Your ChatGPT Traffic Didn’t Disappear Last Week. Your Analytics Just Lost Track of It.

If you opened Google Analytics last week and saw your ChatGPT referrals fall off a cliff, you probably had the same thought most of us did.

Did something break? Did we lose our visibility? Did ChatGPT stop recommending us?

The short answer is no. Most of those visitors were still arriving. Your analytics just stopped recognising where they came from, and quietly dropped them into the “Direct” bucket instead.

Here’s what happened, how to check whether it hit your site, and what to change so the next glitch doesn’t send you into a week of unnecessary panic.

What actually happened between 8 and 15 October

OpenAI rolled out GPT-6 on 7–8 October. Within a day, a lot of sites saw ChatGPT-attributed traffic drop sharply from 9 October.

The cause wasn’t demand. After the rollout, ChatGPT began opening links in a new tab with a setting that withholds the referrer, and it stopped adding the utm_source=chatgpt.com tag its citations had carried since ChatGPT search launched in 2024. Without either signal, GA4 has no way to know a visit came from ChatGPT, so it calls it Direct. Search Engine Journal covered the analysis as it unfolded.

The numbers made it fairly obvious. Across a panel of more than 500 pages, reported ChatGPT traffic fell by roughly 31,000 visits a day while Direct traffic rose by about 26,000 a day. Visits don’t vanish and reappear in another channel by coincidence.

OpenAI reportedly confirmed the issue on 14 October and a fix started rolling out on 15 October. By that morning, one analysis found tagged ChatGPT clicks back at roughly 95–100% of the same hour the week before. Marketing4eCommerce has a good summary of the timeline. The fix may not have reached every account evenly, so some visits may still be arriving without referrer data.

How to check whether your site was affected

This takes about ten minutes.

1. Compare the two channels side by side. In GA4, open Traffic acquisition and set the date range to roughly 1–20 October. Add a comparison for sessions where the source contains “chatgpt”. If ChatGPT dips from the 9th while Direct rises by a similar amount over the same days, you were hit.

2. Look at where those “Direct” visits landed. Real direct traffic mostly lands on your homepage. People type your name or use a bookmark. If “Direct” suddenly includes a lot of visits to deep blog posts, pricing pages or comparison pages, those are almost certainly AI referrals in disguise. Nobody types out a 70-character blog URL by hand.

3. Check that the recovery happened. From the 15th, your ChatGPT line should climb back towards normal. If it hasn’t, keep an eye on it for another week before drawing conclusions.

Clean up the data so this week doesn’t mislead you later

The real risk isn’t the dip itself. It’s someone looking at a quarterly report in January, seeing AI traffic “collapse” in October, and making a budget decision based on a bug.

• Add an annotation. Mark 8–15 October in your reporting tool as “ChatGPT referrer/UTM bug — AI visits misattributed to Direct”. Future you will be grateful.
• Don’t judge AI search work on that week. If you’ve been investing in getting your brand cited by AI tools, exclude those dates when you compare performance.
• Build an “AI assistants” channel group. In GA4 you can create a custom channel that matches sources like chatgpt.com, perplexity, copilot and gemini. It makes AI traffic visible in one line instead of scattered across Referral.

The bigger lesson: referrer data is fragile

This wasn’t the first time AI traffic got miscounted, and it won’t be the last. We wrote about the wider problem in Your GA4 Says “Direct.” Your Customers May Have Found You on ChatGPT: a lot of AI-driven visits never carried a clean referrer in the first place, especially from apps.

So treat your analytics as one witness, not the only one. A few cheap habits help a lot:

• Add a “How did you hear about us?” field to your enquiry and booking forms. Make it optional, but include “ChatGPT / AI assistant” as an answer.
• Ask on sales calls. You’ll be surprised how often people say “I asked ChatGPT for agencies and you came up.”
• Track leads and revenue by month, not just sessions. If enquiries held steady while “ChatGPT traffic” crashed, the traffic didn’t really crash.

So, should you worry?

Probably not about last week. The visits were real, they were still coming, and the tagging is mostly back.

What’s worth worrying about is building decisions on a single number from a single tool. If you’re not sure your tracking is set up properly, our free website audit checks the basics in about 30 seconds, and we’re happy to look at your analytics setup on a call.

Last week was a reminder that the customer didn’t change. The plumbing did.