ChatGPT Bug Sent AI Referral Traffic to Direct After GPT-6

Marketers who were watching website visits from AI tools noticed a big drop in October. Several analytics tools showed fewer visitors coming from ChatGPT after GPT-6 launched. But the drop may have been caused by a change in how visits were counted, rather than fewer people using ChatGPT to visit websites.

According to Search Engine Land, OpenAI appears to have removed UTM tracking tags from links in ChatGPT. Without these tags, visits from ChatGPT may be counted as direct traffic instead of referral traffic. This makes it harder to tell where website visitors actually came from.

What Seems to Have Changed

The change was noticed after GPT-6 launched in October 2026. Links in ChatGPT began appearing without the utm_source=chatgpt.com tag that OpenAI had added in June. This tag helped website owners easily see which visitors came from ChatGPT.

Cedric Huesler from Adobe found another important clue. He noticed that ChatGPT was opening links in a new way that hid information about where visitors came from. Search Engine Roundtable shared Huesler's findings and explained that GPT-6 appeared to use a window.open command with the settings no opener and no referrer. It also removed the UTM tag. Together, these changes meant that websites received visits without either type of tracking information.

According to Relevant Audience, OpenAI's ChatGPT release notes said that GPT-6 began rolling out to paid users on October 7. Because the tracking change happened around the same time, marketers wondered whether the new model was sending fewer visitors to websites.

Barry Schwartz from Search Engine Roundtable reported the same issue on October 14. He said the change seemed to have happened around the GPT-6 release one week earlier. At that time, it was not clear whether OpenAI had made the change on purpose.

Why Missing Tracking Tags Make Visits Look Direct

To understand the problem, it helps to know how website analytics tools identify where visitors come from. A UTM parameter is a small label added to the end of a website link. It tells analytics tools which source, type of traffic, or marketing campaign brought someone to the website. A referrer is another piece of information sent by the browser. It identifies the page the visitor came from.

Usually, either one of these pieces of information is enough to identify where a visitor came from. But when both are missing, tools such as Google Analytics usually count the visit as direct traffic. This category includes people who type a website address into their browser, use a saved bookmark, or arrive without any information showing where they came from.

As a result, someone clicking a link in ChatGPT can look exactly like someone who typed the website address themselves. The visitor has not changed. Only the way the visit is labeled has changed.

What Huesler Found in the Data

Huesler studied about 500 website pages where ChatGPT had previously brought in at least 5% of all visits. He compared traffic from October 9 to October 13 with traffic from the previous week.

The numbers showed a clear pattern.

• ChatGPT traffic: Fell by about 31,000 visits per day.
• Direct traffic: Rose by about 26,000 visits per day on those same pages.
• Search traffic: Fell by 7%, which matched wider trends in search traffic.

About two out of every three pages that lost ChatGPT referral visits also gained direct visits. In the previous week, the chances of seeing this pattern had been about 50-50. This suggests that the increase in direct traffic was a real, noticeable change. Huesler also found that the new direct visitors behaved much like the people who had previously arrived through ChatGPT. This suggests that many of them came from ChatGPT but were no longer being identified that way.

There was still a difference between the two numbers. ChatGPT traffic fell by about 31,000 visits per day, while direct traffic rose by about 26,000. That leaves a difference of around 5,000 visits. Some of those visits may have truly disappeared, while others may have been counted under different traffic categories.

Why the Numbers Need Careful Attention

Search Engine Land explained that the findings point toward a problem with how traffic sources were recorded, rather than necessarily showing a real drop in visitors from ChatGPT. The publication also said that OpenAI had not confirmed whether the change was intentional. The study could not prove that every missing ChatGPT referral had disappeared for the same reason.

Other analysts looked at the issue too. Search Engine Journal reported that the team behind LLM Pulse did not see an increase in direct traffic large enough to match the drop in ChatGPT referrals. Their data combined visits from many sources, so they could not clearly separate visits that had truly disappeared from visits that were simply no longer labeled as coming from ChatGPT. They also could not identify one traffic category that accounted for all the missing OpenAI visits.

This means Huesler's findings provide an important clue, but they do not tell the whole story. His study focused on pages that received a noticeable share of their traffic from ChatGPT. Websites that normally get fewer visitors from ChatGPT might show a different pattern.

What Happened with the Fix

Later reports showed that the story did not end with the original traffic drop. Search Engine Roundtable reported that ChatGPT had started adding the utm_source=chatgpt.com tag to its links again. French publication Siecle Digital said the tag returned on October 15.

According to Search Engine Journal and Dr. Web, Huesler also said that OpenAI had confirmed the problem to him. Several sources described it as a tracking issue that lasted only a short time. However, these publications noted that OpenAI had not released an official explanation. Its GPT-6 launch announcements also did not discuss the missing UTM tags or referrer information.

There is an important detail for anyone looking back at their website data. Dr. Web explained that Google Analytics records where a visit came from when the visit happens. Fixing the links later does not automatically change old records. This means visits from the affected days will probably stay labeled as direct traffic.

Why This Matters for Marketers

AI assistants are becoming a more popular way for people to discover and visit websites. That makes it increasingly important to know which visitors come from these tools. Marketers use this information to decide what content to create, which search engine optimization (SEO) work to focus on, and what results to report to people involved in the business. If visits are counted incorrectly, a marketing team might think ChatGPT has suddenly stopped bringing visitors, even when it has not.

Search Engine Land explained that a large drop in ChatGPT referrals, together with a big rise in direct traffic, can happen simply because tracking stops working correctly. It also said that changes to how AI assistants open outside links can make it harder to measure traffic, even when the actual number of visitors stays the same.

Relevant Audience pointed out a similar problem with direct traffic. It said that the rise in direct visits could be explained by an October-related tracking effect, rather than more people knowing the brand or seeing promotions outside the internet. This shows why marketers should not automatically assume that an increase in direct traffic means more people are choosing to visit the website on their own.

The situation also shows how easily traffic reports from AI assistants can be affected. If another company changes how its links open in people's browsers, website analytics might suddenly show a large rise or fall in traffic even when visitor numbers have not really changed. This means marketers depend partly on other companies to keep AI referral tracking working correctly.

What to Check in Your Analytics

There are several ways to find out whether your website really lost visitors or whether the visits were simply counted under the wrong category.

First, compare ChatGPT referral traffic and direct traffic on pages that usually receive visitors from ChatGPT. Look for pages that normally get a steady number of ChatGPT visits. If ChatGPT referrals suddenly fall while direct visits rise, that may be the same pattern Huesler found.

Next, check the total number of visits. If the total stays about the same and only ChatGPT referrals fall, the problem may be how the visits were labeled. But if total visits also fall, there may be a bigger problem. Check all traffic sources before deciding what caused the change. Search Engine Journal also recommends checking your homepage. In its analysis of links shared in ChatGPT messages, more than half of the direct links pointed to a website's homepage.

For advertising, Search Engine Journal recommends checking how traffic from ads placed in ChatGPT is recorded. The publication found that clicks on ChatGPT Ads were not affected in the same way when advertisers had added their own UTM tracking tags to the links.

Finally, Search Engine Journal recommends adding a note to your analytics data for the period around October 9 to October 15. It suggests calling this period "chatbot migration." This will help people reviewing the data later understand why traffic numbers may have suddenly risen or fallen. Search Engine Journal also reported that Gemini had updated its links to include UTM tracking tags about two weeks earlier.

Frequently Asked Questions

Q1: What did ChatGPT change?
According to Search Engine Land, OpenAI appears to have removed the utm_source=chatgpt.com tracking tag from links in ChatGPT. The links also seemed to open in a way that hid information about which website sent the visitor.

Q2: Why did this make visits look like direct traffic?
When a visit has no UTM tag and no referrer information, Google Analytics usually counts it as a direct visit. The person visiting the website has not changed, but the information showing where they came from is missing.

Q3: When did the problem begin?
The problem was noticed after GPT-6 launched in October 2026. Search Engine Roundtable said it appeared to have started about one week before its October 14 report.

Q4: What did Cedric Huesler discover?
Huesler studied about 500 pages where ChatGPT normally brought in at least 5% of visits. Between October 9 and October 13, visits labeled as coming from ChatGPT fell by about 31,000 per day. At the same time, direct visits rose by about 26,000 per day.

Q5: Did the number of visitors from ChatGPT really fall?
Not necessarily. Huesler's findings suggest that many visits were counted under the wrong traffic category rather than actually disappearing. However, his study could not prove that every missing ChatGPT referral had the same cause.

Q6: Has OpenAI fixed the problem?
Later reports suggest that ChatGPT has started adding UTM tracking tags to its links again. Huesler said OpenAI confirmed the problem to him, but the company had not yet published a detailed explanation.

Q7: Will my old analytics records be corrected?
Probably not. Google Analytics records the source of a visit when it happens. Because of this, visits affected by the tracking problem will most likely remain labeled as direct traffic.

Q8: How can I find out whether my website was affected?
Compare ChatGPT referral visits and direct visits on the same pages during early and middle October. Also check the total number of visits and conversions, such as purchases or sign-ups, to see whether your website actually lost visitors or important actions.

Q9: Were ChatGPT Ads affected?
Search Engine Journal reported that clicks on ChatGPT Ads should not have been affected when advertisers added their own UTM tracking tags to the links.