ChatGPT Links Lost Their UTM Tags, and Direct Traffic Rose

Around 9 October 2026, sites that had been counting visits from ChatGPT began to see that number fall while direct traffic rose. Search Engine Roundtable reported the pattern on 14 October, attributing the analysis to Cedric Huesler of Adobe and a second set of observations to Sébastien Billard. This post separates what they reported from what has been established, because no statement from OpenAI accompanied the report and the cause is not confirmed.

What was reported

The report says OpenAI began adding UTM tags to ChatGPT links in June 2026, and that the tags appear to have stopped around the release of GPT-6. According to Huesler, GPT-6 opens links with noopener,noreferrer and no longer adds utm_source=chatgpt.com, so analytics software has nothing to attribute the visit to and records it as direct.

His figures come from about 500 pages that previously received at least 5% of their visits from ChatGPT, comparing 9 to 13 October with the week before:

• ChatGPT visits fell by about 31,000 a day, while direct visits on the same pages rose by about 26,000 a day.
• Search traffic on those pages fell 7%, the same as elsewhere, so the shift was not a ranking change.
• Direct traffic rose on two in three of the pages that lost ChatGPT traffic.
• On mobile, the share of new direct visitors who saw a consent banner rose from 36% to 44%, closer to the 55% seen for ChatGPT visitors.

Billard separately reported a sharp fall in ChatGPT human traffic and in ChatGPT-User bot traffic since about 9 October, with OAI-SearchBot traffic rising over the same period, and said he was still looking into the UTM parameter.

What is not established

Everything above is third-party observation. The article did not verify the referrer behaviour independently, and it says it is unclear whether the change was intentional. A tag disappearing and a visit being sent by another route are different explanations that produce the same chart, and the figures cited do not separate them. Treat the 26,000 against 31,000 as evidence that most of the lost visits moved into direct, not as a measurement of ChatGPT’s real traffic.

Why the visits land in direct

Google’s documentation for the GA4 default channel group defines Direct as sessions where the source is exactly “(direct)” and the medium is “(not set)” or “(none)”. A visit with no referrer and no UTM tag has no other source to report, so it falls into that definition automatically. The same would happen for an email client, a messaging app or a saved link, which is why direct has always been a mixed bucket. A rise in it after a change like this one cannot be assigned to any single source from analytics alone, as the guide to using GA4 for SEO decisions sets out.

What server logs can and cannot add

OpenAI’s page on its crawlers describes OAI-SearchBot as the agent that surfaces sites in ChatGPT search, and ChatGPT-User as the agent that fetches pages when a user asks ChatGPT a question. Both identify themselves in server logs. They record fetches made by OpenAI’s systems, however, and not clicks made by people, so logs can show that a page was retrieved to answer a question and cannot show whether anyone then followed the link. The distinction between the two agents is explained in more detail in the OpenAI documentation on its bots and user agents.

How to read your own numbers

Three habits keep the analysis honest. First, annotate 9 October in the analytics property, so that any later comparison carries a note that the tagging may have changed. Second, compare direct landing pages week on week for the pages that earned ChatGPT visits earlier, since a rise concentrated on those pages is the pattern in Huesler’s data, while a rise across the whole site is not. Third, do not relabel direct traffic as ChatGPT in a report. An estimate with a stated assumption can be shown beside the measured number, but replacing the measured number with it makes a figure look more certain than it is.

For sites that earn citations rather than clicks, the wider problem is covered in the post on measuring AI search visibility, and for publishers the position of ChatGPT as a source is covered in the post on ChatGPT search and publishers. If OpenAI comments on the change, the correction to this post will be a single paragraph.