ChatGPT Traffic Dropped? Why It's Now Showing as Direct
From 9 October, ChatGPT referrals in Google Analytics dropped by as much as two thirds. People are still clicking. ChatGPT has just stopped telling your analytics where they came from. Here’s what happened, what the numbers show, and why it matters far less if you measure AI search by visibility rather than clicks.
Did ChatGPT Stop Sending Traffic?
No, it stopped labelling it. Since GPT-6 rolled out on 7 and 8 October, most links in ChatGPT answers arrive without a tag or a referrer, so Google Analytics counts those visits as Direct. A smaller share may be genuinely lost, because GPT-6 answers more questions without anyone needing to click.
The Label Was Doing All the Work
Since June, ChatGPT had been adding utm_source=chatgpt.com to the links in its answers. That small tag is how most analytics tools knew a visit came from ChatGPT. The referrer, the usual way a browser says which site sent someone, was only showing up on around one click in eight. So the tag was carrying almost the whole job.
With GPT-6, links open in a way that drops both. Cedric Huesler at Adobe spotted the change in the code. Profound tied it to ChatGPT’s new interactive answers: as they grew to around a third of all answers, the share carrying a tagged link fell from 98% to 66%. OpenAI has not commented or included any information with the rollout.
?utm_source=chatgpt.comnoreferrer and no tagWhat the Numbers Show
For the three weeks before GPT-6, tracked ChatGPT visits bounced between roughly 80 and 110 on Adobe’s index. That’s normal weekly ebb and flow. Then, over the three days the rollout took, the line fell to 31 and stayed there. Nothing about demand changes that fast. A measurement change does.
The more telling part is where those visits went. On around 500 pages where ChatGPT normally drove at least 5% of traffic, most of what ChatGPT lost reappeared as Direct. The new Direct visitors also skew towards first-time visitors, which is what you’d expect from people arriving off an AI answer rather than regulars typing in your address.
Most of what ChatGPT "lost" turns up as Direct on the same pages.
pages that lost ChatGPT visits gained Direct visits (261 of 387). The week before it was a coin flip.
of mobile Direct visitors on these pages are new to the site. For known ChatGPT visitors it is 55%.
Around 500 pages where ChatGPT drove 5% or more of visits.
Is Any of It Genuinely Lost?
Zero Click traffic has been a buzzword in the industy for a while and GPT-6 answers more questions with maps, widgets and small built-in tools, so fewer people need to click through at all. LLM Pulse saw ChatGPT referrals fall 46% across the sites it monitors while Perplexity and Claude held steady, which fits both explanations. Gemini still tags its links, so if your Gemini numbers look normal, that’s another sign the issue is labelling rather than demand.
Clicks Were Never the Whole Picture
Think of an AI recommendation like word of mouth. When a friend tells someone to use you, nobody tracks that conversation, but it’s still the reason they get in touch. ChatGPT works the same way. The moment that matters is when it names you as the answer. The click, if it happens at all, comes after.
That’s why counting ChatGPT referrals was always a partial measure. It only captured the people who clicked, and only when ChatGPT said where they came from. This month showed how quickly even that can disappear.
How We Measure AI Search Instead
In our reports, ChatGPT traffic has never been the headline number. It moves with every product update, and it only tells you about the end of the journey. What we lead with is Share of Voice.
Each month we put the questions your customers actually ask to ChatGPT, Gemini, Google AI Mode and Perplexity. We record how often your brand is mentioned and recommended, and how that compares with the competitors you’re up against. Because it looks at the answers themselves rather than what happens after them, this month’s change has no effect on it at all. Your trend line carries on exactly as before.
Counts who clicked, and only when ChatGPT says so
Broke overnight on 9 October
Misses people who got their answer without clicking
Can't show how you compare with competitors
Measures how often you're named and recommended
Unaffected by tags, referrers or analytics changes
Captures visibility whether or not anyone clicks
Shows where you stand against named competitors
Referral traffic still has a place in our reporting. Where ChatGPT visits are now hidden, we estimate them conservatively, counting around a quarter of the Direct traffic on the pages AI search sends people to rather than assuming all of it is ChatGPT. More important are the enquiries and sales that come in. Visibility tells you whether you’re being recommended. Enquiries tell you whether it’s paying off. Together they give a picture no single tracking tag can.
Checking Your Own Analytics
If you look after your own GA4, four things will keep October’s numbers from misleading you.
01
Mark 9 October as a break
ChatGPT figures before and after that date aren’t measuring the same thing, so month-on-month comparisons will look worse than reality.
02
Don't treat Direct as all ChatGPT
Direct also covers bookmarks, typed visits and links from emails and apps. On the pages AI search sends people to, only part of the rise is likely ChatGPT, so estimate it conservatively.
03
Don't build a filter to fix it
No tag and no referrer means there’s nothing for a GA4 rule to catch. The hidden visits can be estimated, not recovered.
04
Ask people how they found you
A ChatGPT option on your enquiry form costs nothing and doesn’t depend on any tracking, whatever OpenAI changes next.
ChatGPT Traffic and GA4: Quick Answers
Why did my ChatGPT traffic drop in October 2026?
Since GPT-6 rolled out on 7 and 8 October, most ChatGPT links no longer carry the utm_source=chatgpt.com tag or pass a referrer. Google Analytics can’t tell those visits came from ChatGPT, so it records them as Direct. Some questions are also now answered inside ChatGPT without a click.
Why is ChatGPT traffic showing as Direct in GA4?
GA4 identifies a visit’s source from the link’s tag or the referrer. When both are missing, it has nothing to go on and files the visit under (direct) / (none), the same as someone typing your web address.
Did OpenAI remove the UTM tags on purpose?
Nobody knows yet. OpenAI hasn’t commented, and Cedric Huesler, who first published the data, has since said it may be a bug. It could be fixed without warning.
Does this affect Share of Voice reporting?
No. Share of Voice measures how often your brand is named and recommended in AI answers by testing the questions directly, so it doesn’t rely on clicks, UTM tags or referral data.
Do Digital Landscope reports still include ChatGPT traffic?
Yes, as a supporting signal alongside Share of Voice and your enquiries. From October we mark 9 October as a tracking break and estimate the hidden ChatGPT visits conservatively, counting around a quarter of the Direct traffic on the pages AI search sends people to, not all of it.
Want to Know How Visible You Are in AI Search?
Book a free consultation and we’ll show you where you stand on Google and AI search, whatever your analytics say this month.
Sources: Search Engine Roundtable · Cedric Huesler, Adobe · Profound · LLM Pulse · OpenAI