ChatGPT stopped tagging its outbound links for a week, making publisher AI traffic look like it collapsed

During the rollout of the GPT-6 interface, links in ChatGPT answers stopped carrying the utm_source=chatgpt.com parameter and began opening with noopener,noreferrer, which strips the referrer entirely. The effect ran from roughly 8 to 15 October. OpenAI confirmed the behaviour as a bug on 14 October and shipped a fix from the 15th.

The measured scale was substantial. Across around 500 monitored pages, ChatGPT sessions fell by approximately 31,000 a day while Direct traffic rose by approximately 26,000 a day. The share of ChatGPT links carrying a source tag dropped from about 98% on 8 October to about 67% by 14 October.

Why this matters. If you publish anything and you check your analytics, your ChatGPT referral figures for 8 to 15 October are understated, and the missing visits are sitting in your Direct channel. That is an attribution artefact rather than a loss of traffic, and the distinction matters because the wrong conclusion is an expensive one: a publisher looking at that dip could reasonably decide that AI visibility work had stopped paying, or that a crawler or indexing problem had got worse, when neither had happened.

The practical step is to re-baseline from 16 October onwards before drawing any conclusion about AI referral trends, and to treat any week-on-week comparison spanning that window as unusable. It is also a reminder of how fragile referral measurement is for AI search generally: the data depends on a parameter the platform chooses to add, and a single interface change removed it without notice.