How to Track ChatGPT Traffic Now That Most of It Shows Up as Direct
If your ChatGPT referrals fell off a cliff sometime after October 7, there's a good chance ChatGPT stopped labeling its links, so the people it sends you are now being counted as Direct traffic. To track ChatGPT traffic from here on, you have to look at Direct traffic on the pages ChatGPT cites, fix how your analytics groups AI sources, and measure what the engines say about you, since the click report won't carry the whole story anymore.
That's the short answer. The longer one is below, along with what the early data shows and a quick check you can run on your own setup.
The short version
With the GPT-6 rollout, ChatGPT links stopped carrying the utm_source=chatgpt.com tag and stopped passing a referrer, so analytics tools file those visits as Direct (Search Engine Roundtable).
On about 500 pages where ChatGPT was a big traffic source, tracked ChatGPT visits fell by roughly 31,000 a day while Direct rose by roughly 26,000 a day on the same pages (Cedric Huesler via Search Engine Roundtable).
GA4's new AI Assistant channel can only count what arrives labeled, so it will undercount from here on.
Clicks were always a small slice of what AI does for a brand. The bigger effect shows up in direct visits and branded search, and in whether the engines recommend you at all.
What changed with ChatGPT links
OpenAI started rolling out GPT-6 on October 7 (OpenAI). Within a few days people who watch analytics for a living started noticing their ChatGPT numbers sagging. Barry Schwartz wrote it up at Search Engine Roundtable on October 14, and the short of it is that the UTM tag OpenAI added to ChatGPT links back in June 2025 (Search Engine Roundtable) appears to be gone.
Cedric Huesler at Adobe dug into the mechanics. According to his post, GPT-6 opens links in a way that strips the referrer (the noopener,noreferrer setting, for the technical folks) and drops the ?utm_source=chatgpt.com parameter (Huesler on LinkedIn). A visit with no tag and no referrer has nowhere to go in your analytics except Direct. Sébastien Billard reported a sharp drop in human ChatGPT traffic starting around October 9 as well (Billard on X).
Nobody from OpenAI has said whether this was on purpose, and Schwartz said as much in his write-up. I'd guess it's a side effect of how the new interface opens links more than a decision to hide traffic from publishers, though I could be wrong about that, and for your reporting it doesn't matter much since the tag is gone either way.
What the early numbers show
Huesler put numbers on it. He looked at roughly 500 pages where ChatGPT drove at least 5% of visits and compared October 9 to 13 with the week before. Tracked ChatGPT visits fell by about 31,000 a day. Direct visits on those same pages rose by about 26,000 a day. Search traffic on the pages dropped 7%, which he says matched what was happening everywhere else, so search can't explain the shift. Direct went up on two out of every three pages that lost ChatGPT traffic (Search Engine Roundtable).
GPT-6 and your analytics
ChatGPT visits fell, and Direct rose on the same pages
About 500 pages where ChatGPT drove at least 5% of visits, October 9 to 13 compared with the week before.
Source: Cedric Huesler (Adobe), early sampled data, as reported by Search Engine Roundtable, Oct 14, 2026, and Huesler on LinkedIn. "Looks new" means the visitor saw the site's consent banner.
There's one more detail in there I liked a lot. Huesler checked how many of the new Direct visitors on mobile saw the site's cookie consent banner, which is a decent stand-in for "first time on this site." Before the change, 36% of Direct visitors saw it. After, 44% did. ChatGPT visitors see it 55% of the time. So the Direct bucket started behaving more like ChatGPT traffic, which is what you'd expect if a big chunk of it is ChatGPT traffic wearing a different label.
He's careful to say this is early, four days of sampled data, and that only pages where ChatGPT was a big share of traffic can show the signal clearly (Huesler on LinkedIn). I'd treat the exact numbers as provisional, even though the direction is pretty hard to argue with.
Why your AI Assistant channel will undercount
Google added an AI Assistant channel to GA4's default channel group on May 13 of this year (Search Engine Journal). Google's definition says a visit lands there when the medium is "ai-assistant" or when the referrer matches its list of AI assistants (Google Analytics Help). A visit with no referrer and no tag matches neither rule.
So if your dashboard for the CMO has a nice tidy line for AI Assistant traffic, that line is going to drop this month, and somebody is going to ask whether the AI work stopped working, which it probably didn't. The line was already missing things before this, too. Perplexity isn't on Google's list and still lands in Referral, AI Overviews and AI Mode get counted as Organic Search, and a lot of traffic from AI mobile apps already arrived with no referrer at all (Search Engine Journal).
I wrote earlier this year that AEO is measurable if you measure the right thing, and I still think so. This week is a reminder that the referral report was never the right thing to lean on by itself.
The clicks were always the small part
If you run marketing, this part matters more to you than the analytics plumbing. The click that didn't happen has been the bigger story of AI search all year, and the research on what happens after an AI answer keeps pointing the same way.
Similarweb looked at what people did in the seven days after an AI engine recommended a brand. When AI recommended Capital One, 14.2% of those users went on to visit Capital One, and 3.8% visited American Express. When it recommended Kayak, 12.0% visited Kayak and 3.4% visited Skyscanner (PPC Land on Similarweb's State of Ecommerce 2026). Rand Fishkin's read of the same research was that search visits ran about 15% lower and direct visits ran higher for people who'd been influenced by AI (SparkToro).
What happens after an AI answer
When AI names a brand, people go find it
Share of users who visited each site within seven days of an AI engine recommending one of the two brands. Most of these visits never show up as AI referrals.
The recommended brand got the bigger share in every pairing, by as much as 3.7 times (Capital One over American Express).
Source: Similarweb, State of Ecommerce 2026 (Sept 2026), 7-day visit rates, U.S. desktop, July to December 2025, as reported by PPC Land. See also SparkToro.
Now apply that to attribution. Somebody asks ChatGPT for a credit card, sees your name, closes the app, and types your URL into a browser later. That visit was never going to show up as ChatGPT traffic, tag or no tag. The GPT-6 change just pushes even the people who did click into the same bucket.
How to track ChatGPT traffic from here
Number one, mark the date. Put an annotation on October 7 in GA4 (or whatever you use) so nobody reads the drop as a performance problem six months from now. It's a small thing that saves an awkward meeting later.
Two, build a baseline for Direct traffic on your AI-cited pages. Pull the list of pages that got meaningful ChatGPT referrals from, say, July through September. Then watch Direct traffic to those specific pages, and especially to deep pages nobody would type from memory. A product comparison page or a pricing FAQ getting a pile of new "direct" visits is a pretty good tell.
Three, fix your AI channel group instead of trusting the default. Search Engine Journal published a regex-based custom channel group that catches Perplexity, Claude, Copilot and the rest, and it's worth copying and reviewing every quarter (Search Engine Journal). It won't catch the untagged ChatGPT visits, but it'll stop you losing the ones that are still labeled.
Four, ask people. A "How did you hear about us?" field on your forms, with ChatGPT and other AI assistants as options, is old-fashioned and it works. Self-reported attribution isn't perfect, and it's still better than a Direct bucket that can't tell you anything.
Then last, measure the answers along with the clicks. Kevin Indig and Amanda Johnson's planning piece in Growth Memo makes a case I agree with, which is to track exposure (AI visibility and citations), behavior (branded search and direct visits) and business outcomes as one set, and to investigate when they disagree (Growth Memo). They also note that 91% of citations show up in only one of ChatGPT, Perplexity or AI Overviews, so you need to check each engine separately.
Why the answers matter more than the clicks
When I ran the AI Verdict Study, the same buyer asking the same question got a recommendation on some runs and a rejection on others in 21 of 72 brand and engine pairings, or 29%. If an engine is telling some of your buyers yes and some of them no, no traffic report on earth is going to show you that. You have to go ask the engines. I wrote more about where the engines go to make up their minds, and about how long ChatGPT citations last once you have them, if you want to go further down this road.
If you want a quick read on where you stand, the free AI Visibility Quick Check will show you whether the engines mention you and what they say when they do. The thinking behind why some brands get recommended and others don't is the whole subject of my book, Explainable, if you want the longer version.
Self-check, about a minute
Can you still see your AI traffic?
Six questions. Answer for your own site and analytics setup, and you'll get a score with the fixes in the order I'd do them.
0 of 6 answered
What I'd tell the boss
If someone above you sees the ChatGPT line drop this month, my suggestion is to say plainly that the tracking changed, show them the October 7 annotation and the Direct baseline on your AI-cited pages, and then steer the conversation toward what the engines are saying about you, since that's what decides whether buyers ever look for you. The referral count was always a partial view, and as of this month it's a smaller one.
I'd love to hear what you're seeing in your own analytics since GPT-6 shipped, especially if your numbers don't match Huesler's. Drop me a note or find me on LinkedIn.
Jarred Smith is the author of Explainable: Why AI Recommends Some Brands & Ignores Others, an Amazon bestseller on AEO, GEO, and SEO. He's a marketing leader with nearly 20 years of experience across healthcare, public media, retail, and environmental services. Find him at jarredsmith.com.