ChatGPT appears to drop UTM tags and referrer, sending clicks to Direct (unconfirmed)

ChatGPT appears to have stopped adding utm_source=chatgpt.com to the links it shows and stopped passing a referrer, so clicks from ChatGPT appear to be landing in analytics as Direct traffic, Search Engine Roundtable reported on 14 October 2026. The change seems to have started with OpenAI's GPT-6 rollout about a week earlier. OpenAI has not confirmed it: the report contains no statement from OpenAI, and Search Engine Roundtable says it is not sure the change was intentional. The supporting numbers come from one practitioner, Cedric Huesler of Adobe, who analysed about 500 pages and found ChatGPT visits falling by about 31,000 a day while Direct rose by about 26,000 a day on the same pages.

What Search Engine Roundtable reported on 14 October 2026

Barry Schwartz wrote in Search Engine Roundtable on 14 October 2026 that OpenAI appears to have removed or replaced the UTM parameters it had been placing on links in ChatGPT answers. That makes it harder for a site owner to tell whether a visit came from ChatGPT. According to the article, OpenAI expanded UTM tracking on ChatGPT links in June 2025, and the parameters disappeared in October 2026. Schwartz dated the start of the change to the GPT-6 release, roughly a week before his post, and said he was not sure whether it was directly intentional. He credited Chee Lo with the tip.

The article is a report of what practitioners are seeing in their data. It is not an announcement. Until OpenAI says something, the cause, the scope and whether it is permanent are all unknown.

The mechanism Cedric Huesler describes

Cedric Huesler of Adobe posted on LinkedIn that GPT-6 now opens links with a JavaScript window.open call that uses the "noopener,noreferrer" settings, and that it drops the utm_source=chatgpt.com tag from the URL. His summary, quoted by Search Engine Roundtable, is that there is "no tag and no referrer", so analytics files the visit under Direct.

Some background on why that combination matters, which is general web behaviour rather than anything in the source. Analytics tools such as Google Analytics 4 work out where a visit came from using two main signals. The first is campaign parameters in the landing URL, such as utm_source. The second is the referrer, the address of the previous page, which the browser normally sends along with the request. The "noreferrer" instruction tells the browser not to send that referrer. If the UTM tag is also gone, the visit arrives with neither signal, and a session with no source information is reported as direct traffic. In GA4 that shows up as (direct) / (none).

In the comments on his post, Huesler noted that shopping links still carry the parameter. In an update to the post, he wrote that sources told him on 14 October it might be a bug, and that he would update when he knew more. Neither Huesler nor Search Engine Roundtable reported an explanation from OpenAI.

Huesler's analysis of about 500 pages

Huesler looked at roughly 500 pages where ChatGPT had driven 5% or more of visits, comparing 9 to 13 October 2026 with the week before. These figures come from his LinkedIn post as summarised by Search Engine Roundtable. They are one practitioner's analysis, not a published study, and he described his own data as covering only four days, sampled, and showing the signal only on pages where ChatGPT was a large share of traffic.

His findings:

• ChatGPT-attributed visits fell by about 31,000 a day, while Direct visits rose by about 26,000 a day on the same pages.
• Search traffic to those pages fell 7%, in line with what he saw elsewhere, so a general traffic drop does not explain the pattern.
• Direct rose on two in three pages that lost ChatGPT traffic. The week before, it had been roughly a coin flip. Search rose on only 45% of the same pages.
• On mobile, the share of new Direct visitors who saw the consent banner, a sign of a first visit, went from 36% to 44%. For ChatGPT visitors the figure is 55%. His reading is that the new Direct visitors behave like ChatGPT visitors.

Huesler reads these numbers as ChatGPT visitors still arriving but being labelled as Direct; they are aggregate movements on the same pages and do not trace individual visits. The 26,000 a day gained by Direct does not fully match the 31,000 lost by ChatGPT, and the post does not account for the gap. A four-day window is short, and this should be read as an early signal rather than a measured effect.

What Sébastien Billard reported

Search Engine Roundtable also cited Sébastien Billard, who posted on X that he had seen a sharp drop in ChatGPT human traffic since 9 October, along with a drop in ChatGPT-User bot traffic and a rise in OAI-SearchBot traffic since 7 October. He said he would look into the UTM parameter. His observations are about his own sites and were not quantified in the article.

Timeline and figures at a glance

The table below lists the dated facts and figures in the Search Engine Roundtable report and Huesler's LinkedIn post, plus the GPT-6 date from OpenAI's ChatGPT release notes. None of the traffic figures come from OpenAI.

Why ChatGPT reports may look wrong for October

If the reports are right, any dashboard that counts ChatGPT traffic by the chatgpt.com source or referrer will under-count from early October. That includes GA4 custom channel groups built to separate AI referrals, Looker Studio reports that filter on source, and monthly client reports that compare ChatGPT sessions month on month. A drop in "ChatGPT traffic" for October could be a labelling change, not fewer people clicking.

The reverse matters too. Direct traffic may look stronger in October without any change in brand demand or offline activity. Anyone who reads a Direct increase as a sign of brand strength should check whether it lines up with the pages ChatGPT used to send visitors to.

This also affects work on AI search visibility. Teams trying to show that ChatGPT SEO work is producing visits have relied on the utm_source=chatgpt.com tag to count them. If that tag is gone, the evidence gets weaker even when the clicks keep coming.

What to check in GA4 (analysis)

The steps below are Relevant Audience analysis based on Huesler's suggestions and standard GA4 reports. They are not instructions from OpenAI or Search Engine Roundtable.

1. List the landing pages that received chatgpt.com sessions in September. Huesler's own data suggests the signal is clearest on pages where ChatGPT was a meaningful share of visits, so start there.
2. For those pages, compare (direct) / (none) sessions before and after about 8 October. A Direct rise on the same pages that lost ChatGPT sessions matches the pattern he describes.
3. Look at deep pages. Direct visits usually land on the home page or pages people bookmark. Direct visits to a specific article or help page that ChatGPT used to link to are more likely to be relabelled ChatGPT clicks. Huesler suggests watching Direct on those pages as a temporary proxy.
4. Check new-user share and engagement for the new Direct sessions against the ChatGPT sessions from September, the same kind of behavioural comparison Huesler made with his consent-banner signal.
5. Add an annotation in GA4, or in your reporting tool, marking the trend break around 8 to 10 October, as Huesler recommends, so anyone reading the October numbers later knows the series changed.
6. Report ChatGPT and Direct together for affected pages until the situation is clear, rather than presenting a ChatGPT decline on its own.

None of these checks can recover the lost attribution. They can only show whether the pattern fits your site. Server logs may help where they record the full request, but they face the same missing referrer.

What is not known

Search Engine Roundtable does not say whether OpenAI meant to make this change, and the article includes no OpenAI comment. It is not known whether the change applies to every user, device or country, whether it is a bug that will be reversed, or whether OpenAI will offer another way to identify ChatGPT traffic. Huesler's analysis covers about 500 pages over four days of sampled data. Billard's observations come from his own sites. Neither is an audit of ChatGPT traffic in general.

What this means for Thai marketers

The source does not mention Thailand or any specific country. If the change is global, which nobody has confirmed, Thai sites that track ChatGPT referrals in GA4 would see the same under-count from early October. Thai businesses that started reporting AI traffic as a separate line, often as part of GEO work, should check October numbers before presenting them, and note the possible tracking break in any month-on-month comparison. The checks above work the same way on Thai-language sites.

FAQ

Did OpenAI confirm that ChatGPT removed UTM parameters?

No, OpenAI has not confirmed it. Search Engine Roundtable reported the change on 14 October 2026 based on practitioner data and said it was not sure whether it was intentional.

Why is ChatGPT traffic showing as Direct in GA4?

According to Cedric Huesler of Adobe, GPT-6 opens links with noopener,noreferrer and drops the utm_source=chatgpt.com tag, so the visit arrives with no tag and no referrer and is classified as Direct. This is his analysis, not an OpenAI explanation.

How big is the effect?

In Huesler's sample of about 500 pages, ChatGPT visits fell by about 31,000 a day and Direct rose by about 26,000 a day between 9 and 13 October 2026 compared with the week before. He noted the data covers four days and is sampled.

Do I need to change anything?

No tracking-code change is known to fix this, but October reports need context. Compare Direct and ChatGPT sessions on the landing pages ChatGPT used to send traffic to, and annotate the break around 8 to 10 October.

Does this affect Thai websites?

The source does not say which countries are affected. If the change applies globally, Thai sites tracking ChatGPT referrals would see the same shift, so Thai marketers should run the same landing-page checks.

The ChatGPT clicks may still be arriving. What seems to have changed is the label on them, and until OpenAI explains, October's ChatGPT numbers need a footnote. If your AI traffic reporting needs rebuilding around this, Relevant Audience's GA4 team can review your setup.