Why Your ChatGPT Traffic Fell in GA4 After October 8 (and How to Check)

If your ChatGPT traffic in GA4 suddenly fell around October 9, don't rewrite your AI search strategy just yet.

Over the past week, several SEO and analytics people shared the same pattern: ChatGPT referrals dropped sharply, but total traffic barely moved. The most likely explanation isn't that people stopped clicking. It's that ChatGPT stopped labelling its links for a few days, so analytics tools counted those visits as Direct.

Cedric Huesler, a product leader at Adobe, wrote on LinkedIn on October 15 that OpenAI told him it was a bug, and that a fix started rolling out early that day. OpenAI hasn't said anything about it publicly yet.

What people saw

The numbers came from different companies, using different data, but they point the same way:

Sources: Ethan Smith, LLM Pulse, Cedric Huesler and Profound. Most of these companies sell AI search tracking tools, and none of the posts gives a full sample size.

The Adobe numbers are the most telling to me. When one channel loses about 31,000 visits a day and another gains about 26,000 on the same pages, that looks like visits being relabelled, not disappearing.

Why ChatGPT visits turned into Direct

This comes down to how GA4 knows a visit came from ChatGPT. There are two signals:

• The referrer. Your browser normally tells the site which page sent the visitor. But according to Huesler, ChatGPT opens links in a new tab in a way that withholds the referrer, so most ChatGPT clicks don't carry one at all.
• The UTM tag. ChatGPT has been adding utm_source=chatgpt.com to its links. When the referrer is missing, this tag is often the only way GA4 can tell the visit came from ChatGPT.

From around October 8, the inline links inside ChatGPT's answers stopped carrying that tag. No referrer and no tag means GA4 has nothing to go on, so it files the visit as Direct.

Not every link was affected. Links shown in the sources list and in shopping results still had the tag, so some ChatGPT traffic still showed up correctly. That's why the drop looked partial rather than total.

Mobile was hit hardest. On the sites Graphite tracks, Ethan Smith said the tag was the only thing identifying 95% of ChatGPT visits from mobile.

Was GPT-6 the cause?

The timing matches. OpenAI rolled out GPT-6 to paid plans on October 7 and to free plans on October 8. Profound linked the missing tags to the new Intelligent UI answer format that came with it.

LLM Pulse made a useful point here: they saw no drop on October 7, when only paid users had the new model. The drop started when free users got it. Still, matching dates isn't proof, and OpenAI hasn't explained what changed.

Important note

The "it was a bug" part comes from Huesler, who says OpenAI confirmed it to him privately. OpenAI hasn't published a statement, and its release notes for October don't mention link tracking.

Huesler reported on October 15 that tagged ChatGPT clicks had recovered to almost the same hourly levels as the week before (95–100%) on the domains he monitors. Others hadn't confirmed a fix when they published. Check your own data from October 15 onward before assuming everything is back to normal.

Also, a fixed tag only helps from now on. The visits from October 8 to 15 that GA4 filed as Direct won't move back to ChatGPT.

How to check your own site in GA4

You don't need any special tool for this. Here's how I'd check:

1. Add an annotation. Mark October 8 to 15 in GA4 as "ChatGPT UTM tag missing", so nobody on your team reads that dip as a real loss later.
2. Find your ChatGPT landing pages. In Explore, build a free-form report with Landing page as the row and Session source as a filter containing "chatgpt". Use the two weeks before October 8. These are the pages ChatGPT usually sends people to.
3. Compare ChatGPT and Direct on those pages. For the same landing pages, look at Direct sessions from October 8 to 15 against the week before. If Direct rose about as much as ChatGPT fell, the traffic didn't go anywhere.
4. Check deep pages, not just the homepage. People rarely type a long blog URL by hand. A jump in Direct traffic to deep pages that ChatGPT links to is a strong sign of mislabelled visits.
5. Look at new users. Huesler saw a higher share of first-time visitors among Direct traffic on mobile during the gap. A spike in new users arriving as Direct fits the same story.
6. Confirm the recovery. Compare ChatGPT sessions from October 16 onward with the weeks before October 8.

What this teaches us about AI traffic

This episode shows how fragile AI referral tracking is. ChatGPT traffic only showed up properly in GA4 because ChatGPT chose to add a UTM tag. When that changed for a few days, a big share of the visits vanished from reports overnight.

GA4's newer AI Assistant channel depends on the same signals. If an AI tool doesn't pass a referrer or a tag, GA4 can't tell where the visit came from.

So a few habits are worth building:

• Report ChatGPT and Direct together for pages that get AI traffic, so a labelling change can't hide the real trend.
• Track AI visibility separately from clicks. Whether ChatGPT mentions and cites you is a different question from whether its links are tagged.
• Ask customers how they found you. A simple "how did you hear about us" question on forms catches AI discovery that analytics misses.
• Don't make big strategy calls on a few days of data, especially when a new model or interface has just launched.

What we don't know yet

• Whether OpenAI will confirm the bug publicly or explain what changed.
• Whether every link type has its tag back, including inline links in the new answer formats.
• How much of the drop was mislabelled visits and how much was real. With Intelligent UI changing how answers look, some clicks may genuinely have dropped too.

My takeaway

If your ChatGPT numbers fell off a cliff on October 9, most of that traffic was probably still arriving. It was just wearing a Direct label.

Annotate the gap, compare ChatGPT and Direct side by side, and check that your numbers recover after October 15. And treat this as a reminder that AI referral data is only as reliable as the tags AI tools choose to send.

Did your ChatGPT traffic drop around October 9? And has it come back since the fix? I'd love to hear what your data shows.

FAQ

Why did ChatGPT traffic drop in GA4 in October 2026?

From around October 8, many links in ChatGPT's answers stopped carrying utm_source=chatgpt.com. Because ChatGPT links usually don't pass a referrer, GA4 had no way to identify those visits and counted them as Direct.

Did ChatGPT really send less traffic?

Most of the evidence points to a labelling problem, not a real loss. Adobe's data showed Direct traffic rising almost as much as ChatGPT traffic fell on the same pages. Some real change is still possible as new answer formats roll out.

Has OpenAI fixed it?

Cedric Huesler of Adobe says OpenAI confirmed a bug to him and that a fix started rolling out on October 15. OpenAI hasn't made a public statement, so check your own data from October 15 onward.

Can I recover the lost ChatGPT data?

No. Visits that GA4 already recorded as Direct stay that way. Add an annotation for October 8 to 15 and compare ChatGPT and Direct together for that period.

Sources: Search Engine Journal (Matt G. Southern); Cedric Huesler on LinkedIn (Oct 13, updated Oct 15) and follow-up (Oct 15); Ethan Smith on LinkedIn; LLM Pulse; Profound; OpenAI ChatGPT release notes.