since clicks do not tell the whole story, you need a metric that does not depend on clicks. We use share of mentions across tracked prompts. The tracked prompts come from prompt research: the list of questions your customers ask AI assistants, the AI search equivalent of keyword research.
How it works:
- we build a prompt list: for example, 50 questions about choosing, pricing, comparing and trusting companies in your industry;
- once a month we run these prompts in the AI assistants that matter to your customers;
- for each answer we record whether your company is named, whether your website is linked, who is named instead of you and whether there are factual errors;
- we calculate the share of answers that mention you and watch the trend by month and by assistant.
A hypothetical example for illustration: in the first month, the company is mentioned in 4 answers out of 50; six months later, in 13. Over the same period, GA4 shows growth in referrals from AI domains and in branded searches. Together, the two metrics show what neither can show alone. AI answers vary with the wording of the question, so watch the trend, not a single measurement.
How to connect AI traffic to sales
Two simple tools close some of the remaining blind spots:
- A "How did you hear about us?" question. Add an option like "an AI assistant recommended you (ChatGPT, Gemini, etc.)" to your lead form or your sales team's call script. It is rough, but it shows what no tracking code can.
- Passing the source to your CRM. Send the traffic source from GA4 along with each lead, for example through hidden form fields. Then you see not only a lead from an AI assistant but also whether it turned into a deal.
The result: GEO becomes measurable
After setup, you have a dedicated AI channel in GA4 that shows referrals from AI assistants, their landing pages and their key events. You have a monthly share of mentions across tracked prompts. You have form and CRM data. That is enough to make decisions: which pages AI assistants recommend most readily, which questions you are still missing from, and what content to create next. See how this fits into the broader work on your website in our AI search optimization checklist and our guide to getting recommended in ChatGPT.
How we measure and grow AI visibility
AI Search Optimization has been doing SEO since 2002. We have 100+ clients around the world, Google and Yandex certifications, and one team that handles your website and your AI visibility end to end. We optimize for every major AI assistant, in any market, and we set up analytics so you can see what the work delivers.
Analytics setup and a monthly report on both metrics are part of the service, which includes:
- an AI visibility audit of your website and of AI answers in your industry;
- prompt research: the questions your customers ask AI assistants;
- technical fixes on the website: AI crawler access, llms.txt, schema markup, speed, errors;
- new content built for AI answers;
- brand mentions on third-party sites;
- publishing articles in external sources to build the authority of your company and its experts;
- ongoing AI visibility tracking, including share of mentions across tracked prompts.
GEO is built on top of good old SEO, and AI evaluates your website as a whole. That is why we take care of the entire site from day one, from technical fixes to GEO. We guarantee delivery of GEO tasks, error fixes and fast adaptation of your website to changing AI search requirements. We do not promise a position in AI answers, but we show you what is changing, in numbers.
The monthly fee is $290 per month, including the first 10 hours of work. Every additional hour is $45. GEO work is billed on a time and materials basis: we invoice only the work completed and delivered that month. No discounts or free months.
Book a call. We will look at your Google Analytics 4, show you how much traffic already comes from AI assistants and run a first snapshot across your tracked prompts.