ChatGPT clicks are landing in direct traffic after GPT-6 dropped UTM tags and referrers

Websites can no longer see which visitors came from ChatGPT. Search Engine Roundtable reports that since the GPT-6 release, links opened from ChatGPT results arrive without the utm_source=chatgpt.com tag OpenAI added in June. They also arrive without a referrer. Analytics tools file those visits as direct.

What changed

Cedric Huesler of Adobe explained the mechanism on LinkedIn. GPT-6 opens links with window.open(url, "_blank", "noopener,noreferrer") and drops the UTM parameter. That removes both signals at once. Losing only the tag would still have left the referrer to identify ChatGPT. With both gone, nothing in the visit points back to it.

The source does not say whether OpenAI made the change on purpose, and Search Engine Roundtable says it is not sure either.

What the numbers show

Huesler looked at ~500 pages where ChatGPT drove 5%+ of visits and compared Oct 9–13 with the week before:

• ChatGPT: −31k visits a day. Direct: +26k a day on the same pages.

• Search on those pages: −7%, the same as everywhere else.

• Direct rose on 2 in 3 pages that lost ChatGPT traffic. The week before, it was a coin flip. Search rose on only 45% of the same pages.

• On mobile, the share of visitors seeing the consent banner went from 36% to 44%. For ChatGPT visitors it is 55%.

The consent banner figure is the strongest of these. The new direct visitors behave like ChatGPT visitors, which is what you would expect if the traffic stayed the same and only its label moved.

What this does not settle

The market's first read is that ChatGPT traffic fell. In this sample it mostly got relabelled. Mostly is the right word, though: direct picked up 26k of the 31k daily visits ChatGPT lost, and the source does not explain the gap. It is also one analyst's sample. Sébastien Billard separately reported a drop in ChatGPT-User robot traffic since October 9th, with OAI search bot traffic up since October 7th. The source does not link that crawler shift to the missing tags, so some of the decline may be real.

What to do with your reporting

If your AI referral numbers come straight from analytics, the week of October 9th will look like a collapse. Mark the date on every dashboard and client report before someone reads it as lost visibility. Treat ChatGPT referral figures from before GPT-6 and after it as two separate series.

Then anchor the baseline to measures that do not depend on a referrer. One is share of answers: how often your brand is cited across a fixed set of prompts. The other is branded search, which catches people who saw you in an answer and searched for you later. ChatGPT referral data only ever counted the people who clicked. Now it misses many of those too.