Why can't Search Console show me ChatGPT referral traffic?
Because Search Console only reports on Google Search. A person who clicks a link inside a ChatGPT answer never touched a Google results page, so there's nothing for Search Console to record. That's the first thing I tell anyone asking whether their GEO services are paying off: the answer isn't in the report they already check every week.
Google's own AI surfaces are a different case. Google's documentation says sites appearing in AI Overviews and AI Mode are included in the overall search traffic in Search Console, under the Web search type. So those clicks are counted but not split out, and you can't filter down to AI Overview clicks alone.
I've had to write this down on a real project. On a Virginia vacation rental site I prepared for AI search, the case study says plainly that Search Console doesn't report ChatGPT, Perplexity or AI Overview citations separately, which is why the AI work there is described as done rather than proven by its numbers.
So the measuring happens in Google Analytics 4, where the visit actually lands, and it starts from one honest rule: what GA4 counts is the least ChatGPT sent, not the most.
How much ChatGPT referral traffic do websites get?
A small share of total visits, growing fast, and bigger than every other assistant's. Two studies published this year, measured differently, point the same way.
SE Ranking's study of 101,574 websites, published on 9 July 2026, says: "Worldwide, the share of referral traffic from ChatGPT to websites rose from 0.23% in April 2026 to 0.32% in May 2026." BrightEdge's figures, reported by TNW on 24 September 2026, say "ChatGPT sent 95.1% of the traffic that AI chatbots referred to websites in August," and add that "AI tools still send far fewer visitors to websites than search engines do."
They don't agree on everything. SE Ranking puts ChatGPT's share of all AI referral traffic at 75.96% in 2026, well below BrightEdge's August figure, because the two use different samples and periods. That's why I read them as direction, not as a benchmark for your site. Your own number comes from your own property, and the rest of this page is how I get it.
SE Ranking saw its own biggest single-day jump in ChatGPT visits on 13 May 2026, the day Google added the AI Assistant channel. So when I compare a site's ChatGPT visits across that date, I compare by session source, chatgpt.com, rather than by channel, since the channel didn't exist before it.
How I track ChatGPT referrals in GA4, step by step
This is the order I work in on any property I'm given access to. The first pass is the slow one. After that it's a short check each month.
- Step 01
Open the AI Assistant channel first
In the Traffic acquisition report I set the primary dimension to Session default channel group and look for the AI Assistant row. Google's definition is exact: the channel matches when the medium is
ai-assistant, and GA4 sets that medium when the referrer matches its list of AI assistants. I add Session source as a secondary dimension so chatgpt.com, claude.ai and gemini.google.com show as separate rows. - Step 02
Search Referral and Unassigned for AI domains
Not every AI visit gets the new medium. I filter Session source for chatgpt, openai, claude, perplexity, gemini and copilot across every channel. An assistant GA4 doesn't recognise yet arrives as a plain referral, and a tagged link without a matching referrer can land in Unassigned with a medium of
(not set). - Step 03
Build one custom channel group that catches all of them
In Admin, under Data display, I copy the default channel group and add a channel called AI assistants with the condition Source partially matches regex
chatgpt\.com|chat\.openai\.com|claude\.ai|perplexity\.ai|gemini\.google\.com|copilot. I pick partially matches on purpose: plain matches regex has to match the whole source, which is why Google's own AI assistants example pads almost every term with.*. Traffic goes to the first channel it matches, so I move mine above Referral. A standard property gets two custom groups, and Google says they apply to reports retroactively, so the history fills in the moment it's saved. - Step 04
Tie the sessions to key events
A visit that doesn't turn into an enquiry is curiosity, not revenue. I check that the property's real conversions, the form submit, the call click or the booking, are marked as key events, then read them against the AI assistants channel. If there aren't any key events set up, that's the first fix, and it's worth more than the AI number itself.
- Step 05
Record which pages the assistants link to
Landing page plus Session source shows which pages ChatGPT and the others actually send people to. It's often not the homepage: on this site the one ChatGPT visit landed on a case study. The pages that get linked tend to answer one question cleanly, which tells me which pages are worth keeping sharp.
- Step 06
Log it monthly, with the caveats beside it
Each month I note sessions, key events and top landing pages for the AI assistants channel, with the date I pulled them. For a deeper look I save a session segment in Explorations with the same source conditions, which Google describes as a subset of your data for analysis, so I can compare AI visitors' paths with everyone else's. The note always says this is a floor.
Don't build a filter that excludes AI traffic from everything else. A custom channel group only relabels sessions, so the default reports stay exactly as they were, and you can delete the group later without touching the data. Google does warn that deleting a channel group can't be undone, so name it clearly.
Rather have me do this? One person, whole job.
Book a call on UpworkWhich assistants pass referrer data, and which don't?
Google's May 2026 note names ChatGPT, Gemini and Claude as examples of the assistants GA4 now recognises. I don't keep a fixed list beyond that, because what an assistant passes depends on where the click happens: the web app, a phone app, a browser extension or a link someone copied. I read the property instead of trusting a list.
Here's the one case I can show from my own data. This site's GA4 property, checked on 27 September 2026, has a single ChatGPT session since it started collecting data in January. It arrived on 24 September with source chatgpt.com, medium ai-assistant, and a session manual source of chatgpt.com, which is the utm_source value. So that link carried a tag, and GA4 still filed it under AI Assistant. One session isn't a trend, and I won't pretend it is, but it shows the parts fitting together.
| Where the click starts | What GA4 sees | Where it usually lands |
|---|---|---|
| A link in a recognised assistant, opened in a browser | The assistant's domain as source, medium ai-assistant | AI Assistant |
| A link from an assistant GA4 doesn't recognise yet | The assistant's domain as source, medium referral | Referral |
| A tagged link that arrives without a matching referrer | A source from the tag, medium (not set) | Unassigned |
| An address typed in, or a copied link with its tag stripped | No referrer and no tag | Direct |
| An AI Overview or AI Mode link in Google | Google as the source | Organic Search, mixed in with ordinary Google clicks |
The Direct row is the one nobody can fix. A visit with no referrer and no tag has nothing to say where it came from. Under the default referrer policy a cross-site click sends the origin only, so GA4 sees chatgpt.com and never the conversation, and you'll never see the prompt that led to your page, however the tracking's set up.
What does a client saying they found the business on ChatGPT tell me?
It tells me the recommendation happened, and it tells me GA4 probably missed it. When a person asks ChatGPT for a jeweller and gets a name, they may not click the link at all. They search the name in Google, or type the address, and the visit arrives as a brand search or as Direct. The assistant did the persuading, and a different channel gets the credit.
That's the situation on a Canadian lab grown diamond store I've worked with. The owner reports clients who found the store because ChatGPT and Claude recommended it. I treat that as the owner's report, which is exactly what it is. AI recommendations can't be counted in Search Console, and a recommendation followed by a brand search won't show in the AI Assistant channel either.
So I don't dismiss those reports, and I don't turn them into a number. I suggest a plain where-did-you-find-this-business field on the enquiry form, with ChatGPT and other AI assistants as options. It's low-tech, and it's the only method that catches the person who never clicked.
What can't be measured yet?
A lot. Mentions without clicks aren't measurable in any analytics tool: if ChatGPT names a business and nobody clicks, nothing reaches the site. The prompts aren't visible either, as the referrer only carries the assistant's domain. AI Overview clicks are counted in Search Console but can't be separated from ordinary Google clicks. And a visit with neither a referrer nor a tag falls into Direct with everything else.
What I can measure is still useful: sessions GA4 attributes to AI assistants, the key events they produce, the pages assistants link to, and whether all three rise or fall month on month. That's enough to tell whether the work on letting AI crawlers reach the site, starting with the AI bots your WordPress robots.txt allows, is being noticed. It's not enough to promise anyone a number, and I won't.
If you'd like one person to set up the AI assistants channel on your GA4 property, check your key events and read the number each month beside Search Console, that's work I do remotely. You'll get the floor, the caveats and the pages that earn it, in plain words.
Book a call on Upwork
