Practitioner reports recovery in ChatGPT UTM tagging

What changed

Cedric Huesler reports a measurement anomaly in his own telemetry between 8 and 15 October. He says links from ChatGPT stopped carrying the usual chatgpt.com UTM source tag, reducing attributable ChatGPT referral traffic while visits without a referrer could be counted as direct. By 15 October, he reported tagged clicks at roughly 95–100% of the comparable prior-week hour across thousands of domains in his dataset. He also said referrers remained absent in many cases. The reported OpenAI confirmation was private correspondence and has not been inspected; public replies included Glenn Gabe saying he did not see recovery.

PMC suggests treating this as a possible attribution discontinuity in a particular telemetry environment, not proof that all ChatGPT traffic, all analytics implementations or all sites behaved the same way. Create a dated annotation for the reported 8–15 October window where it is relevant to the property being reviewed. Then examine the same landing pages and comparable hours before, during and after that window, rather than relying only on a single channel line.

Compare UTM-tagged ChatGPT referrals, direct traffic, landing-page sessions, conversion paths and raw request-level evidence where the organisation lawfully retains it. Look for a coincident shift: a fall in tagged ChatGPT traffic alongside a rise in direct entries to the same pages may warrant further investigation, but it does not by itself prove a cause. Review tracking changes, consent behaviour, redirects, campaign tagging and analytics configuration that could also affect classification. Keep the analysis separate from ChatGPT Ads measurement; the supplied report concerns organic referral measurement.

The supplied material does not establish the dataset’s methodology, composition or representativeness. It also does not independently verify a platform cause, fix timing or uniform recovery. Avoid converting Huesler’s roughly 95–100% comparison into a general performance percentage, and do not say visits never fell for every site. Report property-level findings as property-level findings.

For reporting, keep the annotation alongside property-level evidence so later comparisons do not turn a possible classification change into a confident traffic-loss or recovery claim. Huesler’s observation remains useful as a diagnostic lead, while the cause and extent of recovery need independent confirmation.

Why it matters for advertisers

It helps analysts investigate a possible direct-versus-ChatGPT attribution shift without presenting one practitioner dataset as universal platform evidence.