The sharp drop in ChatGPT referral traffic that analytics teams saw in early October was a bug, not a change in user behaviour and not a deliberate removal of tracking. Search Engine Land reported on October 20 that OpenAI confirmed the missing UTM parameters to Adobe’s Cedric Huesler on October 14, that a fix began rolling out early on October 15, and that tagged ChatGPT clicks had recovered to roughly 95 to 100 percent of prior levels by 9 a.m. Pacific Time that day, based on telemetry covering thousands of domains.
For about a week, links in ChatGPT answers stopped carrying the utm_source=chatgpt.com parameter that OpenAI introduced in June, and the links were opened in a way that suppressed referral information. With neither a tag nor a referrer, analytics platforms filed those visits under Direct traffic instead of ChatGPT.
The Numbers Behind The Apparent Drop
Huesler examined approximately 500 pages where ChatGPT previously accounted for at least 5 percent of visits, comparing October 9 to 13 with the preceding week. ChatGPT-attributed traffic fell by approximately 31,000 visits per day, while Direct traffic rose by approximately 26,000 visits per day on the same pages. Search traffic fell 7 percent, consistent with broader search trends, and two-thirds of the pages that lost ChatGPT referrals saw Direct traffic increase.
The analysis also found similarities between the behaviour of the newly classified Direct visitors and previously identified ChatGPT visitors — a pattern consistent with the same audience arriving under a different label. The timing lined up with the GPT-6 release: reports of missing parameters followed that rollout, and the tagging had been a relatively recent addition, extended to more ChatGPT links after its June introduction.
What OpenAI Confirmed, And What Is Still Missing
Huesler contacted OpenAI on October 14, and the company confirmed the missing UTM parameters were a known bug, with a fix expected that week. The rollout began early on October 15 and recovery was rapid on the telemetry Huesler cited. OpenAI has not, in the reporting, published a separate public post-mortem; the confirmation reached the industry through Huesler and the trade press.
One gap remains even after the fix. Referral information is still missing from most ChatGPT clicks, Search Engine Land noted, which makes the utm_source=chatgpt.com tag the main attribution signal analytics tools have. If the tag breaks again, there is no referrer fallback to catch the traffic.
What Marketers Are Told To Do With The Gap Week
Huesler’s recommendations, as reported, are practical. First, annotate October 8 to 15 in Google Analytics 4 or Adobe Analytics, so the temporary decline in ChatGPT referrals and the corresponding rise in Direct traffic are explained in the record. Second, do not change attribution models on the strength of that week: the apparent loss was misclassification, not evidence of a genuine decline in visits. Third, keep using UTM tracking, because it is now the load-bearing signal for measuring ChatGPT-driven visits.
Businesses monitoring AI-driven discovery should read referral trends, Direct traffic and landing-page performance together before concluding anything about ChatGPT’s effectiveness, the report added.
Why It Matters
The UTM bug matters because AI referral measurement is still held together by a single courtesy tag. Brands building an AI-visibility practice — the work covered in the Fudan study of AI search tools citing a fictional concept, in the paper that found OpenAI’s API matched only 26 percent of pages ChatGPT cited and in how brands track their visibility inside ChatGPT and Gemini answers — depend on click attribution to prove any of it worked. A one-week tagging failure silently moved tens of thousands of daily visits into the Direct bucket, and only an outside analyst’s cross-check caught it. That fragility, not the bug itself, is the story for every team reporting AI traffic to a board.
Background
OpenAI began adding utm_source=chatgpt.com to links in ChatGPT answers in June 2026 and later extended the tagging to more links, giving publishers their first consistent way to count ChatGPT referrals in standard analytics tools. The October incident followed the GPT-6 rollout, when links began opening with the referrer suppressed and the tag missing; similar attribution gaps have been a recurring complaint as AI assistants have become a measurable source of website discovery alongside search and social.
Source: Search Engine Land
0 Comments