ChatGPT traffic analytics: how to measure ChatGPT referrals in GA4

What ChatGPT traffic analytics can and can't see

ChatGPT traffic analytics means two different things, so settle which one you want first. One is the traffic chatgpt.com gets itself: how many people use ChatGPT each month. That's what the traffic graphs at the top of Google show, and they come from outside estimates. Semrush's published analysis of ChatGPT traffic, for example, says it rests on a panel of US clickstream data. None of those tools can see your GA4 property. The other meaning, and the one this guide covers, is the traffic ChatGPT sends to your own site.

Two clues can travel with a click from ChatGPT to your site. The first is a tag in the link. OpenAI's publisher FAQ says ChatGPT automatically adds utm_source=chatgpt.com to referral URLs from its search results, so analytics tools such as Google Analytics can pick it up. Google's Analytics help says utm_source feeds the Session source dimension.

The second clue is the referrer: the address of the page the person came from, which the browser passes along. MDN's documentation on the Referrer-Policy header says the default policy sends only the origin when a link goes to another site. So the most GA4 learns is chatgpt.com. It never sees the chat, the question or what the answer said about you.

A click also needs a link to click. OpenAI's help on searching the web with ChatGPT says ChatGPT chooses to search based on what you ask, or the person can switch search on, and that responses which use web search may include citations. An answer written without a search has no source links at all. So a person can read your business name in an answer, decide to ring you and leave no trace in GA4. Plan for that gap from the start instead of reading a small referral count as small influence.

One setting decides whether there's anything to measure at all. OpenAI's crawler documentation says sites that opt out of OAI-SearchBot won't be shown in ChatGPT search answers, though they can still appear as navigational links. Check your robots.txt before you conclude ChatGPT sends you nothing. A crawler hit on its own proves little, as the guide on why a crawl does not prove an AI citation explains.

Find ChatGPT referrals in GA4, step by step

Start with the standard report. It needs no set-up, and anyone with access to the property can follow the first five steps.

1. In GA4, go to Reports, then Acquisition, then Traffic acquisition. The table opens on Session default channel group.
2. Look for the AI Assistant row. Google's Analytics release notes date this channel to 13 May 2026, and Google's channel definitions say it covers sources like ChatGPT, Gemini, Deepseek, Copilot and Grok. It leaves out Google's AI Overviews and AI Mode, which GA4 counts in Organic Search. Don't stop here: this row only holds ChatGPT clicks that arrived with a referrer and no tag, as the next section explains.
3. Switch the primary dimension to Session source / medium and type chatgpt into the search box. Tagged ChatGPT clicks show up here as chatgpt.com / (not set). Leave out any row carrying the medium you put on your own paid ChatGPT ads, such as cpc. Every other row with chatgpt in the source belongs in your count, whatever its medium says.
4. Add Landing page + query string as a secondary dimension. It shows which pages people arrived on, the most useful clue you'll get about which answers linked to you.
5. Check the Key events column. If your enquiry form and phone clicks aren't marked as key events, fix that first, or you'll be counting visits and learning nothing about enquiries.
6. Save the filter as its own report, because text in the search box isn't saved. Google's help says creating a detail report needs the Editor or Administrator role. Open Library at the bottom left of Reports, choose + Create new report, then Create detail report, and pick the Traffic acquisition template. In the panel on the right, add a report filter: Session source contains chatgpt. Type it in lower case, since Google says partial matches are case sensitive. Save it under a name you'll recognise. Without that role, use the Explore route in the section after next, or ask whoever manages your GA4 property.
7. For a view that sorts every channel, anyone with Editor access can build a custom channel group with its own AI channel placed above Referral, and Google's help walks through it. Google says custom channel groups apply to your reports retroactively, so a custom group covers the months before the AI Assistant channel existed as well. A rule on Source also catches the tagged clicks the default AI Assistant channel leaves in Unassigned.
8. Test the pattern before you trust it. Google's example AI channel is a matches regex rule, and Google's regex help says GA4 regex is a full match by default. So the first part of that example, ^.*ai, catches any source ending in ai: every.ai domain, AI assistant or not. For ChatGPT alone, a rule of Source contains chatgpt is simpler and easier to check. Google also says each visit goes to the first channel it matches, so if you run ChatGPT ads, keep your paid channel above the AI channel.

Why one ChatGPT visit can wear different labels

A click from ChatGPT can land in several places in GA4, depending on what survives the trip. The AI Assistant channel needs a medium of ai-assistant, and Google's channel rules say GA4 sets that medium when the referrer matches its list of AI assistants. The tag overrules that. Google's help on traffic-source dimensions says that when any UTM parameter is on the URL, GA4 takes all its traffic-source values from the UTM parameters alone, and any that are missing show as (not set). OpenAI's FAQ describes a tag that names only the source, so a tagged click records chatgpt.com / (not set), even when the referrer arrived too. None of Google's default channel rules matches that pair, so it's likely to sit in Unassigned, the label GA4 uses when no rule matches.

Google's Analytics help lists missing UTM tags, redirects that strip them, addresses typed straight into a browser and ad blockers among the reasons a visit shows as (direct) / (none). For those visits GA4 never received the source, so nothing you change later can bring it back. A custom channel group only regroups the source and medium GA4 recorded, so it can't move a (direct) / (none) visit into an AI channel either.

The table at the end of this section sums up where each kind of visit is likely to land. The one gap you control is your own site. Many sites redirect somewhere on the way in: http to https, www to non-www, a missing trailing slash, an old page to its replacement. Any of those can drop the tag. This test takes a few minutes:

1. Pick a page ChatGPT might link to, such as your main service page.
2. Add?utm_source=redirect-test to the end of its address. A made-up value keeps the test out of your ChatGPT count.
3. Try each version people might reach: with and without www, starting with http:// instead of https://, and without the trailing slash.
4. Open each one in a private browser window and check the address bar once the page settles. The tag should still be there.
5. If the tag disappears in any version, ask whoever manages the site to keep query strings through every redirect. The fix sits in the redirect rules on the server or in the site's CMS.

Build a private view in Explore and add the other AI assistants

If you can't create reports, use Explore instead. Google's role definitions say the Viewer role can create, edit and delete its own explorations, and Google's Explorations help says a new exploration is visible only to you until you share it. That suits a business whose agency or web developer owns the GA4 property. Sharing one needs at least the Analyst role, and Google says shared explorations are read-only for everyone else. Here's a free form exploration that does the job:

• Click Explore in the left navigation and open a blank exploration. Free form is the default technique.
• Under Variables, add the dimensions Session source / medium and Landing page + query string, and the metrics Sessions, Engaged sessions, Engagement rate and Key events.
• Drag Session source / medium into Rows and the four metrics into Values.
• Add a filter: Session source contains chatgpt. For every AI assistant at once, switch the filter to matches regex and use a pattern like the one in the table below.
• Set the date range, then add Landing page + query string as a second row dimension to see which pages each source sends people to.
• Check the date range against your data retention. Google's Explorations help says properties keep 2 months of data by default, and Google's data retention help says that setting affects explorations and funnel reports but not standard reports. To compare more than two months in Explore, an Editor can raise retention to 14 months under Admin, Data collection and modification, Data Retention. Google says an increase also applies to data already collected.

Worked example: one quarter of ChatGPT referrals for a local electrician

Say an electrician in Maroochydore wants to know whether ChatGPT is worth watching. The owner builds the exploration above, adds an AI option to the enquiry form, and records one row a month. The figures below are made up to show the method. They are not benchmarks and say nothing about what any real business should expect.

Month one looks exciting: 4 ChatGPT sessions and 1 key event, an enquiry from one visit in four on paper. Organic Search in the same month has hundreds of sessions. The owner resists the urge to say ChatGPT converts better. With 4 sessions, one more or one fewer enquiry moves the rate by a quarter. That's noise. A practical rule is to wait until a single visit moves a rate by no more than a couple of points before comparing channels, which means about 50 sessions, and to compare totals, not rates, until then.

Over the quarter the landing page column tells a clearer story. Most ChatGPT sessions land on the switchboard upgrades page, none on the home page. So the owner writes five questions a customer might type before booking that job, such as whether a switchboard needs upgrading before an EV charger goes in, and asks each one in a fresh ChatGPT chat. Where the page is cited, the owner notes what the answer said about it. Where it isn't, they note which sources were cited instead. One run is a sample, not a ranking: answers can change between chats and over time, which is why the AI visibility report comparison guide treats prompt checks as samples too.

Then the owner reads the self-reported column against the sessions. Two customers in month three named ChatGPT on the phone, but GA4 shows only one ChatGPT key event. That isn't a contradiction. One customer read the answer and rang straight away. The sheet now shows the page that earns ChatGPT clicks, the questions behind it and the enquiries clicks miss, which is far more useful than a traffic total.

Fill the gaps with what customers and Search Console tell you

GA4 only records visits. Someone who reads your name in an answer and then rings you, or searches your business name on Google, leaves no ChatGPT trace at all. Two other sources narrow that gap: what customers tell you, and what Search Console shows about searches for your name. Neither is complete. Customers forget or guess, and branded searches rise for plenty of reasons that have nothing to do with ChatGPT. Phone enquiries also need their own matching process, covered in call tracking and conversion reconciliation.

Search Console's branded queries filter makes the second source easier to read. Google's Search Central blog says a branded query includes your brand name, its variations or misspellings, and brand-related products or services. Search Console's help says the filter isn't available for sites with a low number of impressions, holds data from March 2025, and may classify some queries wrongly. It also says anonymised queries drop out of the totals once a query filter is applied, so a filtered total will read lower than the unfiltered chart. Set it up like this:

• Add a 'How did you hear about us?' question to your enquiry form with a short list: Google search, Google Maps, ChatGPT or another AI tool, a friend or referral, social media, and Other with a free text box. Keep it optional so it doesn't get in the way.
• Ask the same question on the phone and write the answer in your job notes or CRM. If someone says Google, ask once whether they'd seen your name anywhere first, and record both answers.
• In Search Console, open Performance, then Search results. Click the filter bar, choose Query, then Branded. If the option isn't there, use Queries containing with your business name instead.
• Compare the last three months with the same three months a year earlier, so seasonal swings don't read as ChatGPT influence.
• Once a month, record the five figures in the worked example: ChatGPT sessions, engaged sessions, key events, customers naming an AI tool and branded clicks.

How to read the numbers without fooling yourself

No one can guarantee ChatGPT will cite or link to a business. Getting cited in the first place is the work of AI SEO. Measurement has a narrower job: counting the visits when a link does appear, and showing whether the page was ready for them.

Referral counts measure clicks. They can't tell you how often ChatGPT mentions or recommends you, a difference explained in AI recommendations versus mentions. Use these rules when you review the month, alongside the wider checks in how to review a monthly SEO report:

• Judge by key events first. A few ChatGPT visits that become enquiries are worth more than a spike that leaves straight away.
• Compare quality on the same metrics. Put ChatGPT and Organic Search side by side on engagement rate and key events per session, using GA4's definition of an engaged session: one that lasts longer than 10 seconds, has a key event or has 2 or more page views. Only compare rates once the ChatGPT sample is big enough, as the worked example shows.
• Look at landing pages before totals. They show which of your pages ChatGPT links to, and that's where your effort should go.
• Compare like with like. The AI Assistant channel only dates from May 2026, so measure earlier months with a source filter or a custom channel group.
• An empty or small AI Assistant row doesn't mean ChatGPT sends you nothing, because tagged ChatGPT clicks are likely to sit in Unassigned as chatgpt.com / (not set). Check the Session source filter for chatgpt across all channels first. If that shows nothing either but customers say they found you through ChatGPT, check robots.txt for OAI-SearchBot, test your redirects, then look through the Direct rows.
• If ChatGPT visits rise but enquiries don't, look hard at the landing pages. The website conversion checklist covers what a visitor needs before they'll take the next step.
• Keep Google's AI features separate. Clicks from AI Overviews and AI Mode sit in Organic Search, and reading them is its own job: see how to interpret Google AI impressions.
• Keep paid clicks apart. OpenAI's crawler documentation confirms ads can be submitted for ChatGPT, so if you run them, tag those landing URLs with utm_source, utm_medium and utm_campaign. Google's help recommends setting them together, because missing parameters show up as (not set). Give paid clicks a utm_medium you never use for anything else, such as cpc, so their rows split cleanly from referrals.